ACS Publications. Most Trusted. Most Cited. Most Read
CSAR Benchmark Exercise of 2010: Combined Evaluation Across All Submitted Scoring Functions
More

OR SEARCH CITATIONS

My Activity
Publications

Figure 1Loading Img
Download Hi-Res ImageDownload to MS-PowerPointCite This:J. Chem. Inf. Model. 2011, 51, 9, 2115-2131
  • Open Access
Article

CSAR Benchmark Exercise of 2010: Combined Evaluation Across All Submitted Scoring Functions
Click to copy article linkArticle link copied!

View Author Information
† Department of Medicinal Chemistry, University of Michigan, 428 Church Street, Ann Arbor, Michigan 48109-1065, United States
‡ Department of Internal Medicine−Hematology/Oncology, University of Michigan, 1500 East Medical Center Drive, 7216 CC, Ann Arbor, Michigan 48109-0934, United States
*E-mail: [email protected]. Telephone: (734) 615-6841.
Open PDFSupporting Information (1)

Journal of Chemical Information and Modeling

Cite this: J. Chem. Inf. Model. 2011, 51, 9, 2115–2131
Click to copy citationCitation copied!
https://doi.org/10.1021/ci200269q
Published August 3, 2011

Copyright © 2011 American Chemical Society. This publication is licensed under these Terms of Use.

Request reuse permissions

Abstract

Click to copy section linkSection link copied!
High Resolution Image
Download MS PowerPoint Slide

As part of the Community Structure-Activity Resource (CSAR) center, a set of 343 high-quality, protein–ligand crystal structures were assembled with experimentally determined Kd or Ki information from the literature. We encouraged the community to score the crystallographic poses of the complexes by any method of their choice. The goal of the exercise was to (1) evaluate the current ability of the field to predict activity from structure and (2) investigate the properties of the complexes and methods that appear to hinder scoring. A total of 19 different methods were submitted with numerous parameter variations for a total of 64 sets of scores from 16 participating groups. Linear regression and nonparametric tests were used to correlate scores to the experimental values. Correlation to experiment for the various methods ranged R2 = 0.58–0.12, Spearman ρ = 0.74–0.37, Kendall τ = 0.55–0.25, and median unsigned error = 1.00–1.68 pKd units. All types of scoring functions—force field based, knowledge based, and empirical—had examples with high and low correlation, showing no bias/advantage for any particular approach. The data across all the participants were combined to identify 63 complexes that were poorly scored across the majority of the scoring methods and 123 complexes that were scored well across the majority. The two sets were compared using a Wilcoxon rank-sum test to assess any significant difference in the distributions of >400 physicochemical properties of the ligands and the proteins. Poorly scored complexes were found to have ligands that were the same size as those in well-scored complexes, but hydrogen bonding and torsional strain were significantly different. These comparisons point to a need for CSAR to develop data sets of congeneric series with a range of hydrogen-bonding and hydrophobic characteristics and a range of rotatable bonds.

Copyright © 2011 American Chemical Society

Subjects

what are subjects

SPECIAL ISSUE

This article is part of the CSAR 2010 Scoring Exercise special issue.

Introduction

Click to copy section linkSection link copied!
Over the past two decades, docking has advanced from an academic exercise to a useful tool to help develop new leads in the pharmaceutical industry. Structure-based drug discovery (SBDD) methods are very successful in enriching hit rates, but it is possible that current software programs are really more effective at eliminating bad leads than identifying good ones. (1) While the ultimate goal is to properly predict tight binders, removing poor choices is valuable and focuses experiments into more fruitful chemical space.
A study by Warren et al. (2) summarizes many of the current strengths and limitations of SBDD. Many docking and scoring routines did well at binding mode prediction, reproducing ligand poses within 2 Å. Some were successful at virtual screening and yielded enrichments appropriate for lead identification, but none could rank order ligands by affinity. In general, inhibitors with nM-level affinity cannot be consistently ranked over those with μM-level binding. It is not possible to identify “activity cliffs,” small changes that result in significant increases or decreases in affinity.
Overall, there is a clear consensus that docking and scoring is a useful technique with potential to be even better, and the need for better training data is commonly identified as a limitation facing the field. (1) The aim of the Community Structure-Activity Resource (CSAR) is to gather data to help scientists improve their methods. To learn what features are most important to address first, we devised our 2010 benchmark exercise. Our exercise is intended to bring people together to compare different methods and improvements for scoring functions based on crystal structures of protein–ligand complexes (no docking was required in order to remove any limitations or biases arising from the use of different search algorithms). Our data set contains hundreds of diverse proteins and ligands of the highest quality that can be used to identify which systems are influenced by different approaches. For a detailed description of how the data set was curated and prepared for the exercise, the reader is directed to our data set paper in this same issue. (3) Most evaluations in the literature are based on a handful of targets at most, and that limited scope prevents us from properly identifying which features of targets and ligands are most difficult to treat computationally. Furthermore, it does not point toward how to improve our methods.
Here, we present an evaluation across all participants, particularly noting which protein–ligand systems are the most difficult to score and which are scored well. Our hypothesis is that sound statistical methods can be used to identify weaknesses in current scoring functions. Targets that are poorly scored across many, diverse methods point to common deficiencies in SBDD. These are the systems that call for improved approaches, departures from the status quo. For CSAR, the differences in physicochemical properties of the “universally well scored” complexes (GOODs) vs “universally poorly scored” (BADs) help to direct the kinds of data sets we curate. In this analysis, we are particularly interested in the complexes that are consistently scored well versus those that consistently scored poorly. GOOD complexes score within 1.1 pKd of the experimental binding affinity for at least 12 of the 17 scoring functions described below, and BAD must be outliers for 12 or more of the scoring functions, see Figure 1. The BAD complexes fall into two groups: OVERs are weak binders that are consistently overscored, and UNDERs are tight binders that are consistently underscored. Differences in the physicochemical properties of GOODs versus BADs help to identify strengths and weaknesses of current approaches. Below, we show that OVERs tend to have hydrophilic ligands and binding sites, while UNDERs are typically hydrophobic. This highlights the need for efforts like CSAR to develop data sets based on congeneric series with ranging log P and hydrogen-bonding features. This is further supported by the results of three participants in the exercise who found that removing hydrogen-bonding terms and/or Coulombic energies resulted in no change in the agreement between scores and experimental affinities. In fact, the correlations significantly improved for one participant (see the papers of this special issue).

Figure 1

Figure 1. Example of comparing a set of scores, pKd (calculated), to their corresponding experimentally determined affinities. (Top) When fitting a line (black) using least-squares linear regression, the distance in the y direction between each data point and the line is its residual. (Bottom) The residuals for all the data points have a normal distribution around zero. The characteristics are well-defined, including the definition of standard deviation (σ in red, which happens to be 1.4 pKd in this example) and the number of data points with residuals outside ± σ (15.8% in each tail). Higher correlations lead to larger R2 and smaller σ; weaker correlations lead to lower R2 and larger σ, but the distributions remain Gaussian in shape.

High Resolution Image
Download MS PowerPoint Slide
The most important aspect of this analysis is the means we use to combine the scoring functions. Improvement is regularly found by combining scores, and it is typically independent of which scoring functions are combined. (4) Research involving belief theory (5-7) and consensus scoring (4, 8-24) has focused on ways to combine data to identify potential positive outcomes, such as enrichment of hit rates. Here, we use consensus to determine which complexes are not scored well. It is very important to design the methods for combining results so that the statistical treatment makes it extremely unlikely for these choices to result from random chance or noise in the data. The equally important, subsequent task focuses on determining why these complexes are outliers.

Statistics

The methods used are straightforward and well-defined, based on linear regression. The most common assessment for a scoring function is its correlation to experimental binding data. When using a simple, least-squares linear regression to fit data points, the fit line must intersect the point (x̅, y̅), and the slope is defined to minimize the distance in the y direction between the data points and the line. (25) An underlying assumption is that the data has a normal distribution along the y axis. Indeed, the distribution of the 343 affinities in the CSAR-NRC set (3) is normally distributed about its average affinity (average pKd/i = 6.15, median = 6.19, max = 13, min = −0.15) with the skew and kurtosis near 0 (skew = 0.06, kurtosis = −0.04). Therefore, the distribution of the residuals (errors in the y direction off any fit line) is normally distributed and centered at zero, see Figure 1. The standard deviation (σ) of the residuals is directly related to the goodness of fit of the line (smaller σ as R2 approaches 1). R2 is the square of the Pearson correlation coefficient which is the percentage of the total variation that can be fit by the linear relationship.
If we define an outlier as any point with an absolute error ≥1σ, then the probability of that occurrence is well-defined: 15.8% for each tail in the distribution of the residuals. This is true regardless of the R2, so all fits—tighter and weaker—have the same probability of a complex having its residual in the tail regions. Below, we combine 17 scoring functions to identify common outliers or BAD complexes that are poorly scored over most methods. If a complex was poorly scored in all 17 scoring functions, then the probability of it being a random occurrence is 2 × (0.158)17 = 4.77 × 10–14, where the factor of 2 accounts for the fact that an outlier can be either consistently above 1σ or consistently below −1σ to be a common outlier by our definition. The probability of a complex being a common outlier for 16 of 17 scores is 2 × (0.158)16 × (0.842) × 17= 4.32 × 10–12, where the factor of 17 accounts for the number of ways that one scoring function can have an error <1σ. Following this enumeration, we can show that the probability of a complex being a common outlier for 12 or more of the 17 scoring functions due entirely to random occurrence is 1.27 × 10–6. We would need a data set of 787 400 complexes (1/probability) to have one common outlier from random chance. That is 3 orders of magnitude larger than the CSAR-NRC data set, making it basically impossible for any of the common outliers to be due to random chance.
Our definition of ≥12 of 17 scores (at least 70% of the methods) ensures that common outliers do not occur randomly. Therefore, if a complex is a common outlier, there has to be an underlying cause. It has to represent some type of molecular recognition that most methods treat insufficiently. In order to identify the physicochemical properties that may frustrate computational methods, we compared the distribution of those properties between the BAD complexes and the set of GOOD complexes. The set of GOOD complexes was defined as those which scored within 1.1 pKd of their experimental value (|residuals| ≤ 1.5 kcal/mol) for ≥12 of 17 scoring functions. Once these two sets were identified, two-tailed Wilcoxon rank-sum tests were used to assess statistical significance of differences in their distributions of physicochemical properties. Of course, we cannot completely rule out experimental error as a cause for an outlier. Our preparation of the data set removed all common complications, such as crystal contacts or poor electron density for the ligand, but the inherent weaknesses of using crystal structures are still there. We are not accounting for protein flexibility nor are we correcting for any differences in ions or pH between the binding assay and the crystallization conditions. There could be errors in the affinity data, particularly for very weak or very strong affinity where measurements are pushed to their limits. For all outliers, we searched the literature for updated affinity data, but none was found.

Contributors

Most of the scores were calculated by authors featured in this special issue, some by the CSAR team, and a few by participants who spoke at the ACS symposium (26) but were unable to submit papers due to various time constraints. Participants were promised anonymity to encourage submission of scores, even those with poor agreement with experiment, but attendees of the ACS symposium felt that hiding the identity of the scoring functions made it impossible to assess the results and know what inherent weaknesses might underlie the analysis. For that reason, the scoring functions are listed in the Methods Section below.
However, we stress that the list of scoring functions is ordered alphabetically, and it is not related to the ordering used in the Results and Discussion Section. In the discussions below, each scoring function is denoted with the generic term “code X”, where X = 1–17. We have chosen not to link the identity of the scoring functions with their performance to avoid trivializing this work into “winners vs losers.” This benchmark exercise is not a contest, and ranking current scoring functions was not our mission. Our goal is to combine the data across all participants and identify the most important and universal deficiencies in scoring protein–ligand binding. Only by knowing where the most significant pitfalls lie can we prioritize which data are needed most to help the community develop their new methodologies. This information has helped direct the focus of CSAR’s future data sets.

Methods

Click to copy section linkSection link copied!
The CSAR-NRC data set (3) is 343 protein–ligand complexes with binding affinity data (Kd or Ki which we abbreviate as pKd/i) from Binding MOAD (27, 28) and PDBbind. (29, 30) The challenge to participants was to calculate absolute free energies of binding over a very diverse set of proteins and small molecule ligands. The set was originally divided into two subsets so that participants could examine training and testing on related sets if they wished, but for this analysis, we are examining scoring across the full set of data.
We wanted an equivalent comparison across all the methods. To remove bias/error from different docking search routines, we asked participants to simply score poses from the crystal structures of the complex. The electron densities for ligands in the CSAR-NRC set are exceptional (RSCC ≥ 0.9 for the ligands), (3) so using poses from the crystals should be an unbiased treatment for all methods. However, we found that force field (FF)-based methods required minimization of the complexes. Small overlaps, within the error of the coordinates, were enough to create very large van der Waals (vdW) penalties. Therefore, a set of minimized structures were made available in addition to the set of crystallographic complexes. All FF methods used the same minimized structures for consistency. Though we could have asked for each FF method to use structures minimized in that FF, it would have removed the emphasis of an even comparison. The minimizations were simply meant to remove the vdW overlaps that undermined their performance. While this is not ideal, it aims to create an even basis for comparison. Participants were welcome to minimize the structures in their own FF, score again as part of their analysis, and report the results in their papers. They were also able to use crystallographic water if they chose, but none were used in the core scores below.
Sixteen groups participated in the benchmark exercise: 11 academic and 5 from the private sector, both software vendors and pharma groups. The submissions for the CSAR-NRC set included 64 variations on 19 scoring functions. Most participants submitted more than one set of scores, varying different parametric choices to determine their influence upon scoring. Two of the methods were only trained on the CSAR-NRC set and could not be included in this analysis. For each of the other 17 methods submitted, an optimal “core” score was chosen for our combined analysis across all participants. Only standard approaches were considered (only pre-existing functional choices, not functions fit to the CSAR-NRC data set). The option most appropriate to avoid artifacts was chosen. For example, FF scores require minimization of the complexes to remove the artifact of high vdW energies. Minimization was usually unnecessary for soft potentials or knowledge-based potentials, and use of the minimized structures often showed decreased agreement with experimental values. If more than one standard approach was submitted, then the option with better root-mean-square error (RMSE) or R2 to the experimental data was chosen. At times, this resulted in minimized structures being used with knowledge-based or soft scoring functions. As noted above, the order below is alphabetical, but the tables, figures, and various performance metrics are ordered by correlation to the experimental values.

AutoDock 4.2.3 (31)

This is a FF-based function, so scores for the minimized complexes were chosen. We also chose the scores calculated using the default charges (Gasteiger charges from AutodockTools 1.5.4.2) over using AM1-BCC charges provided with the data set because AutoDock is parametrized to use Gasteiger charges.

AutoDock Vina 1.1.1.1 (32)

Standard, default scoring options were chosen, and the protein and ligand were prepared with AutodockTools 1.5.4.2. Scores calculated with the unminimized complexes were chosen because this is a knowledge-based potential.

DOCK 4.0.1 (33)

This scoring function is based on the AMBER FF, so scores for the minimized complexes were chosen. Scores were submitted using both the full DOCK scoring function and the function with only the vdW terms included. The score chosen for the core set used the full scoring function.

DrugScore 1.1 (34)

The default scoring parameters were used. Even though this is a knowledge-based function, scores for the minimized complexes were chosen because they had slightly better correlation with experimental affinities.

eHiTS (35)

The scores chosen for the core set were calculated with the default scoring options using the minimized set of structures. A second set of scores was also submitted with the function tuned on the entire data set. The standard score provided by eHiTS was chosen over the scores calculated from the function trained on the data set. However, the scores fit to the CSAR-NRC set were used as a type of benchmark because the large number of parameters allowed for a very tight correlation to experimental values. This was taken as a limiting case for the maximal performance possible when fitting to the data set.

FRED 2.2.5 (Chemgauss3) (36)

Standard, default scoring was chosen, calculated with the unminimized complexes because this is a soft, shape-based scoring method. The AM1-BCC charges provided with the data set were used. Scores were calculated using Shapegauss, Chemgauss3, Chemscore, OEChemscore 1.4.2, Screenscore, and PLP scoring functions with a box cutoff of 4 Å larger than the ligand. Both the minimized and unminimized structures were tested. The Chemgauss3 function using the unminimized structures was chosen because it achieved the highest R2 in linear fit to experiment for the CSAR-NRC data set.

Glide 5.5 (SP) (37)

The minimized set of complexes was used because Glide is based upon the optimized potentials for liquid simulations (OPLS) FF for atom typing and charges. The grid was centered at the average ligand coordinate and the box extended 25 Å. The standard precision (SP) and the extra precision (XP) scores were calculated. SP was chosen over XP because it was able to score more of systems in the CSAR-NRC data set. Also, SP had a higher R2 in linear fit to experiment for the CSAR-NRC data set.

GOLD 4.0.1 (ChemScore) (38, 39)

The scores of the minimized structures were chosen for the core set, since the performance based on R2 was better. ChemScore was chosen over GoldScore and ASP because it had the best correlation with experiment. The binding site was defined as the region within 12 Å of the ligand’s center of mass. Non-natural amino acids and water molecules are not considered in the rescoring. GOLD used primarily knowledge-based scoring functions (or statistical potential functions), which relied only on atom typing of the ligands, so the charge information of the ligands was not used in the scoring.

ITScore 2.0 (40)

Standard scoring was chosen over the new form that includes a correction for rotatable bonds in the ligand. Scores calculated for the unminimized complexes were chosen because this is a knowledge-based potential. ITScore was trained on 1152 protein–ligand complexes from PDBbind (29, 30) (excluding those in the CSAR data set), to develop the pairwise statistical potentials for the scoring function.

Lead Finder (41)

Scores were based on two preparation protocols. The first used the structures provided in the CSAR-NRC set of minimized complexes. In the second, the structures were prepared using MolTech’s software Model Builder, which calculates the pKas of the ligand to suggest proper ionization states. (42) Though the second preparation showed a slightly improved correlation to the experimental values, the scores from the first preparation were chosen for the core set for consistency with other methods.

MedusaScore (43)

The participant provided two scores based on MedusaScore and a QSAR model. MedusaScore is a FF based function, but the vdW repulsion terms are removed in order to avoid sensitivity to possible atomic clashes in the structures. MedusaScore was chosen over the QSAR approach because the latter was based on descriptors determined from the CSAR-NRC data set.

MOE 2010.10 (ASE and AffinityDG) (44)

Two scoring functions from MOE were chosen as core scores because they used fundamentally different and independent approaches. All MOE scoring functions (ASE, (45) Alpha HB, London dG, and Affinity dG) were used with their default settings and were computed on both the crystal structures and minimized structures provided. Both ASE and Alpha HB are shape-based methods, and ASE was chosen because of its better correlation with the experimental affinities. Both LondonDG and AffinityDG functions attempt to estimate the free energy of binding. Affinity DG had a lower RMSE and was chosen for the core set. The chosen core scores were calculated using the minimized data set, as it provided better agreement with experiment in both cases.

M-Score (46)

The default scoring parameters were used for this knowledge-based scoring function. The minimized structures resulted in scores with slightly better correlation to experiment and were chosen for the core set.

S2 (47)

The S2 function is a linear interaction energy function based on the number of interacting types of pairs, with the weights calculated using linear regression fit to the LPDB data set. Scores for the minimized structures were chosen. The S2 function was chosen over the S1 function because the S1 function only accounts for the size of the molecule.

SIE (48)

Solvated interaction energy (SIE) scores use a FF-based method, so the minimized complexes were used. Parameters were assigned using AMBER/GAFF, but AM1-BCC charges were used for the ligand, cofactor, and any other modified residues. Two parameters of the SIE function (α and C) are fit to experimental binding free energies. The standard approach used 99 protein–ligand complexes. The α and C values were also refit using the CSAR-NRC data set. The standard scores provided from the SIE scoring function were chosen over the scores fit to our data set, to remove any bias and maintain consistent treatment across all core scores.

X-Score 2.0 (49)

No special preparation was performed, and the default scoring parameters were used. X-Score reports three scores (HPScore, HMScore, and HSScore) and the average of all three. We chose the average score for the core set, based on the minimized set of structures. X-Score was developed from the PDBbind data set which has significant overlap with the CSAR-NRC set.

Caveats

Of the 17 core methods above, 12 were able to score all 343 complexes. Two methods left one complex unscored, and the other three were unable to score two, four, and five complexes, respectively. Any complex left unscored was counted as an outlier for that method. On very rare occasion, we needed to drop an individual complex from a method’s set of scores (a single complex was removed from three methods, and three complexes were removed from one method). In these cases, the complexes’ scores were well off from the rest of the set, and the residuals were greater than 3.5σ (very unlikely in a set this size). These were very likely errors in the calculation, and they greatly skewed the linear regression analysis outlined below. Of course, any removed complex was treated the same as an unscored complex.

Correlation between Scores and Experimental Binding Affinities

The statistical package “R” (50) was used to calculate Pearson’s R, Spearman ρ, and Kendall τ values. Fisher transformations coupled with standard deviations were used to determine 95% confidence intervals. (51) Every method was normalized so that the scaled scores ranged from 0 to 1 (i.e., scores were converted by first making all scores positive numbers and then scaling by [(scorei – scoremin)/(scoremax – scoremin)]). This simply shifts the values and scales them. It does not change the value of R2, but it does change any Person’s R, Spearman ρ, or Kendall τ with a negative value to its absolute value. This treatment makes it easier to compare across the methods because R, ρ, or τ of 1.0 always indicates perfect correlation and rank ordering, regardless of whether the original scores were given in positive or negative numbers.

Selecting BAD and GOOD Complexes by Linear Regression

Each method’s scores were compared to the experimental pKd/i through least-squares linear regression analysis in JMP. (52) As noted in Figure 1, the residuals are normally distributed with σ proportional to the goodness of fit. JMP (52) was used to compare the residuals across all the fits and determine the list of BAD complexes. These BAD complexes fell into two groups. The UNDER set consisted of those complexes with residuals ≥1σ (under scored) in at least 12 of 17 functions, and the OVER set were those with residuals ≤ –1σ (over scored) in at least 12 of the 17 functions. Before progressing to the comparison of BAD and GOOD complexes, we searched the literature to identify whether any subsequent research on those targets had identified any errors in the processing or disagreements in affinity measurements. No errors in set up or changes to affinity data were found.
When identifying the GOOD structures, we wanted to remove any bias in the linear regression arising from trying to fit the BAD complexes. Therefore, the BAD structures were removed, and then the remaining complexes were fit again in JMP. (52) The set of GOOD complexes was defined as those with an absolute value of its residual less than 1.1 pKd/i (<1.5 kcal/mol) in at least 12 of the 17 core scoring functions.

Comparison between the Physical Properties of BAD and GOOD Sets

We calculated over 400 physicochemical properties for each complex based on the ligand, the protein, and the interactions between the two. For the ligand, all two- and three-dimensional (2D and 3D) properties available in MOE (44) were calculated, except for those requiring semiempirical quantum mechanics. Energy descriptors were calculated with the MMFF94x force field in MOE. This provided 319 ligand descriptors.
The proteins, binding sites, and the protein–ligand interactions were examined in many ways. Properties describing the quality of the crystal structure included clash scores calculated by MolProbity, (53) all Z scores calculated by WhatIf, (54) DPI, (55) and the Rfree and resolution reported in the original publication for each complex. The chemical interactions between the ligand and the protein were determined by the prolig_Calculate function available in MOE, (44) which yielded hydrogen-bonding, ionic, arene, and metal interactions within the binding sites. The buried and exposed molecular surface area (SA) of the binding pocket was calculated with GoCav. (56) The hydrophobic buried SA was estimated by determining which nonhydrogen atom of the protein was closest to the buried surface grid point determined by GoCav. If a carbon atom of the protein or the sulfur atom of a methionine residue was closest, then the point was considered hydrophobic. All other buried SA points were considered hydrophilic. Bridging water molecules were required to be less than 50% solvent exposed and to be within 4 Å of a nonhydrogen atom of both the ligand and the protein. Any natural amino acid, modified residue, and metal atom with a nonhydrogen atom within 4 Å of the ligand’s nonhydrogen atoms were considered part of the protein’s binding site. These contacts were determined and tallied using in-house code written in Perl. Each binding site was then described by its %amino-acid content (number of each of the 20 amino acids in the binding site divided by the total number in the binding site, where metals and modified residues were counted as a 21st residue called “other”). Then averages and standard deviations for the amino acid content of the binding sites were determined by bootstrapping for 1000 iterations, randomly combining two-thirds of the data set each time. GOOD, OVER, and UNDER complexes were each bootstrapped as separate sets.
For every physicochemical property, JMP (52) was used to compare the distribution of values for the GOOD complexes to the distribution of values for the BAD complexes. UNDER and OVER complexes were compared separately to the GOOD complexes. A nonparametric, two-tailed, Wilcoxon rank-sum test was performed to calculate the likelihood that distributions of physicochemical properties were the same. Only properties with p values ≤0.05 were considered relevant.

Results and Discussion

Click to copy section linkSection link copied!

Factor Xa (FXa) Complexes Were Removed Early in the Analysis

The initial set of identified outliers contained several FXa structures. Each had ligands with sub-nM-level affinities, but the pockets were well exposed and the complementarity appeared poor. All FXa structures are missing an N-terminal domain, and its effect on ligand binding is unclear. In vivo, the domain is required for calcium activation of FXa, and the anticoagulant warfarin works by inhibiting the modification of this domain’s key residues that chelate calcium. (57) Therefore, we removed all 11 FXa structures from the analysis of BAD and GOOD structures.
The subsequent analysis below is based on the 332 remaining structures in the CSAR-NRC set. Note that after dropping these complexes, the characteristics of the set are basically unchanged. The maximum and minimum affinities are the same. The average and median affinities of the 332 set are 6.07 and 6.115 pKd/i, respectively. The standard deviation is 2.20 pKd/i, and the median unsigned error (Med |Err|) is 1.47 pKd/i. The distribution is still Gaussian with near-zero skew and kurtosis of 0.09 and 0.04, respectively.
However, FXa has been a platform for successful structure-based design (58) and testing of modeling techniques. (59, 60) There is a wealth of additional data on FXa in the pharmaceutical industry (61-69) that would help the field overcome its limitations. In fact, we are currently negotiating with two companies for FXa data. There are strong electronic influences and π–π stacking effects that can shift ligand affinities over 4 orders of magnitude. Structure 2p3t (67) (set 1, complex no. 141) is a good example of the difficulty in modeling some high-affinity inhibitors of FXa. It has a binding affinity of 5 pM but appears to have poor complementarity and is largely solvent exposed, see Figure 2. The effects of halogenation and π–π stacking provide a strong driving force for the association. Below, we show that hydrophobic interactions in tight-binding complexes are underestimated by the majority of methods examined in the benchmark exercise.

Figure 2

Figure 2. Crystal structure of FXa bound with a 5 pM ligand (PDB id 2p3t). The ligand is very exposed with few hydrogen bonds to the protein.

High Resolution Image
Download MS PowerPoint Slide
It should be noted that several structures of HIV-1 protease were in the BAD set. At the same time, there were twice as many in the GOOD set, and the complete protein is present in the crystal structures. Therefore, we chose not to eliminate those complexes from the analysis. A table in the Supporting Information lists the proteins that have an appropriate ligand series in the data set, including FXa. Proteins were clustered at 100% sequence identity to avoid data on mutants muddling the analysis of affinities. Sixteen proteins (listed with PDB ids in the data set paper) (3) had three ligands or more, but only 10 had ligand affinities that ranged more than 1 order of magnitude. No limit was placed on chemical similarity across the ligands to define some measure of congeneric series, though some clearly are. Each code’s relative ranking of the ligand series for each of the 10 proteins is also provided in the table in the Supporting Information. A list of all protein families and their complexes has been included in the download of the data set since the project was initiated (www.CSARdock.org, accessed July 25, 2011).

Correlation between Experimental Affinities and the 17 Core Methods

Table 1 presents both parametric and nonparametric assessments of the correlation between the submitted scores and the experimental binding affinities of the 332 entries used. For methods that estimated affinities, the Med |Err| was 1.00–1.68 pKd/i (1.4–2.3 kcal/mol), and the RMSE ranged from 1.51 to 2.99 pKd/i (2.1–4.1 kcal/mol). Across all 17 core scores, the σ values for the residuals from the linear regression were 1.43–2.07 pKd/i (1.9–2.9 kcal/mol). For FF-based, knowledge-based, and empirical scoring functions, all had examples with high and low correlation. There was no obvious advantage to choosing one type over another.
Table 1. Parametric and Nonparametric Measures of Correlation Between the Scores and Experimental Binding Affinitiesa
methodPearson RSpearman ρKendall τR2σbRMSEbMed |Err|b
code 10.76 (0.80–0.71)0.74 (0.79–0.68)0.55 (0.60–0.50)0.58 (0.64–0.50)1.431.511.00
code 20.72 (0.77–0.66)0.73 (0.78–0.67)0.54 (0.59–0.49)0.52 (0.59–0.44)1.53  
code 30.67 (0.72–0.60)0.68 (0.74–0.61)0.49 (0.54–0.43)0.45 (0.52–0.37)1.641.651.05
code 40.64 (0.70–0.58)0.64 (0.70–0.56)0.46 (0.52–0.40)0.42 (0.49–0.33)1.682.091.5
code 50.63 (0.69–0.56)0.64 (0.71–0.57)0.46 (0.52–0.40)0.40 (0.48–0.32)1.71  
code 60.62 (0.68–0.55)0.61 (0.68–0.53)0.43 (0.49–0.38)0.39 (0.47–0.30)1.721.811.26
code 70.62 (0.68–0.55)0.61 (0.68–0.53)0.43 (0.49–0.37)0.38 (0.46–0.30)1.72  
code 80.61 (0.67–0.54)0.59 (0.66–0.51)0.42 (0.48–0.36)0.37 (0.45–0.29)1.75  
code 90.61 (0.67–0.53)0.60 (0.67–0.52)0.43 (0.49–0.37)0.37 (0.45–0.28)1.75  
code 100.60 (0.66–0.52)0.60 (0.67–0.52)0.43 (0.48–0.37)0.36 (0.44–0.27)1.772.991.67
code 110.59 (0.66–0.52)0.57 (0.64–0.49)0.40 (0.46–0.34)0.35 (0.43–0.27)1.771.921.36
code 120.57 (0.63–0.49)0.57 (0.65–0.49)0.41 (0.47–0.35)0.32 (0.40–0.24)1.822.181.28
code 130.56 (0.63–0.48)0.60 (0.67–0.52)0.42 (0.48–0.36)0.32 (0.40–0.24)1.822.521.68
code 140.56 (0.63–0.48)0.54 (0.62–0.45)0.38 (0.44–0.31)0.32 (0.40–0.23)1.82  
code 150.56 (0.63–0.48)0.56 (0.63–0.47)0.39 (0.45–0.33)0.31 (0.39–0.23)1.83  
code 160.53 (0.60–0.45)0.53 (0.61–0.44)0.37 (0.43–0.31)0.28 (0.36–0.20)1.871.901.23
code 170.35 (0.44–0.25)0.37 (0.46–0.27)0.25 (0.32–0.18)0.12 (0.20–0.06)2.07  
Yardsticks (Maximum and “Null” Correlations)
trained on 343 setc0.93 (0.94–0.91)0.93 (0.94–0.90)0.77 (0.80–0.74)0.86 (0.89–0.83)0.820.950.48
heavy atoms0.51 (0.58–0.42)0.49 (0.57–0.40)0.35 (0.41–0.28)0.26 (0.34–0.18)1.90  
Slog P0.46 (0.54–0.38)0.50 (0.58–0.41)0.34 (0.40–0.28)0.22 (0.30–0.14)1.95  
a

Values obtained through analysis of the set of 332 complexes (FXa structures removed from the CSAR-NRC set). 95% confidence interval in parentheses, units of pKd for σ, RMSE, and Med |Err|.

b

Metrics appropriate for the methods that estimated absolute binding affinities, rather than relative ranking; units are pKd.

c

One of the 17 methods above, fit with many adjustable parameters specifically to reproduce the 343 complexes of the full CSAR-NRC set.

Modest correlations were expected because of the difficulty in predicting absolute free energies of binding, but the correlations are just as good as those for the easier problem of relative ranking to a single protein target. (2) Spearman ρ and Kendall τ are nonparametric and reflect the relative ranking across the complexes, with ρ nearly equaling the Pearson R values (0.76–0.35), while τ is less.
The R2 range is 0.58–0.12 with the bulk of the methods falling between 0.4 and 0.3. Caution must be used in making a statistically significant evaluation across the codes. Though a 95% confidence interval of R2 can be analytically determined, (51) the difference in the 95% confidence intervals is not the same as a 95% confidence in the difference. It is most appropriate to evaluate the statistical significance between the R2 values by examining the residuals that underlie the correlation. As shown in Figure 1, residuals for all fits are normally distributed around zero, so a rank-sum test is not appropriate. Instead, the difference in the spread of the distribution can be evaluated using Levene’s F-test for the equality of variance (calculated using the “R” statistical package). (50) This is more stringent than simply comparing the confidence intervals. Codes 1 and 3 have a small overlap in their 95% confidence intervals of R2, but the F-test of their residuals provides a p value of 0.015, meaning that they are statistically significant in their difference. (Levene’s tests for code 3 show it to be statistically comparable to codes 2 and 4–11, but it is a method parametrized on the PDBbind data set, (29, 30) which has a great deal of overlap with the CSAR-NRC set. A “performance” comparison to other methods is not particularly meaningful.) Levene’s test for the residuals of codes 1 and 2 gives p = 0.23; therefore, the performance of codes 1 and 2 are comparable. F-test comparisons of codes 4–16 have p > 0.05, making them equivalent. The very low R2 for code 17 is statistically significant in its difference to the other core methods.

Yardsticks for Linear Regression

The equivalence of the overwhelming majority of methods further underscores why the benchmark exercise is not a contest. More importantly, only codes 1–6 are statistically significant in their difference to common “null cases.” Perhaps the most appropriate null case for scoring functions is the correlation between affinity and the number of nonhydrogen atoms in the ligand, (70, 71) which is R2 = 0.26 for this set with a 95% confidence interval of 0.34–0.18. This is a particularly useful counter example because the additive, pairwise potentials used in most scoring functions lead to ever increasing scores as more atoms are added to the ligand. (70-72) Another null case to consider is the correlation between the affinity and the hydrophobicity of a ligand. It is well-known that adding hydrophobic moieties to a ligand will increase its affinity. (73-75) This is a bit of a “cheat” because one is simply disfavoring the unbound state. We have used SlogP values (76) calculated with MOE to provide this null case (R2 = 0.22 with a 95% confidence interval of 0.30–0.14). We should caution the reader that, unlike the count of nonhydrogen atoms, predictions of hydrophobicity are parametrized methods just like scoring functions, except that the values are independent of the protein target.
We should note that the highest correlation to experiment was obtained when the method with the most adjustable parameters was refit using the 343 CSAR-NRC complexes: R2 = 0.86, R = 0.93, ρ = 0.93, and τ = 0.77. This is provided in Table 1 as an example of maximum performance possible with the data set. As our paper on the data set noted, the experimental uncertainty should limit the correlation to an R2 of ∼0.83 when fitting to this data without overparameterizing. (3) Of the 64 total submissions, 4 others also fit to the whole data set, obtaining R2 of 0.54–0.42, R of 0.73–0.64, ρ of 0.71–0.64, and τ of 0.52–0.46.

Identification of 63 BAD and 123 GOOD Complexes by Linear Regression and σ

A complete list of the BAD and GOOD complexes is given in the Supporting Information. Figure 3 compares the 17 core scoring functions to the experimental affinities. The red lines highlight complexes with residuals within and outside ±1σ, where any point outside is an outlier for that individual method. The BAD complexes, defined by having residuals outside ±1σ for at least 12 of 17 methods, were composed of 34 OVER (weak binders scored too high) and 29 UNDER (strong binders scored too low). Figure 3 shows that every method may score a few BAD complexes well (red and blue data points between the red lines).

Figure 3

Figure 3. Least-squares linear regression of the 17 core scoring functions. Black lines are the linear regression fit. Red lines indicate +σ and −σ, the standard deviation of the residuals. Blue points are UNDER complexes which were underscored in ≥12 of the 17 functions. The red points are OVER complexes which were overscored in ≥12 of the 17 functions.

High Resolution Image
Download MS PowerPoint Slide
From the linear regression of the 332 complexes, 116 had residuals within ±1.1 pKd/i (1.5 kcal/mol) for ≥12 of 17 methods. However, it is best to keep the BAD structures from influencing the linear regression and subsequent identification of GOOD systems. Therefore, we removed the 63 BAD complexes and refit the remaining 269 complexes for each method. Based on the ≥12 of 17 requirement, the number of systems with residuals within ±1.1 pKd/i increased to 123. The Supporting Information outlines the procedure in a figure and provides a discussion of our metrics for identifying the GOOD complexes.

Methods that Estimate Absolute Binding Affinities

Nine of the core methods estimated binding affinities, rather than providing simple rank scores. The Med |Err| ranged from 1.00 to 1.68 pKd/i, and the RMSE was 1.51–2.99 pKd/i, see Table 1. The agreement with experiment ranked the codes in the order (1, 3) > (16, 6) > (11, 12, 4) > (13, 10). For this ranking, greater importance was given to Med |Err| because RMSE heavily weighs the farthest outliers. For estimates of affinities, having less error for more complexes is more important than having outliers that are closer but still quite far off.
A suggested null case is to calculate the RMSE and Med |Err| while setting every score to the average experimental value (6.07 pKd/i). For RMSE, that simply gives the standard deviation of the data set, 2.2 pKd/i, and the Med |Err| is 1.47 pKd/i. Five of the 9 methods have RMSE and Med |Err| less than the null case (codes 1, 3, 16, 6, and 11). Two methods have errors less than one null metric but nearly equal to the other (codes 12 and 4). Two methods have errors in excess of the null case (codes 13 and 10). Of course, all of the methods have RMSE larger than their standard deviation from the linear regression, but it is most pronounced for codes 10 and 13 which seem to be biased to overscore and underscore, respectively, across the full range of complexes (see Figure 4). Codes 3 and 16 rank very well and have low errors, but they appear to have some bias that limits the scoring range to roughly 4–10 pKd.

Figure 4

Figure 4. Comparison of experimental and calculated values from the nine functions which predicted absolute binding affinity, listed roughly in order of increasing Med |Err| and RMSE. Black lines represent perfect agreement. The red lines indicate +Med |Err| and −Med |Err| from the black line. The blue circles denote complexes for which ≥7 of the 9 methods have consistently underestimated the affinity by at least Med |Err|, while the red circles are those where the affinity was overestimated.

High Resolution Image
Download MS PowerPoint Slide
Figure 4 compares the estimated affinities with the experimental values for these nine core methods, where any estimated free energies were converted to pKd. We have identified universal outliers much as we did through linear regression. Using RMSE for a cutoff would parallel the previous analysis, but as stated above, low Med |Err| is more important in this type of scoring. OVER complexes had errors (pKd/iexperiment – pKdscore) less than −1 × Med |Err| in at least 7 of the 9 functions (78% of the methods, a larger percentage than the 12 of 17 requirement for the sets defined by linear regression). UNDER complexes were determined by the errors greater than 1 × Med |Err| for ≥7 of 9 methods. Again, structures were determined to be well scored if their error was <1.1 pKd. This cutoff was maintained even if the Med |Err| was less than 1.1. This lead to 36 OVER, 28 UNDER, and 34 GOOD complexes based on Med |Err|. Unfortunately, there is no way to estimate the statistical significance of the sets determined in this manner, but the overwhelming majority of complexes are also in the sets determined by linear regression. The complexes are listed in the Supporting Information.

Comparison of the GOOD versus BAD Complexes

The comparison of GOOD and BAD complexes below focuses only on the sets determined through linear regression because of the solid statistics outlined in the Introduction Section. We next applied the concept of a null hypothesis to this portion of the analysis and developed a null set of complexes (NULL) to characterize a type of signal-to-noise metric. The first graph in Figure 5 shows the distribution of affinities for the GOOD, OVER, and UNDER sets. There is a large bias for OVER complexes to have low affinities, UNDER complexes to have high affinities, and GOOD complexes to lie in between. Therefore, we defined the NULL cases based on affinities and compared the characteristics of the signal to the inherent background. Within this framework, the signal is the comparison of GOOD to OVER and UNDER complexes, and the NULL sets simply compare complexes with midlevel affinity to weak binders and tight binders, respectively. First, we divided the 332 complexes into three subsets, using cutoffs of ≤50 nM and ≥50 μM, as shown in gray shading in Figure 5. We then removed any UNDER complexes from the high-affinity subset, any GOOD complexes from the midrange subset, and any OVER complexes from the low-affinity subset. The NULL set contained 179 complexes: 65 high-, 69 mid-, 45 low-affinity complexes. We would like to ensure that the differences in physical properties are not simply a reflection of affinity. Obviously, those properties are important in scoring and will be represented across the sets, but the use of a NULL set helps us identify potential bias arising from the definition of a difficult-to-score system.

Figure 5

Figure 5. Distribution of binding affinities in the GOOD and BAD complexes (left) are compared to those of the NULL case (right). The NULL case is generated by the sets of all complexes with affinities ≤50 nM (high), 50 nM–50 μM (middle), and ≥50 μM (low). This midrange of affinities is highlighted with a wide, gray bar on both figures.

High Resolution Image
Download MS PowerPoint Slide
Our goal in this analysis is to identify the physical characteristics of the proteins and the ligands that are most difficult to model. However, we cannot base this determination solely on the properties of the BAD complexes; they must be compared to the GOOD complexes. Any properties common to both sets cannot be the root of the problem. We need to identify characteristics of difficult systems that are significantly different from the easy ones. For instance, there are studies in the literature indicating that metalloenzymes are difficult to model, (77-79) and if many of the BADs contained metals in their binding sites, then it would be tempting to conclude that metals are a major stumbling block for our field. However, that would not be true if many metalloenzymes were also in the GOOD set. We can easily calculate p values for differences in the distributions of various system properties in GOOD vs OVER and GOOD vs UNDER sets. This allows us to identify statistically significant differences between the sets. In fact, we found that there is no difference between GOOD, OVER, or UNDER complexes with respect to metals in the binding sites (medians of 0 for all three sets; means of 0.32 for GOOD, 0.24 for OVER, and 0.24 for UNDER; p values of 0.85 for OVER vs GOOD and 0.79 for UNDER vs GOOD). Of course, this does not mean that metalloenzymes are easy to model; that would require a bias for metals in GOOD only. It was interesting to find that there was a statistically significant bias for metals in the binding sites of low-affinity complexes in the NULL sets (mean of 0.62 for midrange and 1.11 for low with p = 0.03). Having a bias in the NULL set that is not found in the OVER vs GOOD comparison further supports the finding that metals are not a strong influence on BAD complexes. It should be emphasized that this finding is for a general analysis over many methods, and it is still possible that an individual scoring technique could find metals to be its greatest limitation (this was the finding of one participating group that was unable to submit a paper to this special issue).
Table 2 lists the most relevant physicochemical properties of the ligands and pockets of the GOOD, UNDER, OVER, and NULL sets. Medians and p values for each property are provided.
Table 2. Comparison of the Distribution of Physicochemical Properties for UNDER, GOOD, OVER, and NULL Sets. a
 ≥12 of 17 UNDER vs GOOD complexesNULL hypothesis (high vs midrange)≥12 of 17 OVER vs GOOD complexesNULL hypothesis (low vs midrange)
physicochemical characteristicsmedian (GOOD)median (UNDER)Δmedian (UNDER – GOOD)pmedian (middle)median (high)Δmedian (high – middle)pmedian (GOOD)median (OVER)Δmedian (OVER – GOOD)pmedian (middle)median (low)Δmedian (low – middle)p
Ligand Properties
ligand heavy atoms (HA)2120–10.63213615<0.012119.5–1.50.102111–10<0.01
calculated log S–2.15–3.87–1.720.45–0.70–6.06–5.36<0.01–2.15–0.092.06<0.01–0.70–0.200.500.02
calculated SlogP–0.601.422.02<0.01–1.823.425.20<0.01–0.60–2.58–1.98<0.01–1.82–2.55–0.730.97
Lipinski violation0000.100110.030000.860000.06
Oprea violation10–10.05132<0.0110.5–0.50.8110–10.03
Nrot4510.71495<0.014400.3243–1<0.01
Nrot/HA0.220.230.010.830.20.270.07<0.010.220.2200.650.20.17–0.030.10
Etor8.592.89–5.70<0.019.6415.676.030.108.5911.182.590.119.645.94–3.70.28
Etor/Nrot1.430.53–0.900.022.481.32–1.160.061.433.071.640.022.484.301.820.22
no. oxygens43–10.094510.884620.054400.20
no. oxygens/HA0.190.14–0.050.120.290.14–0.15<0.010.190.380.19<0.010.290.360.070.20
no. hydrophobic101330.5792617<0.01108–20.0196–3<0.01
no. hydrophobic/HA0.560.670.110.100.440.710.27<0.010.560.43–0.13<0.010.440.460.020.53
no. acc + no. donors65–10.127700.566710.4674–30.28
(no. acc + no. don)/HA0.240.23–0.010.260.360.21–0.15<0.010.240.380.140.040.360.400.040.51
Hydrogen Bonds and Water in the Binding Pocket
protein–ligand Hbonds95–4<0.018800.159900.1285–3<0.01
pro–lig Hbonds/HAb0.350.19–0.160.060.50.20–0.25<0.010.350.590.24<0.010.50.42–0.080.45
pro–lig Hbonds/(lig no. acc+ no. don)1.000.80–0.200.131.050.94–0.110.331.001.110.110.511.050.67–0.38<0.01
bridging H2O53–20.064620.045610.354310.03
bridging H2O/HAb0.220.20–0.020.060.210.17–0.040.060.220.310.09<0.010.210.2–0.010.80
total H2O in pocket53–20.055720.045610.3354–10.06
total H2O/HAb0.240.20–0.040.030.230.19–0.04<0.010.240.330.09<0.010.230.250.020.56
Surface Properties of the Ligand and Binding Pocket
ligand vdW SA311300–110.81294525231<0.01311293–180.06294182–102<0.01
lig vdW SA/HA14.915.20.30.0914.614.70.10.6214.914.5–0.40.7514.615.30.70.02
hydrophobic lig vdW SA138196580.50122367245<0.01138112–26<0.0112285–370.01
hydrophobic lig vdW SA/HA7.699.491.800.056.289.873.59<0.017.695.66–2.03<0.016.285.64–0.640.65
polar lig vdW SA71.452.1–19.30.1071.075.54.50.3771.490.819.40.2771.059.3–11.70.02
polar lig vdW SA/HA3.592.71–0.880.294.271.89–2.38<0.013.595.051.46<0.014.274.560.890.57
pocket exposed SA83.134.6–48.50.1473.7108.334.60.0783.181.6–1.50.7073.767.9–5.80.91
pocket ESA/HAb2.932.20–0.730.143.612.87–0.740.182.933.810.880.323.615.351.740.02
pocket %ESA0.170.14–0.030.160.180.17–0.010.300.170.190.020.470.180.250.070.11
pocket buried SA301278–230.43282510228<0.01301251–500.12282196–86<0.01
pocket BSA/HAb14.213.8–0.40.6214.113.4–0.70.1414.215.71.50.1914.116.52.4<0.01
%hydrophobic pocket BSA0.520.570.050.010.480.570.09<0.010.520.46–0.060.060.480.520.040.61
%hydrophilic pocket BSA0.480.43–0.050.010.520.43–0.09<0.010.480.540.060.060.520.48–0.040.61
hydrophobic pocket BSA167177100.69139287158<0.01167120–470.02139107–32<0.01
hydrophobic pocket BSA/HAb7.367.770.410.086.957.480.530.037.367.14–0.220.806.957.590.640.16
hydrophilic pocket BSA13586–490.0313520570<0.01135131–40.5513590–45<0.01
hydrophilic pocket BSA/HAb6.546.02–0.520.036.695.81–0.88<0.016.547.741.20<0.016.697.670.980.52
a

Entries in bold have significant p values.

b

Properties of the pocket are divided by the number of nonhydrogen atoms in the ligands, not the HA in the pocket itself.

Physicochemical Properties of GOOD versus UNDER Complexes

Very few statistically significant differences were observed between properties of the GOOD and UNDER sets. Their ligands are roughly the same size, which is in stark disagreement with the NULL case. Many physicochemical properties are proportional to size, so the fact that UNDER and GOOD ligands are similar in size makes our key comparisons between the sets more straightforward. However, comparisons to the NULL case must be done cautiously because high-affinity NULLs are much larger. Therefore, all properties were examined by the raw values and values corrected for size by dividing by the number of ligand heavy atoms (HA).
A striking difference between GOOD and UNDER ligands is the fact that both have roughly the same number of rotatable bonds (Nrot and Nrot/HA), but UNDERs have much lower torsional energies (Etor). There is less torsional strain in the UNDER ligands. Even when corrected for size (Etor/Nrot), UNDERs are less strained than the high-affinity ligands in the NULL case.
Lipinski (80) and Oprea (81) described various counts of calculated properties that aid in indentifying compounds with good oral absorption (Lipinski) and drug-like compounds (Oprea). A count of the number of ligands in each set that violate these empirical rules shows that UNDER ligands are more drug-like than the GOOD ligands and the high-affinity set in the NULL case. UNDERs are more lipophilic than GOODs (higher SlogP and lower log S), but the difference is more pronounced in the NULL case. The counts of hydrophobic and hydrogen-bonding atoms are not significantly different between UNDER and GOOD ligands, unlike the NULL case. What is most striking is that—despite the ligands being roughly the same size with the same number of hydrogen-bonding features—there are significantly fewer hydrogen bonds between the protein and the ligand in UNDER complexes. The pockets of UNDER complexes contain fewer water molecules as well. Many of the trends for hydrogen bonding indicate a more hydrophobic environment for the UNDER pockets but just miss the arbitrary cutoff of p = 0.05. There is significantly less hydrophilic buried surface area (BSA) in the UNDER pockets, which shifts the %hydrophilic and %hydrophobic BSA. It is unclear if these trends in the pocket BSA are a reflection of the NULL case because the GOOD and UNDER ligands are very similar, whereas the mid- and high-affinity sets are starkly different. In the same vein, the hydrophobic vdW surface per HA of the ligand is larger for UNDERs, but it is also similar to the value obtained for the high-affinity complexes in the NULL set.

Physicochemical Properties of GOOD versus OVER Complexes

Like the UNDERs, OVER ligands are roughly the same size as the GOOD ligands. There is a large size difference in the NULL case, but with low-affinity ligands being much smaller than the midrange, so again, comparisons to the NULL must be made cautiously.
OVER complexes have two patterns in opposition to the UNDER complexes. OVER ligands have higher Etor/Nrot, making them more strained. Also, OVER ligands are much more hydrophilic and soluble than GOOD (or UNDER) complexes. While the UNDER ligands had no differences in the count of hydrophobic and hydrophilic atoms, the OVER ligands are very different from the GOOD. There are fewer hydrophobic atoms and notably more oxygen atoms. On a per HA basis, there are more protein–ligand hydrogen bonds and more bridging water. The ligands have less hydrophobic vdW SA/HA and more polar vdW SA/HA. For the pockets, the hydrophilic BSA increases roughly the same degree as the UNDER complexes decrease, but the p value is just shy of the 0.05 cutoff.
In the NULL case of mid- vs low-affinity complexes, there are many SA properties that are significantly different, but many are due to the large size difference in the NULL case that is not seen in OVER vs GOOD complexes. There is one interesting trend in the pockets of the NULL case. The low-affinity complexes have more exposed ligands (ESA/HA). Though the OVER complexes are more exposed than the GOOD, it is not statistically significant nor is it as extreme as the trend in the NULLs.

Comparison of Amino Acids in the Binding Sites

Figure 6 shows the distribution of amino acids in the binding sites of GOOD, UNDER, OVER, and NULL sets. The comparison of UNDER to GOOD binding sites shows that the increase in hydrophobic character of the pockets comes from a marked increase in the aliphatic residues Val, Ile, and Leu and not a change in the aromatic amino acids. However, large contributions from Ile and Leu are also seen in high-affinity NULLs. The decrease in hydrogen-bonding interactions and hydrophilic BSA for UNDER complexes comes from significant decreases in Lys, Arg, and Ser. It is very interesting that there is a decrease in the positively charged residues but not the presence of the acidic amino acids. While there is a decrease in Asn and Gln, this is also seen in the NULL set.

Figure 6

Figure 6. Distribution of amino acids in the binding sites of the GOOD and BAD complexes meeting the ≥12 of 17 definition (left) are compared to those of the NULL case (right). The graph in the lower left provides the distribution of all amino acids in the full protein sequences to show that the important trends do not result from inherent differences in composition of the proteins (the same is true of the NULLs, data not shown). Metals and modified residues are denoted as other, “OTH”. Averages and error bars for the amino acid content were determined by bootstrapping.

High Resolution Image
Download MS PowerPoint Slide
The comparison between physicochemical properties of OVER and GOOD complexes revealed more hydrophilic ligands and pockets for the OVERs. However, the only significant difference in the composition of the binding sites is more Ser in the OVER pockets. There are decreases in Gly and Ile, but they are in good agreement with the content of low-affinity NULL pockets. It is possible that the similarity in the amino-acid composition may explain why the OVERs score too well. The ligands are the same size as GOOD ligands, not small like the low-affinity NULL ligands, and the pockets are more similar to GOODs than to the NULLs.

Impact on CSAR’s Future Data Sets

CSAR’s goal is to provide better, more complete data to the computational community, a widely available resource that scientists can use to improve their docking and scoring methods. This analysis across all participants in the first benchmark exercise is intended to shed light on the most pressing needs of the field as a whole. It is clear that hydrogen-bonding features should be our immediate focus, followed by rotatable bonds.
Our comparison of GOOD and BAD complexes shows that the field is underestimating the impact of hydrophobic interactions and overestimating the contribution of hydrogen bonding. This was supported by the contributions of three participants who showed that the correlation to affinity was unchanged or improved when hydrogen-bonding/Coulombic interactions were removed from their chosen scoring functions (data not shown). Scoring is usually based only on the bound complex, but binding is an equilibrium between bound and unbound states. A hydroxyl group can be added to a ligand to improve its interaction with a binding site, but experiment often shows little change in affinity. (82) The added hydrogen bonding also favors the interaction with water in the unbound state, and it is very difficult for a binding site to provide better hydrogen bonding than water. The electrostatic interactions in most scoring functions provide very favorable contributions to the estimated affinity because desolvation penalties are often overlooked. Desolvation is also an important driving force for hydrophobic association, providing a boost that simple vdW energies cannot model.
It is important that we provide data for many different protein targets. The affinities of the ligands must range at least 3 orders of magnitude so that experimental error will not limit the development of statistically meaningful models. (83) For each protein, we want to provide data on at least three congeneric series of ligands which provide the widest range of hydrogen-bonding features possible for the target. This should have a higher impact than varying size of the ligands. Of course, the data needed to tackle rotatable bonds will likely require ligand series with a range of sizes.
Our in-house efforts to enhance available data include isothermal titration calorimetry (ITC) and the determination of binding kinetics (www.CSARdock.org). These complementary techniques are independent means of determining exact binding constants, as opposed to inhibition constants or IC50s. They each provide valuable insights into the contributions to binding: entropy and enthalpy for calorimetry and kon and koff for kinetics. These data are particularly important for understanding the hydrophobic effects and hydrogen bonding implicated here. Furthermore, ITC can be used to determine changes in protonation states upon ligand binding. (84) This complication is a factor for a few of the BAD complexes. For instance, the crystal structure 1tok (85) (set 2, complex no. 96) is aspartate transferase binding maleic acid (HOOC–HC═CH–COOH). In the binding site, both ends of the diacid are complemented by bidentate hydrogen bonding from Arg side chains, indicating that the ligand is doubly deprotonated. However, maleic acid can be singly deprotonated in solution, depending on the conditions (second pKa of maleic acid is ∼6.3). (86)

Conclusion

Click to copy section linkSection link copied!
The most common test to evaluate docking and scoring involves relatively ranking compounds against a single target because pharma tackles this practical challenge every day. However, the findings of those exercises are often system-dependent and cannot necessarily be extrapolated. Tackling absolute free energies of binding across diverse proteins is difficult, but we have the potential to learn something new by posing unique challenges. We have outlined a means for statistically evaluating our data set across multiple methods, emphasizing the insights possible by combining the results of many participants.
These insights help us prioritize the design of new data sets to address specific shortcomings of our methods. If the answer could be found by conducting individual, single-target studies, then the solutions would have been found long ago. It is important to keep an eye on the big picture—the global landscape of docking and scoring—to understand what model systems are most needed to improve the field.
Future benchmarks from CSAR will involve blind rankings of chosen model systems, sets of data designed to address the shortcomings we identify as a community. As always, we will strive to provide as many systems as possible to avoid system-dependent insights. Confirmed data on inactive compounds will be provided. We greatly appreciate the efforts of all of our colleagues in the pharmaceutical industry for the donation of data for these future benchmark exercises.

Supporting Information

Click to copy section linkSection link copied!

Table of 10 proteins with ligand series and the performance of each of the 17 core codes on relative ranking, a discussion of methods and metrics for identifying the GOOD complexes, and a complete listing of GOOD and BAD complexes. This material is available free of charge via the Internet at http://pubs.acs.org.

Terms & Conditions

Most electronic Supporting Information files are available without a subscription to ACS Web Editions. Such files may be downloaded by article for research use (if there is a public use license linked to the relevant article, that license may permit other uses). Permission may be obtained from ACS for other uses through requests via the RightsLink permission system: http://pubs.acs.org/page/copyright/permissions.html.

Author Information

Click to copy section linkSection link copied!
  • Corresponding Author
    • Heather A. Carlson - Department of Medicinal Chemistry, University of Michigan, 428 Church Street, Ann Arbor, Michigan 48109-1065, United States;  Email: [email protected]
  • Authors
    • Richard D. Smith - Department of Medicinal Chemistry, University of Michigan, 428 Church Street, Ann Arbor, Michigan 48109-1065, United States
    • James B. Dunbar Jr. - Department of Medicinal Chemistry, University of Michigan, 428 Church Street, Ann Arbor, Michigan 48109-1065, United States
    • Peter Man-Un Ung - Department of Medicinal Chemistry, University of Michigan, 428 Church Street, Ann Arbor, Michigan 48109-1065, United States
    • Emilio X. Esposito - Department of Medicinal Chemistry, University of Michigan, 428 Church Street, Ann Arbor, Michigan 48109-1065, United States
    • Chao-Yie Yang - Department of Internal Medicine−Hematology/Oncology, University of Michigan, 1500 East Medical Center Drive, 7216 CC, Ann Arbor, Michigan 48109-0934, United States
    • Shaomeng Wang - Department of Internal Medicine−Hematology/Oncology, University of Michigan, 1500 East Medical Center Drive, 7216 CC, Ann Arbor, Michigan 48109-0934, United States

Acknowledgment

Click to copy section linkSection link copied!

We would like to thank all participants in the benchmark exercise! Whether you submitted a paper to this special issue, gave a talk at the symposium, submitted scores for this analysis, or just attended the talks and the discussions at the symposium, everyone’s feedback was valuable to our efforts. We thank numerous colleagues for helpful discussions, particularly John Liebeschuetz (CCDC) for his insights in Factor Xa. The CSAR Center is funded by the National Institute of General Medical Sciences (U01 GM086873). We also thank the Chemical Computing Group and OpenEye Scientific Software for generously donating the use of their software.

Added in Proof

Click to copy section linkSection link copied!

We thank Yu Zhou of the National Institute of Biological Sciences, Beijing for informing us that the ligand in 1x8d (set 2, no. 121) was incorrectly protonated. It is a sugar, and one of the hydroxyl groups was misinterpreted to be a ketone. As one might expect, it was indeed one of the BAD structures, improperly scored across most methods. A corrected version is available for download on the CSAR website (www.CSARdock.org, accessed August 24, 2011).

References

Click to copy section linkSection link copied!

This article references 86 other publications.

  1. 1
    Leach, A. R.; Shoichet, B. K.; Peishoff, C. E. Prediction of protein-ligand interactions. Docking and scoring: successes and gaps J. Med. Chem. 2006, 49, 5851– 5855
    Google Scholar
    1
    Prediction of protein-ligand interactions. Docking and scoring: Successes and gaps
    Leach, Andrew R.; Shoichet, Brian K.; Peishoff, Catherine E.
    Journal of Medicinal Chemistry (2006), 49 (20), 5851-5855CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
    A review. The current state of docking and scoring technol. are summarized, highlighting the key areas that must be addressed. Benchmarks and measurements to take up the data infrastructure on which new algorithm development depends; and possible industrial, academic, and governmental partnerships that may be needed to fulfill the potential of docking and scoring, are discussed.
  2. 2
    Warren, G. L.; Andrews, C. W.; Capelli, A.-M.; Clarke, B.; LaLonde, J.; Lambert, M. H.; Lindvall, M.; Nevins, N.; Semus, S. F.; Senger, S.; Tedesco, G.; Wall, I. D.; Woolven, J. M.; Peishoff, C. E.; Head, M. S. A critical assessment of docking programs and scoring functions J. Med. Chem. 2006, 49, 5912– 5931
    Google Scholar
    2
    A Critical Assessment of Docking Programs and Scoring Functions
    Warren, Gregory L.; Andrews, C. Webster; Capelli, Anna-Maria; Clarke, Brian; LaLonde, Judith; Lambert, Millard H.; Lindvall, Mika; Nevins, Neysa; Semus, Simon F.; Senger, Stefan; Tedesco, Giovanna; Wall, Ian D.; Woolven, James M.; Peishoff, Catherine E.; Head, Martha S.
    Journal of Medicinal Chemistry (2006), 49 (20), 5912-5931CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
    Docking is a computational technique that samples conformations of small mols. in protein binding sites; scoring functions are used to assess which of these conformations best complements the protein binding site. An evaluation of 10 docking programs and 37 scoring functions was conducted against eight proteins of seven protein types for three tasks: binding mode prediction, virtual screening for lead identification, and rank-ordering by affinity for lead optimization. All of the docking programs were able to generate ligand conformations similar to crystallog. detd. protein/ligand complex structures for at least one of the targets. However, scoring functions were less successful at distinguishing the crystallog. conformation from the set of docked poses. Docking programs identified active compds. from a pharmaceutically relevant pool of decoy compds.; however, no single program performed well for all of the targets. For prediction of compd. affinity, none of the docking programs or scoring functions made a useful prediction of ligand binding affinity.
  3. 3
    Dunbar, J. B., Jr.; Smith, R. D.; Yang, C. Y.; Ung, P. M.; Lexa, K. W.; Khazanov, N. A.; Stuckey, J. A.; Wang, S.; Carlson, H. A. CSAR Benchmark Exercise of 2010: Selection of the protein-ligand complexes J. Chem. Inf. Model. 2011,  DOI: 10.1021/ci200082t
    Google Scholar
    There is no corresponding record for this reference.
  4. 4
    Wang, R.; Lu, Y.; Wang, S. Comparative evaluation of 11 scoring functions for molecular docking J. Med. Chem. 2003, 46, 2287– 2303
    Google Scholar
    4
    Comparative Evaluation of 11 Scoring Functions for Molecular Docking
    Wang, Renxiao; Lu, Yipin; Wang, Shaomeng
    Journal of Medicinal Chemistry (2003), 46 (12), 2287-2303CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
    Eleven popular scoring functions have been tested on 100 protein-ligand complexes to evaluate their abilities to reproduce exptl. detd. structures and binding affinities. They include four scoring functions implemented in the LigFit module in Cerius2 (LigScore, PLP, PMF, and LUDI), four scoring functions implemented in the CScore module in SYBYL (F-Score, G-Score, D-Score, and ChemScore), the scoring function implemented in the AutoDock program, and two stand-alone scoring functions (DrugScore and X-Score). These scoring functions are not tested in the context of a particular docking program. Instead, conformational sampling and scoring are sepd. into two consecutive steps. First, an exhaustive conformational sampling is performed by using the AutoDock program to generate an ensemble of docked conformations for each ligand mol. This conformational ensemble is required to cover the entire conformational space as much as possible rather than to focus on a few energy min. Then, each scoring function is applied to score this conformational ensemble to see if it can identify the exptl. obsd. conformation from all of the other decoys. Among all of the scoring functions under test, six of them, i.e., PLP, F-Score, LigScore, DrugScore, LUDI, and X-Score, yield success rates higher than the AutoDock scoring function. The success rates of these six scoring functions range from 66% to 76% if using root-mean-square deviation ≤2.0 Å as the criterion. Combining any two or three of these six scoring functions into a consensus scoring scheme further improves the success rate to nearly 80% or even higher. However, when applied to reproduce the exptl. detd. binding affinities of the 100 protein-ligand complexes, only X-Score, PLP, DrugScore, and G-Score are able to give correlation coeffs. over 0.50. All of the 11 scoring functions are further inspected by their abilities to construct a descriptive, funnel-shaped energy surface for protein-ligand complexation. The results indicate that X-Score and DrugScore perform better than the other ones at this aspect.
  5. 5
    Muchmore, S. W.; Debe, D. A.; Metz, J. T.; Brown, S. P.; Martin, Y. C.; Hajduk, P. J. Application of belief theory to similarity data fusion for use in analog searching and lead hopping J. Chem. Inf. Model. 2008, 48, 941– 948
    Google Scholar
    5
    Application of Belief Theory to Similarity Data Fusion for Use in Analog Searching and Lead Hopping
    Muchmore, Steven W.; Debe, Derek A.; Metz, James T.; Brown, Scott P.; Martin, Yvonne C.; Hajduk, Philip J.
    Journal of Chemical Information and Modeling (2008), 48 (5), 941-948CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
    A wide variety of computational algorithms have been developed that strive to capture the chem. similarity between two compds. for use in virtual screening and lead discovery. One limitation of such approaches is that, while a returned similarity value reflects the perceived degree of relatedness between any two compds., there is no direct correlation between this value and the expectation or confidence that any two mols. will in fact be equally active. A lack of a common framework for interpretation of similarity measures also confounds the reliable fusion of information from different algorithms. Here, we present a probabilistic framework for interpreting similarity measures that directly correlates the similarity value to a quant. expectation that two mols. will in fact be equipotent. The approach is based on extensive benchmarking of 10 different similarity methods (MACCS keys, Daylight fingerprints, max. common subgraphs, rapid overlay of chem. structures (ROCS) shape similarity, and six connectivity-based fingerprints) against a database of more than 150 000 compds. with activity data against 23 protein targets. Given this unified and probabilistic framework for interpreting chem. similarity, principles derived from decision theory can then be applied to combine the evidence from different similarity measures in such a way that both capitalizes on the strengths of the individual approaches and maintains a quant. est. of the likelihood that any two mols. will exhibit similar biol. activity.
  6. 6
    Muchmore, S. W.; Edmunds, J. J.; Stewart, K. D.; Hajduk, P. J. Cheminformatic tools for medicinal chemists J. Med. Chem. 2010, 53, 4830– 4841
    Google Scholar
    There is no corresponding record for this reference.
  7. 7
    Swann, S. L.; Brown, S. P.; Muchmore, S. W.; Patel, H.; Merta, P.; Locklear, J.; Hajduk, P. J. A unified, probabilistic framework for structure- and ligand-based virtual screening J. Med. Chem. 2011, 54, 1223– 1232
    Google Scholar
    7
    A Unified, Probabilistic Framework for Structure- and Ligand-Based Virtual Screening
    Swann, Steven L.; Brown, Scott P.; Muchmore, Steven W.; Patel, Hetal; Merta, Philip; Locklear, John; Hajduk, Philip J.
    Journal of Medicinal Chemistry (2011), 54 (5), 1223-1232CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
    The authors present a probabilistic framework for interpreting structure-based virtual screening that returns a quant. likelihood of observing bioactivity and can be quant. combined with ligand-based screening methods to yield a cumulative prediction that consistently outperforms any single screening metric. The approach has been developed and validated on more than 30 different protein targets. Transforming structure-based in silico screening results into robust probabilities of activity enables the general fusion of multiple structure- and ligand-based approaches and returns a quant. expectation of success that can be used to prioritize (or deprioritize) further discovery activities. This unified probabilistic framework offers a paradigm shift in how docking and scoring results are interpreted, which can enhance early lead-finding efforts by maximizing the value of in silico computational tools.
  8. 8
    Baber, J. C.; Shirley, W. A.; Gao, Y.; Feher, M. The use of consensus scoring in ligand-based virtual screening J. Chem. Inf. Model. 2006, 46, 277– 288
    Google Scholar
    8
    The use of consensus scoring in ligand-based virtual screening
    Baber, J. Christian; Shirley, William A.; Gao, Yinghong; Feher, Miklos
    Journal of Chemical Information and Modeling (2006), 46 (1), 277-288CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
    A new consensus approach has been developed for ligand-based virtual screening. It involves combining highly disparate properties in order to improve performance in virtual screening. The properties include structural, 2D pharmacophore and property-based fingerprints, scores derived using BCUT descriptors, and 3D pharmacophore approaches. Different approaches for the combination of all or some of these methods have been tested. Logistic regression and sum ranks were found to be the most advantageous in different pharmaceutical applications. The three major reasons consensus scoring appears to enrich data sets better than single scoring functions are (1) using multiple scoring functions is similar to repeated samplings, in which case the mean is closer to the true value than any single value, (2) due to the better clustering of actives, multiple sampling will recover more actives than inactives, and (3) different methods seem to agree more on the ranking of the actives than on the inactives. Furthermore, consensus results are not only better but are also more consistent across receptor systems.
  9. 9
    Bar-Haim, S.; Aharon, A.; Ben-Moshe, T.; Marantz, Y.; Senderowitz, H. SeleX-CS: a new consensus scoring algorithm for hit discovery and lead optimization J. Chem. Inf. Model. 2009, 49, 623– 633
    Google Scholar
    9
    SeleX-CS: A New Consensus Scoring Algorithm for Hit Discovery and Lead Optimization
    Bar-Haim, Shay; Aharon, Ayelet; Ben-Moshe, Tal; Marantz, Yael; Senderowitz, Hanoch
    Journal of Chemical Information and Modeling (2009), 49 (3), 623-633CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
    Identifying active compds. (hits) that bind to biol. targets of pharmaceutical relevance is the cornerstone of drug design efforts. Structure based virtual screening, namely, the in silico evaluation of binding energies and geometries between a protein and its putative ligands, has emerged over the past few years as a promising approach in this field. The success of the method relies on the availability of reliable 3-dimensional (3D) structures of the target protein and its candidate ligands (the screening library), a reliable docking method that can fit the different ligands into the protein's binding site, and an accurate scoring function that can rank the resulting binding modes in accord with their binding affinities. This last requirement is arguably the most difficult to meet due to the complexity of the binding process. A potential soln. to this so-called scoring problem is the usage of multiple scoring functions in an approach known as consensus scoring. Several consensus scoring methods were suggested in the literature and have generally demonstrated an improved ranking of screening libraries relative to individual scoring functions. Nevertheless, current consensus scoring strategies suffer from several shortcomings, in particular, strong dependence on the initial parameters and an incomplete treatment of inactive compds. In this work we present a new consensus scoring algorithm (SeleX-Consensus Scoring abbreviated to SeleX-CS) specifically designed to address these limitations. (i) A subset of the initial set of the scoring functions is allowed to form the consensus score, and this subset is optimized via a Monte Carlo/Simulated Annealing procedure. (ii) Rank redundancy between the members of the screening library is removed. (iii) The method explicitly considers the presence of inactive compds. The new algorithm was applied to the ranking of screening libraries targeting two G-protein coupled receptors (GPCR). Excellent enrichment factors were obtained in both cases: For the cannabinoid receptor 1 (CB1), SeleX-CS outperformed the best single score and afforded an enrichment factor of 41 at 1% of the screening library compared with the best single score value of 15 (GOLD_Fitness). For the chemokine receptor type 2 (CCR2) SeleX-CS afforded an enrichment factor of 72 (again at 1% of the screening library) once more outperforming any single score (enrichment factor of 20 by G_SCORE). Moreover, SeleX-CS demonstrated success rates of 67% (CCR2) and 73% (CB1) when applied to ranking an external test set. In both cases, the new algorithm also afforded good derichment of inactive compds. (i.e., the ability to push inactive compds. to the bottom of the ranked library). The method was then extended to rank a lead optimization series targeting the Kv4.3 potassium ion channel, resulting in a Spearman's correlation coeff., ρ = 0.63 (n = 40), between the SeleX-CS-based rank and the actual pKi values. These results suggest that SeleX-CS is a powerful method for ranking screening libraries in the lead discovery phase and also merits consideration as a lead optimization tool.
  10. 10
    Betzi, S.; Suhre, K.; Chetrit, B.; Guerlesquin, F.; Morelli, X. GFscore: a general nonlinear consensus scoring function for high-throughput docking J. Chem. Inf. Model. 2006, 46, 1704– 1712
    Google Scholar
    10
    GFscore: A General Nonlinear Consensus Scoring Function for High-Throughput Docking
    Betzi, Stephane; Suhre, Karsten; Chetrit, Bernard; Guerlesquin, Francoise; Morelli, Xavier
    Journal of Chemical Information and Modeling (2006), 46 (4), 1704-1712CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
    Most of the recent published works in the field of docking and scoring protein/ligand complexes have focused on ranking true positives resulting from a Virtual Library Screening (VLS) through the use of a specified or consensus linear scoring function. In this work, the authors present a methodol. to speed up the High Throughput Screening (HTS) process, by allowing focused screens or for hit list triaging when a prohibitively large no. of hits is identified in the primary screen, where the authors have extended the principle of consensus scoring in a nonlinear neural network manner. This led us to introduce a nonlinear Generalist scoring Function, GFscore, which was trained to discriminate true positives from false positives in a data set of diverse chem. compds. This original Generalist scoring Function is a combination of the five scoring functions found in the CScore package from Tripos Inc. GFscore eliminates up to 75% of mols., with a confidence rate of 90%. The final result is a Hit Enrichment in the list of mols. to investigate during a research campaign for biol. active compds. where the remaining 25% of mols. would be sent to in vitro screening expts. GFscore is therefore a powerful tool for the biologist, saving both time and money.
  11. 11
    Bissantz, C.; Folkers, G.; Rognan, D. Protein-based virtual screening of chemical databases. 1. Evaluation of different docking/scoring combinations J. Med. Chem. 2000, 43, 4759– 4767
    Google Scholar
    11
    Protein-Based Virtual Screening of Chemical Databases. 1. Evaluation of Different Docking/Scoring Combinations
    Bissantz, Caterina; Folkers, Gerd; Rognan, Didier
    Journal of Medicinal Chemistry (2000), 43 (25), 4759-4767CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
    Three different database docking programs (Dock, FlexX, Gold) have been used in combination with seven scoring functions (Chemscore, Dock, FlexX, Fresno, Gold, Pmf, Score) to assess the accuracy of virtual screening methods against two protein targets (thymidine kinase, estrogen receptor) of known three-dimensional structure. For both targets, it was generally possible to discriminate about 7 out of 10 true hits from a random database of 990 ligands. The use of consensus lists common to two or three scoring functions clearly enhances hit rates among the top 5% scorers from 10% (single scoring) to 25-40% (double scoring) and up to 65-70% (triple scoring). However, in all tested cases, no clear relationships could be found between docking and ranking accuracies. Moreover, predicting the abs. binding free energy of true hits was not possible whatever docking accuracy was achieved and scoring function used. As the best docking/consensus scoring combination varies with the selected target and the physicochem. of target-ligand interactions, we propose a two-step protocol for screening large databases: (i) screening of a reduced dataset contg. a few known ligands for deriving the optimal docking/consensus scoring scheme, (ii) applying the latter parameters to the screening of the entire database.
  12. 12
    Charifson, P. S.; Corkery, J. J.; Murcko, M. A.; Walters, W. P. Consensus scoring: A method for obtaining improved hit rates from docking databases of three-dimensional structures into proteins J. Med. Chem. 1999, 42, 5100– 5109
    Google Scholar
    12
    Consensus Scoring: A Method for Obtaining Improved Hit Rates from Docking Databases of Three-Dimensional Structures into Proteins
    Charifson, Paul S.; Corkery, Joseph J.; Murcko, Mark A.; Walters, W. Patrick
    Journal of Medicinal Chemistry (1999), 42 (25), 5100-5109CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
    The authors present the results of an extensive computational study in which the authors show that combining scoring functions in an intersection-based consensus approach results in an enhancement in the ability to discriminate between active and inactive enzyme inhibitors. This is illustrated in the context of docking collections of three-dimensional structures into three different enzymes of pharmaceutical interest: p38 MAP kinase, inosine monophosphate dehydrogenase, and HIV protease. An anal. of two different docking methods and thirteen scoring functions provides insights into which functions perform well, both singly and in combination. The data shows that consensus scoring further provides a dramatic redn. in the no. of false positives identified by individual scoring functions, thus leading to a significant enhancement in hit-rates.
  13. 13
    Clark, R. D.; Strizhev, A.; Leonard, J. M.; Blake, J. F.; Matthew, J. B. Consensus scoring for ligand/protein interactions J. Mol. Graphics Modell. 2002, 20, 281– 295
    Google Scholar
    13
    Consensus scoring for ligand/protein interactions
    Clark, Robert D.; Strizhev, Alexander; Leonard, Joseph M.; Blake, James F.; Matthew, James B.
    Journal of Molecular Graphics & Modelling (2002), 20 (4), 281-295CODEN: JMGMFI; ISSN:1093-3263. (Elsevier Science Inc.)
    Several different functions have been put forward for evaluating the energetics of ligand binding to proteins. Those employed in the DOCK, GOLD and FlexX docking programs have been esp. widely used, particularly in connection with virtual high-throughput screening (vHTS) projects. Until recently, such evaluation functions were usually considered only in conjunction with the docking programs that relied on them. In such studies, the evaluation function in question actually fills two distinct roles: it serves as the objective function being optimized (fitness function), but is also the scoring function used to compare the candidate docking configurations generated by the program. We have used descriptions available in the open literature to create free-standing scoring functions based on those used in DOCK and GOLD, and have implemented the more recently formulated PMF [J. Med. Chem. 42 (1999) 791] scoring function as well. The performance of these functions was examd. individually for each of several data sets for which both crystal structures and affinities are available, as was the performance of the FlexX scoring function. Various ways of combining individual scores into a consensus score (CScore) were also considered. The individual and consensus scores were also used to try to pick out configurations most similar to those found in crystal structures from among a set of candidate configurations produced by FlexX docking runs. We find that the reliability and interpretability of results can be improved by combining results from all four functions into a CScore.
  14. 14
    Feher, M. Consensus scoring for protein-ligand interactions Drug Discovery Today 2006, 11, 421– 428
    Google Scholar
    14
    Consensus scoring for protein-ligand interactions
    Feher, Miklos
    Drug Discovery Today (2006), 11 (9 & 10), 421-428CODEN: DDTOFS; ISSN:1359-6446. (Elsevier)
    A review. This article reviews the application of consensus scoring for cases when the target 3D structure is known. Comparing the performance of different methods is not a trivial task, and it appears that consensus scoring usually substantially improves virtual screening performance, contributing to better enrichments. It also seems to improve - albeit less dramatically - the prediction of bound conformations and poses. The prediction of binding energies is still rather inaccurate and although consensus scoring generally improves these predictions, more development is required before it can be used for this purpose in routine lead optimization.
  15. 15
    Garcia-Sosa, A. T.; Sild, S.; Maran, U. Design of multi-binding-site inhibitors, ligand efficiency, and consensus screening of avian influenza H5N1 wild-type neuraminidase and of the oseltamivir-resistant H274Y variant J. Chem. Inf. Model. 2008, 48, 2074– 2080
    Google Scholar
    There is no corresponding record for this reference.
  16. 16
    Krovat, E. M.; Langer, T. Impact of scoring functions on enrichment in docking-based virtual screening: an application study on renin inhibitors J. Chem. Inf. Comput. Sci. 2004, 44, 1123– 1129
    Google Scholar
    16
    Impact of Scoring Functions on Enrichment in Docking-Based Virtual Screening: An Application Study on Renin Inhibitors
    Krovat, Eva M.; Langer, Thierry
    Journal of Chemical Information and Computer Sciences (2004), 44 (3), 1123-1129CODEN: JCISD8; ISSN:0095-2338. (American Chemical Society)
    The docking program LigandFit/Cerius2 has been used to perform shape-based virtual screening of databases against the aspartic protease renin, a target of detd. three-dimensional structure. The protein structure was used in the induced fit binding conformation that occurs when renin is bound to the highly active renin inhibitor (IC50 = 2 nM). The scoring was calcd. using several different scoring functions to get insight into the predictability of the magnitude of binding interactions. A database of 1000 diverse and drug-like compds., comprised of 990 members of a virtual database generated by using the iLib diverse software and 10 known active renin inhibitors, was docked flexibly and scored to det. appropriate scoring functions. All seven scoring functions used (LigScore1, LigScore2, PLP1, PLP2, JAIN, PMF, LUDI) were able to retrieve at least 50% of the active compds. within the first 20% (200 mols.) of the entire test database. A hit rate of 90% in the top 1.4% resulted using the quadruple consensus scoring of LigScore2, PLP1, PLP2, and JAIN. Addnl., a focused database was created with the iLib diverse software and used for the same procedure as the test database. Docking and scoring of the 990 focused compds. and the 10 known actives were performed. A hit rate of 100% in the top 8.4% resulted with use of the triple consensus scoring of PLP1, PLP2, and PMF. As expected, a ranking of the known active compds. within the focused database compared to the test database was obsd. Adequate virtual screening conditions were derived empirically. They can be used for proximate docking and scoring application of compds. with putative renin inhibiting potency.
  17. 17
    Oda, A.; Tsuchida, K.; Takakura, T.; Yamaotsu, N.; Hirono, S. Comparison of consensus scoring strategies for evaluating computational models of protein-ligand complexes J. Chem. Inf. Model. 2006, 46, 380– 391
    Google Scholar
    17
    Comparison of Consensus Scoring Strategies for Evaluating Computational Models of Protein-Ligand Complexes
    Oda, Akifumi; Tsuchida, Keiichi; Takakura, Tadakazu; Yamaotsu, Noriyuki; Hirono, Shuichi
    Journal of Chemical Information and Modeling (2006), 46 (1), 380-391CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
    Here, the comparisons of performance of nine consensus scoring strategies, in which multiple scoring functions were used simultaneously to evaluate candidate structures for a protein-ligand complex, in combination with nine scoring functions (FlexX score, GOLD score, PMF score, DOCK score, ChemScore, DrugScore, PLP, ScreenScore, and X-Score), were carried out. The systematic naming of consensus scoring strategies was also proposed. The authors' results demonstrate that choosing the most appropriate type of consensus score is essential for model selection in computational docking; although the vote-by-no. strategy was an effective selection method, the no.-by-no. and rank-by-no. strategies were more appropriate when computational tractability was taken into account. By incorporating these consensus scores into the FlexX program, reasonable complex models can be obtained more efficiently than those selected by independent FlexX scores. These strategies might also improve the scoring of other docking programs, and more-effective structure-based drug design should result from these improvements.
  18. 18
    Omigari, K.; Mitomo, D.; Kubota, S.; Nakamura, K.; Fukunishi, Y. A method to enhance the hit ratio by a combination of structure-based drug screening and ligand-based screening Adv. Appl. Bioinf. Chem. 2008, 1, 19– 28
    Google Scholar
    There is no corresponding record for this reference.
  19. 19
    Paul, N.; Rognan, D. ConsDock: A new program for the consensus analysis of protein-ligand interactions Proteins: Struct., Funct., Bioinf. 2002, 47, 521– 533
    Google Scholar
    There is no corresponding record for this reference.
  20. 20
    Renner, S.; Derksen, S.; Radestock, S.; Morchen, F. Maximum common binding modes (MCBM): consensus docking scoring using multiple ligand information and interaction fingerprints J. Chem. Inf. Model. 2008, 48, 319– 332
    Google Scholar
    20
    Maximum Common Binding Modes (MCBM): Consensus Docking Scoring Using Multiple Ligand Information and Interaction Fingerprints
    Renner, Steffen; Derksen, Swetlana; Radestock, Sebastian; Moerchen, Fabian
    Journal of Chemical Information and Modeling (2008), 48 (2), 319-332CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
    Improving the scoring functions for small mol.-protein docking is a highly challenging task in current computational drug design. Here we present a novel consensus scoring concept for the prediction of binding modes for multiple known active ligands. Similar ligands are generally believed to bind to their receptor in a similar fashion. The presumption of our approach was that the true binding modes of similar ligands should be more similar to each other compared to false pos. binding modes. The no. of conserved (consensus) interactions between similar ligands was used as a docking score. Patterns of interactions were modeled using ligand receptor interaction fingerprints. Our approach was evaluated for four different data sets of known cocrystal structures (CDK-2, dihydrofolate reductase, HIV-1 protease, and thrombin). Docking poses were generated with FlexX and rescored by our approach. For comparison the CScore scoring functions from Sybyl were used, and consensus scores were calcd. thereof. Our approach performed better than individual scoring functions and was comparable to consensus scoring. Anal. of the distribution of docking poses by self-organizing maps (SOM) and interaction fingerprints confirmed that clusters of docking poses composed of multiple ligands were preferentially obsd. near the native binding mode. Being conceptually unrelated to commonly used docking scoring functions our approach provides a powerful method to complement and improve computational docking expts.
  21. 21
    Teramoto, R.; Fukunishi, H. Structure-based virtual screening with supervised consensus scoring: evaluation of pose prediction and enrichment factors J. Chem. Inf. Model. 2008, 48, 747– 754
    Google Scholar
    21
    Structure-Based Virtual Screening with Supervised Consensus Scoring: Evaluation of Pose Prediction and Enrichment Factors
    Teramoto, Reiji; Fukunishi, Hiroaki
    Journal of Chemical Information and Modeling (2008), 48 (4), 747-754CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
    Since the evaluation of ligand conformations is a crucial aspect of structure-based virtual screening, scoring functions play significant roles in it. However, it is known that a scoring function does not always work well for all target proteins. When one cannot know which scoring function works best against a target protein a priori, there is no std. scoring method to know it even if 3D structure of a target protein-ligand complex is available. Therefore, development of the method to achieve high enrichments from given scoring functions and 3D structure of protein-ligand complex is a crucial and challenging task. To address this problem, we applied SCS (supervised consensus scoring), which employs a rough linear correlation between the binding free energy and the root-mean-square deviation (rmsd) of a native ligand conformations and incorporates protein-ligand binding process with docked ligand conformations using supervised learning, to virtual screening. We evaluated both the docking poses and enrichments of SCS and five scoring functions (F-Score, G-Score, D-Score, ChemScore, and PMF) for three different target proteins: thymidine kinase (TK), thrombin (thrombin), and peroxisome proliferator-activated receptor gamma (PPARγ). Our enrichment studies show that SCS is competitive or superior to a best single scoring function at the top ranks of screened database. We found that the enrichments of SCS could be limited by a best scoring function, because SCS is obtained on the basis of the five individual scoring functions. Therefore, it is concluded that SCS works very successfully from our results. Moreover, from docking pose anal., we revealed the connection between enrichment and av. centroid distance of top-scored docking poses. Since SCS requires only one 3D structure of protein-ligand complex, SCS will be useful for identifying new ligands.
  22. 22
    Teramoto, R.; Fukunishi, H. Consensus scoring with feature selection for structure-based virtual screening J. Chem. Inf. Model. 2008, 48, 288– 295
    Google Scholar
    22
    Consensus Scoring with Feature Selection for Structure-Based Virtual Screening
    Teramoto, Reiji; Fukunishi, Hiroaki
    Journal of Chemical Information and Modeling (2008), 48 (2), 288-295CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
    The evaluation of ligand conformations is a crucial aspect of structure-based virtual screening, and scoring functions play significant roles in it. While consensus scoring (CS) generally improves enrichment by compensating for the deficiencies of each scoring function, the strategy of how individual scoring functions are selected remains a challenging task when few known active compds. are available. To address this problem, the authors propose feature selection-based consensus scoring (FSCS), which performs supervised feature selection with docked native ligand conformations to select complementary scoring functions. The authors evaluated the enrichments of five scoring functions (F-Score, D-Score, PMF, G-Score, and ChemScore), FSCS, and RCS (rank-by-rank consensus scoring) for four different target proteins: acetylcholine esterase (AChE), thrombin, phosphodiesterase 5 (PDE5), and peroxisome proliferator-activated receptor gamma (PPARγ). The results indicated that FSCS was able to select the complementary scoring functions and enhance ligand enrichments and that it outperformed RCS and the individual scoring functions for all target proteins. They also indicated that the performances of the single scoring functions were strongly dependent on the target protein. An esp. favorable result with implications for practical drug screening is that FSCS performs well even if only one 3D structure of the protein-ligand complex is known. Moreover, the authors found that one can infer which scoring functions significantly enrich active compds. by using feature selection before actual docking and that the selected scoring functions are complementary.
  23. 23
    Wang, R.; Wang, S. How does consensus scoring work for virtual library screening? An idealized computer experiment J. Chem. Inf. Comput. Sci. 2001, 41, 1422– 1426
    Google Scholar
    23
    How Does Consensus Scoring Work for Virtual Library Screening? An Idealized Computer Experiment
    Wang, Renxiao; Wang, Shaomeng
    Journal of Chemical Information and Computer Sciences (2001), 41 (5), 1422-1426CODEN: JCISD8; ISSN:0095-2338. (American Chemical Society)
    It has been reported recently that consensus scoring, which combines multiple scoring functions in binding affinity estn., leads to higher hit-rates in virtual library screening studies. This method seems quite independent to the target receptor, the docking program, or even the scoring functions under investigation. Here we present an idealized computer expt. to explore how consensus scoring works. A hypothetical set of 5000 compds. is used to represent a chem. library under screening. The binding affinities of all its member compds. are assigned by mimicking a real situation. Based on the assumption that the error of a scoring function is a random no. in a normal distribution, the predicted binding affinities were generated by adding such a random no. to the "obsd." binding affinities. The relation between the hit-rates and the no. of scoring functions employed in scoring was then investigated. The performance of several typical ranking strategies for a consensus scoring procedure was also explored. Our results demonstrate that consensus scoring outperforms any single scoring for a simple statistical reason: the mean value of repeated samplings tends to be closer to the true value. Our results also suggest that a moderate no. of scoring functions, three or four, are sufficient for the purpose of consensus scoring. As for the ranking strategy, both the rank-by-no. and the rank-by-rank strategy work more effectively than the rank-by-vote strategy.
  24. 24
    Yang, J. M.; Chen, Y. F.; Shen, T. W.; Kristal, B. S.; Hsu, D. F. Consensus scoring criteria for improving enrichment in virtual screening J. Chem. Inf. Model. 2005, 45, 1134– 1146
    Google Scholar
    24
    Consensus Scoring Criteria for Improving Enrichment in Virtual Screening
    Yang, Jinn-Moon; Chen, Yen-Fu; Shen, Tsai-Wei; Kristal, Bruce S.; Hsu, D. Frank
    Journal of Chemical Information and Modeling (2005), 45 (4), 1134-1146CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
    Motivation: Virtual screening of mol. compd. libraries is a potentially powerful and inexpensive method for the discovery of novel lead compds. for drug development. The major weakness of virtual screening-the inability to consistently identify true positives (leads)-is likely due to our incomplete understanding of the chem. involved in ligand binding and the subsequently imprecise scoring algorithms. It has been demonstrated that combining multiple scoring functions (consensus scoring) improves the enrichment of true positives. Previous efforts at consensus scoring have largely focused on empirical results, but they have yet to provide a theor. anal. that gives insight into real features of combinations and data fusion for virtual screening. Results: The authors demonstrate that combining multiple scoring functions improves the enrichment of true positives only if (a) each of the individual scoring functions has relatively high performance and (b) the individual scoring functions are distinctive. Notably, these two prediction variables are previously established criteria for the performance of data fusion approaches using either rank or score combinations. This work, thus, establishes a potential theor. basis for the probable success of data fusion approaches to improve yields in in silico screening expts. Furthermore, it is similarly established that the second criterion (b) can, in at least some cases, be functionally defined as the area between the rank vs. score plots generated by the two (or more) algorithms. Because rank-score plots are independent of the performance of the individual scoring function, this establishes a second theor. defined approach to detg. the likely success of combining data from different predictive algorithms. This approach is, thus, useful in practical settings in the virtual screening process when the performance of at least two individual scoring functions (such as in criterion a) can be estd. as having a high likelihood of having high performance, even if no training sets are available. The authors provide initial validation of this theor. approach using data from five scoring systems with two evolutionary docking algorithms on four targets, thymidine kinase, human dihydrofolate reductase, and estrogen receptors of antagonists and agonists. Our procedure is computationally efficient, able to adapt to different situations, and scalable to a large no. of compds. as well as to a greater no. of combinations. Results of the expt. show a fairly significant improvement (vs. single algorithms) in several measures of scoring quality, specifically "goodness-of-hit" scores, false pos. rates, and "enrichment". This approach (available online at http://gemdock.life. nctu.edu.tw/dock/download.php) has practical utility for cases where the basic tools are known or believed to be generally applicable, but where specific training sets are absent.
  25. 25
    Hogg, R. V.; Tanis, E. A. Probability and Statistical Inference; Prentice Hall College Division: Englewood Cliffs, NJ, 2001, pp 402– 411.
    Google Scholar
    There is no corresponding record for this reference.
  26. 26
    Books of Abstracts; 240th American Chemical Society National Meeting, Boston, MA, August 22–28, 2010; ACS: Washington, D.C., 2010.
    Google Scholar
    There is no corresponding record for this reference.
  27. 27
    Benson, M. L.; Smith, R. D.; Khazanov, N. A.; Dimcheff, B.; Beaver, J.; Dresslar, P.; Nerothin, J.; Carlson, H. A. Binding MOAD, a high-quality protein-ligand database Nucleic Acids Res. 2008, 36, D674– D678
    Google Scholar
    There is no corresponding record for this reference.
  28. 28
    Hu, L.; Benson, M. L.; Smith, R. D.; Lerner, M. G.; Carlson, H. A. Binding MOAD (Mother Of All Databases) Proteins: Struct., Funct., Bioinf. 2005, 60, 333– 340
    Google Scholar
    28
    Binding MOAD (Mother of All Databases)
    Hu, Liegi; Benson, Mark L.; Smith, Richard D.; Lerner, Michael G.; Carlson, Heather A.
    Proteins: Structure, Function, and Bioinformatics (2005), 60 (3), 333-340CODEN: PSFBAF ISSN:. (Wiley-Liss, Inc.)
    Binding MOAD (Mother of All Databases) is the largest collection of high-quality, protein-ligand complexes available from the Protein Data Bank. At this time, Binding MOAD contains 5331 protein-ligand complexes comprised of 1780 unique protein families and 2630 unique ligands. We have searched the crystallog. papers for all 5000 + structures and compiled binding data for 1375 (26%) of the protein-ligand complexes. The binding-affinity data ranges 13 orders of magnitude. This is the largest collection of binding data reported to date in the literature. We have also addressed the issue of redundancy in the data. To create a nonredundant dataset, one protein from each of the 1780 protein families was chosen as a representative. Representatives were chosen by tightest binding, best resoln., etc. For the 1780 "best" complexes that comprise the nonredundant version of Binding MOAD, 475 (27%) have binding data. This significant collection of protein-ligand complexes will be very useful in elucidating the biophys. patterns of mol. recognition and enzymic regulation. The complexes with binding-affinity data will help in the development of improved scoring functions and structure-based drug discovery techniques. The dataset can be accessed at http://www.BindingMOAD.org.
  29. 29
    Wang, R.; Fang, X.; Lu, Y.; Wang, S. The PDBbind database: collection of binding affinities for protein-ligand complexes with known three-dimensional structures J. Med. Chem. 2004, 47, 2977– 2980
    Google Scholar
    29
    The PDBbind database: Collection of binding affinities for protein-ligand complexes with known three-dimensional structures
    Wang, Renxiao; Fang, Xueliang; Lu, Yipin; Wang, Shaomeng
    Journal of Medicinal Chemistry (2004), 47 (12), 2977-2980CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
    We have screened the entire Protein Data Bank (Release No. 103, Jan. 2003) and identified 5671 protein-ligand complexes out of 19 621 exptl. structures. A systematic examn. of the primary refs. of these entries has led to a collection of binding affinity data (Kd, Ki, and IC50) for a total of 1359 complexes. The outcomes of this project have been organized into a Web-accessible database named the PDBbind database.
  30. 30
    Wang, R.; Fang, X.; Lu, Y.; Yang, C.-Y.; Wang, S. The PDBbind database: methodologies and updates J. Med. Chem. 2005, 48, 4111– 4119
    Google Scholar
    30
    The PDBbind Database: Methodologies and Updates
    Wang, Renxiao; Fang, Xueliang; Lu, Yipin; Yang, Chao-Yie; Wang, Shaomeng
    Journal of Medicinal Chemistry (2005), 48 (12), 4111-4119CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
    The authors have developed the PDBbind database to provide a comprehensive collection of binding affinities for the protein-ligand complexes in the Protein Data Bank (PDB). This paper gives a full description of the latest version, i.e., version 2003, which is an update to our recently reported work. Out of 23 790 entries in the PDB release No.107 (Jan. 2004), 5897 entries were identified as protein-ligand complexes that meet our definition. Exptl. detd. binding affinities (Kd, Ki, and IC50) for 1622 of these were retrieved from the refs. assocd. with these complexes. A total of 900 complexes were selected to form a "refined set", which is of particular value as a std. data set for docking and scoring studies. All of the final data, including binding affinity data, ref. citations, and processed structural files, have been incorporated into the PDBbind database accessible online at http:// www.pdbbind.org/.
  31. 31
    Morris, G. M.; Goodsell, D. S.; Huey, R.; Olson, A. J. Distributed automated docking of flexible ligands to proteins: parallel applications of AutoDock 2.4 J. Comput.-Aided Mol. Des. 1996, 10, 293– 304
    Google Scholar
    31
    Distributed automated docking of flexible ligands to proteins: parallel applications of AutoDock 2.4
    Morris, Garrett M.; Goodsell, David S.; Huey, Ruth; Olson, Arthur J.
    Journal of Computer-Aided Molecular Design (1996), 10 (4), 293-304CODEN: JCADEQ; ISSN:0920-654X. (ESCOM)
    AutoDock 2.4 predicts the bound conformations of a small, flexible ligand to a nonflexible macromol. target of known structure. The technique combines simulated annealing for conformation searching with a rapid grid-based method of energy evaluation based on the AMBER force field. AutoDock has been optimized in performance without sacrificing accuracy; it incorporates many enhancements and addns., including an intuitive interface. We have developed a set of tools for launching and analyzing many independent docking jobs in parallel on a heterogeneous network of UNIX-based workstations. This paper describes the current release, and the results of a suite of diverse test systems. We also present the results of a systematic investigation in to the effects of varying simulated-annealing parameters on mol. docking. We show that even for ligands with a large no. of degrees of freedom, root-mean-square deviations of less than 1 Å from the crystallog. conformation are obtained for the lowest-energy dockings, although fewer dockings find the crystallog. conformation when there are more degrees of freedom.
  32. 32
    Trott, O.; Olson, A. J. Software News and Update AutoDock Vina: Improving the Speed and Accuracy of Docking with a New Scoring Function, Efficient Optimization and Multithreading J. Comput. Chem. 2010, 31, 455– 461
    Google Scholar
    There is no corresponding record for this reference.
  33. 33
    Shoichet, B. K.; Bodian, D. L.; Kuntz, I. D. Molecular Docking Using Shape Descriptors J. Comput. Chem. 1992, 13, 380– 397
    Google Scholar
    33
    Molecular docking using shape descriptors
    Shoichet, Brian K.; Bodian, Dale L.; Kuntz, Irwin D.
    Journal of Computational Chemistry (1992), 13 (3), 380-97CODEN: JCCHDD; ISSN:0192-8651.
    Mol. docking explores the binding modes of two interacting mols. The technique is increasingly popular for studying protein-ligand interactions and for drug design. A fundamental problem with mol. docking is that orientation space is very large and grows combinatorially with the no. of degrees of freedom of the interacting mols. Here, algorithms are described and evaluated that improve the efficiency and accuracy of a shape-based docking method. Mol. organization and sampling techniques are used to remove the exponential time dependence on mol. size in docking calcns. The new techniques allow one to study systems that were prohibitively large for the original method. The new algorithms are tested in 10 different protein-ligand systems, including systems, including 7 systems where the ligand is itself a protein. In all cases, the new algorithms successfully reproduce the exptl. detd. configurations of the ligand in the protein.
  34. 34
    Gohlke, H.; Hendlich, M.; Klebe, G. Knowledge-based scoring function to predict protein-ligand interactions J. Mol. Biol. 2000, 295, 337– 356
    Google Scholar
    34
    Knowledge-based Scoring Function to Predict Protein-Ligand Interactions
    Gohlke, Holger; Hendlich, Manfred; Klebe, Gerhard
    Journal of Molecular Biology (2000), 295 (2), 337-356CODEN: JMOBAK; ISSN:0022-2836. (Academic Press)
    The development and validation of a new knowledge-based scoring function (DrugScore) to describe the binding geometry of ligands in proteins is presented. It discriminates efficiently between well-docked ligand binding modes (root-mean-square deviation <2.0 Å with respect to a crystallog. detd. ref. complex) and those largely deviating from the native structure, e.g. generated by computer docking programs. Structural information is extd. from crystallog. detd. protein-ligand complexes using ReLiBase and converted into distance-dependent pair-preferences and solvent-accessible surface (SAS) dependent singlet preferences for protein and ligand atoms. Definition of an appropriate ref. state and accounting for inaccuracies inherently present in exptl. data is required to achieve good predictive power. The sum of the pair preferences and the singlet preferences is calcd. based on the 3D structure of protein-ligand binding modes generated by docking tools. For two test sets of 91 and 68 protein-ligand complexes, taken from the Protein Data Bank (PDB), the calcd. score recognizes poses generated by FlexX deviating <2 Å from the crystal structure on rank 1 in three quarters of all possible cases. Compared to FlexX, this is a substantial improvement. For ligand geometries generated by DOCK, DrugScore is superior to the "chem. scoring" implemented into this tool, while comparable results are obtained using the "energy scoring" in DOCK. None of the presently known scoring functions achieves comparable power to ext. binding modes in agreement with expt. It is fast to compute, regards implicitly solvation and entropy contributions and produces correctly the geometry of directional interactions. Small deviations in the 3D structure are tolerated and, since only contacts to non-hydrogen atoms are regarded, it is independent from assumptions of protonation states. (c) 2000 Academic Press.
  35. 35
    Zsoldos, Z.; Reid, D.; Simon, A.; Sadjad, S. B.; Johnson, A. P. eHiTS: a new fast, exhaustive flexible ligand docking system J. Mol. Graph. Model. 2007, 26, 198– 212
    Google Scholar
    35
    eHiTS: A new fast, exhaustive flexible ligand docking system
    Zsoldos, Zsolt; Reid, Darryl; Simon, Aniko; Sadjad, Sayyed Bashir; Johnson, A. Peter
    Journal of Molecular Graphics & Modelling (2007), 26 (1), 198-212CODEN: JMGMFI; ISSN:1093-3263. (Elsevier Inc.)
    The flexible ligand docking problem is divided into two subproblems: pose/conformation search and scoring function. For successful virtual screening the search algorithm must be fast and able to find the optimal binding pose and conformation of the ligand. Statistical anal. of exptl. data of bound ligand conformations is presented with conclusions about the sampling requirements for docking algorithms. EHiTS is an exhaustive flexible-docking method that systematically covers the part of the conformational and positional search space that avoids severe steric clashes, producing highly accurate docking poses at a speed practical for virtual high-throughput screening. The customizable scoring function of eHiTS combines novel terms (based on local surface point contact evaluation) with traditional empirical and statistical approaches. Validation results of eHiTS are presented and compared to three other docking software on a set of 91 PDB structures that are common to the validation sets published for the other programs.
  36. 36
    FRED; version 2.2.5; OpenEye Scientific Software, Inc.: Santa FRED, NM 87508, 2009.
    Google Scholar
    There is no corresponding record for this reference.
  37. 37
    Friesner, R. A.; Banks, J. L.; Murphy, R. B.; Halgren, T. A.; Klicic, J. J.; Mainz, D. T.; Repasky, M. P.; Knoll, E. H.; Shelley, M.; Perry, J. K.; Shaw, D. E.; Francis, P.; Shenkin, P. S. Glide: a new approach for rapid, accurate docking and scoring. 1. Method and assessment of docking accuracy J. Med. Chem. 2004, 47, 1739– 1749
    Google Scholar
    37
    Glide: A new approach for rapid, accurate docking and scoring. 1. method and assessment of docking accuracy
    Friesner, Richard A.; Banks, Jay L.; Murphy, Robert B.; Halgren, Thomas A.; Klicic, Jasna J.; Mainz, Daniel T.; Repasky, Matthew P.; Knoll, Eric H.; Shelley, Mee; Perry, Jason K.; Shaw, David E.; Francis, Perry; Shenkin, Peter S.
    Journal of Medicinal Chemistry (2004), 47 (7), 1739-1749CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
    A review. Unlike other methods for docking ligands to the rigid 3D structure of a known protein receptor, Glide approximates a complete systematic search of the conformational, orientational, and positional space of the docked ligand. In this search, an initial rough positioning and scoring phase that dramatically narrows the search space is followed by torsionally flexible energy optimization on an OPLS-AA nonbonded potential grid for a few hundred surviving candidate poses. The very best candidates are further refined via a Monte Carlo sampling of pose conformation; in some cases, this is crucial to obtaining an accurate docked pose. Selection of the best docked pose uses a model energy function that combines empirical and force-field-based terms. Docking accuracy is assessed by redocking ligands from 282 cocrystd. PDB complexes starting from conformationally optimized ligand geometries that bear no memory of the correctly docked pose. Errors in geometry for the top-ranked pose are less than 1 Å in nearly half of the cases and are greater than 2 Å in only about one-third of them. Comparisons to published data on rms deviations show that Glide is nearly twice as accurate as GOLD and more than twice as accurate as FlexX for ligands having up to 20 rotatable bonds. Glide is also found to be more accurate than the recently described Surflex method.
  38. 38
    Verdonk, M. L.; Cole, J. C.; Hartshorn, M. J.; Murray, C. W.; Taylor, R. D. Improved protein-ligand docking using GOLD Proteins: Struct., Funct., Bioinf. 2003, 52, 609– 623
    Google Scholar
    There is no corresponding record for this reference.
  39. 39
    Kramer, C.; Gedeck, P. Global free energy scoring functions based on distance-dependent atom-type pair descriptors J.Chem. Inf. Model. 2011, 51, 707– 720
    Google Scholar
    39
    Global free energy scoring functions based on distance-dependent atom-type pair descriptors
    Kramer, Christian; Gedeck, Peter
    Journal of Chemical Information and Modeling (2011), 51 (3), 707-720CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
    Scoring functions for protein-ligand docking have received much attention in the past two decades. In many cases, remarkable success has been demonstrated in predicting the correct geometry of interaction. On independent test sets, however, the predicted binding energies or scores correlate only slightly with the obsd. free energies of binding. In this study, we analyze how well free energies of binding can be predicted on the basis of crystal structures using traditional QSAR techniques in a proteochemometric approach. We introduce a new set of protein-ligand interaction descriptors on the basis of distance-binned Crippen-like atom type pairs. A subset of the publicly available PDBbind09-CN refined set (MW < 900 g/mol, #P < 2, ndon + nacc < 20; N = 1387) is being used as data set. It is demonstrated how simple, yet surprisingly good, scoring functions can be generated for the whole diverse database (R2out-of-bag = 0.48, Rp = 0.69, RMSE = 1.44, MUE = 1.14) and individual protein family subsets. This performance is significantly better than the performance of almost all other scoring functions published that have been validated on a test set as large and diverse as the PDBbind refined set. We also find that on some protein families surprisingly good scoring functions can be obtained using simple ligand-only descriptors like logS, logP, and mol. wt. The ligand-descriptor based scoring function equals or even outperforms commonly used scoring functions, highlighting the need for better scoring functions. We demonstrate how the obsd. performance depends on the validation strategy, and we outline a general validation protocol for future free energy scoring functions.
  40. 40
    Huang, S.-Y.; Zou, X. An iterative knowledge-based scoring function to predict protein-ligand interactions: I. Derivation of interaction potentials J. Comput. Chem. 2006, 27, 1866– 1875
    Google Scholar
    40
    An iterative knowledge-based scoring function to predict protein-ligand interactions: I. Derivation of interaction potentials
    Huang, Sheng-You; Zou, Xiaoqin
    Journal of Computational Chemistry (2006), 27 (15), 1866-1875CODEN: JCCHDD; ISSN:0192-8651. (John Wiley & Sons, Inc.)
    Using a novel iterative method, the authors have developed a knowledge-based scoring function (ITScore) to predict protein-ligand interactions. The pair potentials for ITScore were derived from a training set of 786 protein-ligand complex structures in the Protein Data Bank. Twenty-six atom types were used based on the atom type category of the SYBYL software. The iterative method circumvents the long-standing ref. state problem in the derivation of knowledge-based scoring functions. The basic idea is to improve pair potentials by iteration until they correctly discriminate exptl. detd. binding modes from decoy ligand poses for the ligand-protein complexes in the training set. The iterative method is efficient and normally converges within 20 iterative steps. The scoring function based on the derived potentials was tested on a diverse set of 140 protein-ligand complexes for affinity prediction, yielding a high correlation coeff. of 0.74. Because ITScore uses SYBYL-defined atom types, this scoring function is easy to use for mol. files prepd. by SYBYL or converted by software such as BABEL.
  41. 41
    Stroganov, O. V.; Novikov, F. N.; Stroylov, V. S.; Kulkov, V.; Chilov, G. G. Lead finder: an approach to improve accuracy of protein-ligand docking, binding energy estimation, and virtual screening J. Chem. Inf. Model. 2008, 48, 2371– 2385
    Google Scholar
    41
    Lead Finder: An Approach To Improve Accuracy of Protein-Ligand Docking, Binding Energy Estimation, and Virtual Screening
    Stroganov, Oleg V.; Novikov, Fedor N.; Stroylov, Viktor S.; Kulkov, Val; Chilov, Ghermes G.
    Journal of Chemical Information and Modeling (2008), 48 (12), 2371-2385CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
    An innovative mol. docking algorithm and three specialized high accuracy scoring functions are introduced in the Lead Finder docking software. Lead Finder's algorithm for ligand docking combines the classical genetic algorithm with various local optimization procedures and resourceful exploitation of the knowledge generated during docking process. Lead Finder's scoring functions are based on a mol. mechanics functional which explicitly accounts for different types of energy contributions scaled with empiric coeffs. to produce three scoring functions tailored for (a) accurate binding energy predictions; (b) correct energy-ranking of docked ligand poses; and (c) correct rank-ordering of active and inactive compds. in virtual screening expts. The predicted values of the free energy of protein-ligand binding were benchmarked against a set of exptl. measured binding energies for 330 diverse protein-ligand complexes yielding rmsd of 1.50 kcal/mol. The accuracy of ligand docking was assessed on a set of 407 structures, which included almost all published test sets of the following programs: FlexX, Glide SP, Glide XP, Gold, LigandFit, MolDock, and Surflex. Rmsd of 2 Å or less was obsd. for 80-96% of the structures in the test sets (80.0% on the Glide XP and FlexX test sets, 96.0% on the Surflex and MolDock test sets). The ability of Lead Finder to distinguish between active and inactive compds. during virtual screening expts. was benchmarked against 34 therapeutically relevant protein targets. Impressive enrichment factors were obtained for almost all of the targets with the av. area under receiver operator curve being equal to 0.92.
  42. 42
    Build_model; version 2.0.1 build 07.30; MolTech Ltd.: 2008–2011.
    Google Scholar
    There is no corresponding record for this reference.
  43. 43
    Yin, S.; Biedermannova, L.; Vondrasek, J.; Dokholyan, N. V. MedusaScore: an accurate force field-based scoring function for virtual drug screening J. Chem. Inf. Model. 2008, 48, 1656– 1662
    Google Scholar
    43
    MedusaScore: An Accurate Force Field-Based Scoring Function for Virtual Drug Screening
    Yin, Shuangye; Biedermannova, Lada; Vondrasek, Jiri; Dokholyan, Nikolay V.
    Journal of Chemical Information and Modeling (2008), 48 (8), 1656-1662CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
    Virtual screening is becoming an important tool for drug discovery. However, the application of virtual screening has been limited by the lack of accurate scoring functions. Here, we present a novel scoring function, MedusaScore, for evaluating protein-ligand binding. MedusaScore is based on models of phys. interactions that include van der Waals, solvation, and hydrogen bonding energies. To ensure the best transferability of the scoring function, we do not use any protein-ligand exptl. data for parameter training. We then test the MedusaScore for docking decoy recognition and binding affinity prediction and find superior performance compared to other widely used scoring functions. Statistical anal. indicates that one source of inaccuracy of MedusaScore may arise from the unaccounted entropic loss upon ligand binding, which suggests avenues of approach for further MedusaScore improvement.
  44. 44
    Molecular Operating Environment (MOE); version 2010.10; Chemical Computing Group: Montreal, C.N., 2010.
    Google Scholar
    There is no corresponding record for this reference.
  45. 45
    Goto, J.; Kataoka, R.; Muta, H.; Hirayama, N. ASEDock-docking based on alpha spheres and excluded volumes J. Chem. Inf. Model. 2008, 48, 583– 590
    Google Scholar
    45
    ASEDock - docking based on alpha spheres and excluded volumes
    Goto, Junichi; Kataoka, Ryoichi; Muta, Hajime; Hirayama, Noriaki
    Journal of Chemical Information and Modeling (2008), 48 (3), 583-590CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
    ASEDock is a novel docking program based on a shape similarity assessment between a concave portion (i.e., concavity) on a protein and the ligand. We have introduced two novel concepts into ASEDock. One is an ASE model, which is defined by the combination of alpha spheres generated at a concavity in a protein and the excluded vols. around the concavity. The other is an ASE score, which evaluates the shape similarity between the ligand and the ASE model. The ASE score selects and refines the initial pose by maximizing the overlap between the alpha spheres and the ligand, and minimizing the overlap between the excluded vol. and the ligand. Because the ASE score makes good use of the Gaussian-type function for evaluating and optimizing the overlap between the ligand and the site model, it can pose a ligand onto the docking site relatively faster and more effectively than using potential energy functions. The posing stage through the use of the ASE score is followed by full atomistic energy minimization. Because the posing algorithm of ASEDock is free from any bias except for shape, it is a very robust docking method. A validation study using 59 high-quality X-ray structures of the complexes between drug-like mols. and the target proteins has demonstrated that ASEDock can faithfully reproduce exptl. detd. docking modes of various druglike mols. in their target proteins. Almost 80% of the structures were reconstructed within the estd. exptl. error. The success rate of ∼98% was attained based on the docking criterion of the root-mean-square deviation (RMSD) of non-hydrogen atoms (≤2.0 Å). The markedly high success of ASEDock in redocking expts. clearly indicates that the most important factor governing the docking process is shape complementarity.
  46. 46
    Yang, C.-Y.; Wang, R.; Wang, S. M-score: a knowledge-based potential scoring function accounting for protein atom mobility J. Med. Chem. 2006, 49, 5903– 5911
    Google Scholar
    46
    M-Score: A Knowledge-Based Potential Scoring Function Accounting for Protein Atom Mobility
    Yang, Chao-Yie; Wang, Renxiao; Wang, Shaomeng
    Journal of Medicinal Chemistry (2006), 49 (20), 5903-5911CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
    A knowledge-based potential scoring function, named M-Score, has been developed based upon 2331 high-resoln. crystal structures of protein-ligand complexes. M-Score considers the mobility of protein atoms, describing the location of each protein atom by a Gaussian distribution instead of a fixed position based upon the isotropic B-factors. This leads to an increase in the no. of atom-pairs in the construction of knowledge-based potentials and a smoothing effect on the pairwise distribution functions. M-Score was validated using 896 complexes which were not included in the 2331 data set and whose exptl. detd. binding affinities were available. The overall linear correlation coeff. (r) between the calcd. scores and exptl. detd. binding affinities (pKi or pKd) for these 896 complexes is -0.49. Evaluation of M-Score against 17 protein families showed that we obtained good to excellent correlations for six protein families, modest correlations for four protein families, and poor correlations for the remaining seven protein families.
  47. 47
    Rahaman, O.; Estrada, T.; Doran, D.; Taufer, M.; Brooks, C.; Armen, R. Evaluation of Several Two-step Scoring Functions Based on Linear Interaction Energy, Effective Ligand Size, and Empirical Pair Potentials for Prediction of Protein-Ligand Binding Geometry and Free Energy J. Chem. Inf. Model. 2011,  DOI: 10.1021/ci1003009
    Google Scholar
    There is no corresponding record for this reference.
  48. 48
    Naim, M.; Bhat, S.; Rankin, K. N.; Dennis, S.; Chowdhury, S. F.; Siddiqi, I.; Drabik, P.; Sulea, T.; Bayly, C. I.; Jakalian, A.; Purisima, E. O. Solvated interaction energy (SIE) for scoring protein-ligand binding affinities. 1. Exploring the parameter space J. Chem. Inf. Model. 2007, 47, 122– 133
    Google Scholar
    48
    Solvated interaction energy (SIE) for scoring protein-ligand binding affinities. 1. Exploring the parameter space
    Naim Marwen; Bhat Sathesh; Rankin Kathryn N; Dennis Sheldon; Chowdhury Shafinaz F; Siddiqi Imran; Drabik Piotr; Sulea Traian; Bayly Christopher I; Jakalian Araz; Purisima Enrico O
    Journal of chemical information and modeling (2007), 47 (1), 122-33 ISSN:1549-9596.
    We present a binding free energy function that consists of force field terms supplemented by solvation terms. We used this function to calibrate the solvation model along with the binding interaction terms in a self-consistent manner. The motivation for this approach was that the solute dielectric-constant dependence of calculated hydration gas-to-water transfer free energies is markedly different from that of binding free energies (J. Comput. Chem. 2003, 24, 954). Hence, we sought to calibrate directly the solvation terms in the context of a binding calculation. The five parameters of the model were systematically scanned to best reproduce the absolute binding free energies for a set of 99 protein-ligand complexes. We obtained a mean unsigned error of 1.29 kcal/mol for the predicted absolute binding affinity in a parameter space that was fairly shallow near the optimum. The lowest errors were obtained with solute dielectric values of Din = 20 or higher and scaling of the intermolecular van der Waals interaction energy by factors ranging from 0.03 to 0.15. The high apparent Din and strong van der Waals scaling may reflect the anticorrelation of the change in solvated potential energy and configurational entropy, that is, enthalpy-entropy compensation in ligand binding (Biophys. J. 2004, 87, 3035-3049). Five variations of preparing the protein-ligand data set were explored in order to examine the effect of energy refinement and the presence of bound water on the calculated results. We find that retaining water in the final protein structure used for calculating the binding free energy is not necessary to obtain good results; that is the continuum solvation model is sufficient. Virtual screening enrichment studies on estrogen receptor and thymidine kinase showed a good ability of the binding free energy function to recover true hits in a collection of decoys.
  49. 49
    Wang, R.; Lai, L.; Wang, S. Further development and validation of empirical scoring functions for structure-based binding affinity prediction J. Comput. Aid. Mol. Des. 2002, 16, 11– 26
    Google Scholar
    49
    Further development and validation of empirical scoring functions for structure-based binding affinity prediction
    Wang, Renxiao; Lai, Luhua; Wang, Shaomeng
    Journal of Computer-Aided Molecular Design (2002), 16 (1), 11-26CODEN: JCADEQ; ISSN:0920-654X. (Kluwer Academic Publishers)
    New empirical scoring functions have been developed to est. the binding affinity of a given protein-ligand complex with known three-dimensional structure. These scoring functions include terms accounting for van der Waals interaction, hydrogen bonding, deformation penalty, and hydrophobic effect. A special feature is that three different algorithms have been implemented to calc. the hydrophobic effect term, which results in three parallel scoring functions. All three scoring functions are calibrated through multivariate regression anal. of a set of 200 protein-ligand complexes and they reproduce the binding free energies of the entire training set with std. deviations of 2.2 kcal/mol, 2.1 kcal/mol, and 2.0 kcal/mol, resp. These three scoring functions are further combined into a consensus scoring function, X-CSCORE. When tested on an independent set of 30 protein-ligand complexes, X-CSCORE is able to predict their binding free energies with a std. deviation of 2.2 kcal/mol. The potential application of X-CSCORE to mol. docking is also investigated. Our results show that this consensus scoring function improves the docking accuracy considerably when compared to the conventional force field computation used for mol. docking.
  50. 50
    R: A Language and Environment for Statistical Computing; Team, R.; D. C.; version 2.9.2; R Project for Statistical Computing: Vienna, Austria, 2009.
    Google Scholar
    There is no corresponding record for this reference.
  51. 51
    Bonett, D. G.; Wright, T. A. Sample Size Requirements for Estimating Pearson, Kendall and Spearman Correlatons Psychometrika 2000, 65, 23– 28
    Google Scholar
    There is no corresponding record for this reference.
  52. 52
    JMP; version 9.0.0; SAS institute Inc.: Cary, N.C.: 2010.
    Google Scholar
    There is no corresponding record for this reference.
  53. 53
    Davis, I. W.; Leaver-Fay, A.; Chen, V. B.; Block, J. N.; Kapral, G. J.; Wang, X.; Murray, L. W.; Arendall, W. B., 3rd; Snoeyink, J.; Richardson, J. S.; Richardson, D. C. MolProbity: all-atom contacts and structure validation for proteins and nucleic acids Nucleic Acids Res. 2007, 35, W375– W383
    Google Scholar
    53
    MolProbity: all-atom contacts and structure validation for proteins and nucleic acids
    Davis Ian W; Leaver-Fay Andrew; Chen Vincent B; Block Jeremy N; Kapral Gary J; Wang Xueyi; Murray Laura W; Arendall W Bryan 3rd; Snoeyink Jack; Richardson Jane S; Richardson David C
    Nucleic acids research (2007), 35 (Web Server issue), W375-83 ISSN:.
    MolProbity is a general-purpose web server offering quality validation for 3D structures of proteins, nucleic acids and complexes. It provides detailed all-atom contact analysis of any steric problems within the molecules as well as updated dihedral-angle diagnostics, and it can calculate and display the H-bond and van der Waals contacts in the interfaces between components. An integral step in the process is the addition and full optimization of all hydrogen atoms, both polar and nonpolar. New analysis functions have been added for RNA, for interfaces, and for NMR ensembles. Additionally, both the web site and major component programs have been rewritten to improve speed, convenience, clarity and integration with other resources. MolProbity results are reported in multiple forms: as overall numeric scores, as lists or charts of local problems, as downloadable PDB and graphics files, and most notably as informative, manipulable 3D kinemage graphics shown online in the KiNG viewer. This service is available free to all users at http://molprobity.biochem.duke.edu.
  54. 54
    Vriend, G. WHAT IF: a molecular modeling and drug design program J. Mol. Graph. 1990, 8, 52– 56
    Google Scholar
    54
    WHAT IF: a molecular modeling and drug design program
    Vriend, G.
    Journal of Molecular Graphics (1990), 8 (1), 52-6, 29CODEN: JMGRDV; ISSN:0263-7855.
    A FORTRAN 77 computer program has been written to aid with macromol. modeling and drug design. Called WHAT IF, it provides an intelligent and flexible environment for displaying, manipulating, and analyzing small mols., proteins, nucleic acids, and their interactions. A relational protein structure database is incorporated to be queried. The program is suitable for most common crystallog. work. The menu-driven operation of WHAT IF, combined with the use of default values wherever user input is required, makes it very easy to use for a novice user while keeping full flexibility for more sophisticated studies. Although there are not too many unique features in WHAT IF, the fact that everything is integrated in one program makes if a unique tool for many purposes.
  55. 55
    Cruickshank, D. W. Remarks about protein structure precision Acta Crystallogr. D 1999, 55, 583– 601
    Google Scholar
    55
    Remarks about protein structure precision
    Cruickshank D W
    Acta crystallographica. Section D, Biological crystallography (1999), 55 (Pt 3), 583-601 ISSN:0907-4449.
    Full-matrix least squares is taken as the basis for an examination of protein structure precision. A two-atom protein model is used to compare the precisions of unrestrained and restrained refinements. In this model, restrained refinement determines a bond length which is the weighted mean of the unrestrained diffraction-only length and the geometric dictionary length. Data of 0.94 A resolution for the 237-residue protein concanavalin A are used in unrestrained and restrained full-matrix inversions to provide standard uncertainties sigma(r) for positions and sigma(l) for bond lengths. sigma(r) is as small as 0.01 A for atoms with low Debye B values but increases strongly with B. The results emphasize the distinction between unrestrained and restrained refinements and between sigma(r) and sigma(l). Other full-matrix inversions are reported. Such inversions require massive calculations. Several approximate methods are examined and compared critically. These include a Fourier map formula [Cruickshank (1949). Acta Cryst. 2, 65-82], Luzzati plots [Luzzati (1952). Acta Cryst. 5, 802-810] and a new diffraction-component precision index (DPI). The DPI estimate of sigma(r, Bavg) is given by a simple formula. It uses R or Rfree and is based on a very rough approximation to the least-squares method. Many examples show its usefulness as a precision comparator for high- and low-resolution structures. The effect of restraints as resolution varies is examined. More regular use of full-matrix inversion is urged to establish positional precision and hence the precision of non-dictionary distances in both high- and low-resolution structures. Failing this, parameter blocks for representative residues and their neighbours should be inverted to gain a general idea of sigma(r) as a function of B. The whole discussion is subject to some caveats about the effects of disordered regions in the crystal.
  56. 56
    Smith, R. D.; Hu, L.; Falkner, J. A.; Benson, M. L.; Nerothin, J. P.; Carlson, H. A. Exploring protein-ligand recognition with Binding MOAD J. Mol. Graphics Modell. 2006, 24, 414– 425
    Google Scholar
    56
    Exploring protein-ligand recognition with Binding MOAD
    Smith, Richard D.; Hu, Liegi; Falkner, Jayson A.; Benson, Mark L.; Nerothin, Jason P.; Carlson, Heather A.
    Journal of Molecular Graphics & Modelling (2006), 24 (6), 414-425CODEN: JMGMFI; ISSN:1093-3263. (Elsevier Inc.)
    We have recently announced the largest database of protein-ligand complexes, Binding MOAD (Mother of All Databases). After the August 2004 update, Binding MOAD contains 6816 complexes. There are 2220 protein families and 3316 unique ligands. After searching 6000+ crystallog. papers, we have obtained binding data for 1793 (27%) of the complexes. We have also created a non-redundant set of complexes with only one complex from each protein family; in that set, 630 (28%) of the unique complexes have binding data. Here, we present information about the data provided at the Binding MOAD website. We also present the results of mining Binding MOAD to map the degree of solvent exposure for binding sites. We have detd. that most cavities and ligands (70-85%) are well buried in the complexes. This fits with the common paradigm that a large degree of contact between the ligand and protein is significant in mol. recognition. GoCAV and the GoCAVviewer are the tools we created for this study. To share our data and make our online dataset more useful to other research groups, we have integrated the viewer into the Binding MOAD website (www.BindingMOAD.org).
  57. 57
    Wallin, R.; Hutson, S. M. Warfarin and the vitamin K-dependent gamma-carboxylation system Trends Mol. Med. 2004, 10, 299– 302
    Google Scholar
    There is no corresponding record for this reference.
  58. 58
    Liebeschuetz, J. W.; Jones, S. D.; Morgan, P. J.; Murray, C. W.; Rimmer, A. D.; Roscoe, J. M.; Waszkowycz, B.; Welsh, P. M.; Wylie, W. A.; Young, S. C.; Martin, H.; Mahler, J.; Brady, L.; Wilkinson, K. PRO_SELECT: combining structure-based drug design and array-based chemistry for rapid lead discovery. 2. The development of a series of highly potent and selective factor Xa inhibitors J. Med. Chem. 2002, 45, 1221– 1232
    Google Scholar
    There is no corresponding record for this reference.
  59. 59
    Huang, N.; Shoichet, B. K.; Irwin, J. J. Benchmarking sets for molecular docking J. Med. Chem. 2006, 49, 6789– 6801
    Google Scholar
    59
    Benchmarking Sets for Molecular Docking
    Huang, Niu; Shoichet, Brian K.; Irwin, John J.
    Journal of Medicinal Chemistry (2006), 49 (23), 6789-6801CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
    Ligand enrichment among top-ranking hits is a key metric of mol. docking. To avoid bias, decoys should resemble ligands phys., so that enrichment is not simply a sepn. of gross features, yet be chem. distinct from them, so that they are unlikely to be binders. We have assembled a directory of useful decoys (DUD), with 2950 ligands for 40 different targets. Every ligand has 36 decoy mols. that are phys. similar but topol. distinct, leading to a database of 98 266 compds. For most targets, enrichment was at least half a log better with uncorrected databases such as the MDDR than with DUD, evidence of bias in the former. These calcns. also allowed 40×40 cross-docking, where the enrichments of each ligand set could be compared for all 40 targets, enabling a specificity metric for the docking screens. DUD is freely available online as a benchmarking set for docking at http://blaster.docking.org/dud/.
  60. 60
    Murcia, M.; Ortiz, A. R. Virtual screening with flexible docking and COMBINE-based models. Application to a series of factor Xa inhibitors J. Med. Chem. 2004, 47, 805– 820
    Google Scholar
    There is no corresponding record for this reference.
  61. 61
    Nazare, M.; Will, D. W.; Matter, H.; Schreuder, H.; Ritter, K.; Urmann, M.; Essrich, M.; Bauer, A.; Wagner, M.; Czech, J.; Lorenz, M.; Laux, V.; Wehner, V. Probing the subpockets of factor Xa reveals two binding modes for inhibitors based on a 2-carboxyindole scaffold: a study combining structure-activity relationship and X-ray crystallography J. Med. Chem. 2005, 48, 4511– 4525
    Google Scholar
    61
    Probing the Subpockets of Factor Xa Reveals Two Binding Modes for Inhibitors Based on a 2-Carboxyindole Scaffold: A Study Combining Structure-Activity Relationship and X-ray Crystallography
    Nazare, Marc; Will, David W.; Matter, Hans; Schreuder, Herman; Ritter, Kurt; Urmann, Matthias; Essrich, Melanie; Bauer, Armin; Wagner, Michael; Czech, Joerg; Lorenz, Martin; Laux, Volker; Wehner, Volkmar
    Journal of Medicinal Chemistry (2005), 48 (14), 4511-4525CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
    Structure-activity relationships within a series of highly potent 2-carboxyindole-based factor Xa inhibitors incorporating a neutral P1 ligand are described with particular emphasis on the structural requirements for addressing subpockets of the factor Xa enzyme. Interactions with the subpockets were probed by systematic substitution of the 2-carboxyindole scaffold, in combination with privileged P1 and P4 substituents. Combining the most favorable substituents at the indole nucleus led to the discovery of a remarkably potent factor Xa inhibitor displaying a Ki value of 0.07 nM. X-ray crystallog. of inhibitors bound to factor Xa revealed substituent-dependent switching of the inhibitor binding mode and provided a rationale for the SAR obtained. These results underscore the key role played by the P1 ligand not only in detg. the binding affinity of the inhibitor by direct interaction but also in modifying the binding mode of the whole scaffold, resulting in a nonlinear SAR.
  62. 62
    Pinto, D. J.; Orwat, M. J.; Koch, S.; Rossi, K. A.; Alexander, R. S.; Smallwood, A.; Wong, P. C.; Rendina, A. R.; Luettgen, J. M.; Knabb, R. M.; He, K.; Xin, B.; Wexler, R. R.; Lam, P. Y. Discovery of 1-(4-methoxyphenyl)-7-oxo-6-(4-(2-oxopiperidin-1-yl)phenyl)-4,5,6,7-tetrah ydro-1H-pyrazolo[3,4-c]pyridine-3-carboxamide (apixaban, BMS-562247), a highly potent, selective, efficacious, and orally bioavailable inhibitor of blood coagulation factor Xa J. Med. Chem. 2007, 50, 5339– 5356
    Google Scholar
    There is no corresponding record for this reference.
  63. 63
    Qiao, J. X.; Chang, C. H.; Cheney, D. L.; Morin, P. E.; Wang, G. Z.; King, S. R.; Wang, T. C.; Rendina, A. R.; Luettgen, J. M.; Knabb, R. M.; Wexler, R. R.; Lam, P. Y. SAR and X-ray structures of enantiopure 1,2-cis-(1R,2S)-cyclopentyldiamine and cyclohexyldiamine derivatives as inhibitors of coagulation Factor Xa Bioorg. Med. Chem. Lett. 2007, 17, 4419– 4427
    Google Scholar
    There is no corresponding record for this reference.
  64. 64
    Qiao, J. X.; Cheng, X.; Smallheer, J. M.; Galemmo, R. A.; Drummond, S.; Pinto, D. J.; Cheney, D. L.; He, K.; Wong, P. C.; Luettgen, J. M.; Knabb, R. M.; Wexler, R. R.; Lam, P. Y. Pyrazole-based factor Xa inhibitors containing N-arylpiperidinyl P4 residues Bioorg. Med. Chem. Lett. 2007, 17, 1432– 1437
    Google Scholar
    There is no corresponding record for this reference.
  65. 65
    Senger, S.; Convery, M. A.; Chan, C.; Watson, N. S. Arylsulfonamides: a study of the relationship between activity and conformational preferences for a series of factor Xa inhibitors Bioorg. Med. Chem. Lett. 2006, 16, 5731– 5735
    Google Scholar
    There is no corresponding record for this reference.
  66. 66
    Watson, N. S.; Brown, D.; Campbell, M.; Chan, C.; Chaudry, L.; Convery, M. A.; Fenwick, R.; Hamblin, J. N.; Haslam, C.; Kelly, H. A.; King, N. P.; Kurtis, C. L.; Leach, A. R.; Manchee, G. R.; Mason, A. M.; Mitchell, C.; Patel, C.; Patel, V. K.; Senger, S.; Shah, G. P.; Weston, H. E.; Whitworth, C.; Young, R. J. Design and synthesis of orally active pyrrolidin-2-one-based factor Xa inhibitors Bioorg. Med. Chem. Lett. 2006, 16, 3784– 3788
    Google Scholar
    There is no corresponding record for this reference.
  67. 67
    Ye, B.; Arnaiz, D. O.; Chou, Y. L.; Griedel, B. D.; Karanjawala, R.; Lee, W.; Morrissey, M. M.; Sacchi, K. L.; Sakata, S. T.; Shaw, K. J.; Wu, S. C.; Zhao, Z.; Adler, M.; Cheeseman, S.; Dole, W. P.; Ewing, J.; Fitch, R.; Lentz, D.; Liang, A.; Light, D.; Morser, J.; Post, J.; Rumennik, G.; Subramanyam, B.; Sullivan, M. E.; Vergona, R.; Walters, J.; Wang, Y. X.; White, K. A.; Whitlow, M.; Kochanny, M. J. Thiophene-anthranilamides as highly potent and orally available factor Xa inhibitors J. Med. Chem. 2007, 50, 2967– 2980
    Google Scholar
    67
    Thiophene-Anthranilamides as Highly Potent and Orally Available Factor Xa Inhibitors
    Ye, Bin; Arnaiz, Damian O.; Chou, Yuo-Ling; Griedel, Brian D.; Karanjawala, Rushad; Lee, Wheeseong; Morrissey, Michael M.; Sacchi, Karna L.; Sakata, Steven T.; Shaw, Kenneth J.; Wu, Shung C.; Zhao, Zuchun; Adler, Marc; Cheeseman, Sarah; Dole, William P.; Ewing, Janice; Fitch, Richard; Lentz, Dao; Liang, Amy; Light, David; Morser, John; Post, Joseph; Rumennik, Galina; Subramanyam, Babu; Sullivan, Mark E.; Vergona, Ron; Walters, Janette; Wang, Yi-Xin; White, Kathy A.; Whitlow, Marc; Kochanny, Monica J.
    Journal of Medicinal Chemistry (2007), 50 (13), 2967-2980CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
    There remains a high unmet medical need for a safe oral therapy for thrombotic disorders. The serine protease factor Xa (fXa), with its central role in the coagulation cascade, is among the more promising targets for anticoagulant therapy and has been the subject of intensive drug discovery efforts. Investigation of a hit from high-throughput screening identified a series of thiophene-substituted anthranilamides as potent nonamidine fXa inhibitors. Lead optimization by incorporation of hydrophilic groups led to the discovery of compds. with picomolar inhibitory potency and micromolar in vitro anticoagulant activity. Based on their high potency, selectivity, oral pharmacokinetics, and efficacy in a rat venous stasis model of thrombosis, compds. ZK 814048 (10b), ZK 810388 (13a), and ZK 813039 (17m) were advanced into development.
  68. 68
    Young, R. J.; Brown, D.; Burns-Kurtis, C. L.; Chan, C.; Convery, M. A.; Hubbard, J. A.; Kelly, H. A.; Pateman, A. J.; Patikis, A.; Senger, S.; Shah, G. P.; Toomey, J. R.; Watson, N. S.; Zhou, P. Selective and dual action orally active inhibitors of thrombin and factor Xa Bioorg. Med. Chem. Lett. 2007, 17, 2927– 2930
    Google Scholar
    68
    Selective and dual action orally active inhibitors of thrombin and factor Xa
    Young, Robert J.; Brown, David; Burns-Kurtis, Cynthia L.; Chan, Chuen; Convery, Maire A.; Hubbard, Julia A.; Kelly, Henry A.; Pateman, Anthony J.; Patikis, Angela; Senger, Stefan; Shah, Gita P.; Toomey, John R.; Watson, Nigel S.; Zhou, Ping
    Bioorganic & Medicinal Chemistry Letters (2007), 17 (10), 2927-2930CODEN: BMCLE8; ISSN:0960-894X. (Elsevier Ltd.)
    The synthetic entry to new classes of dual fXa/thrombin and selective thrombin inhibitors with significant oral bioavailability is described. This was achieved through minor modifications to the sulfonamide group in our potent and selective fXa inhibitor (E)-2-(5-chlorothien-2-yl)-N-{(3S)-1-[(1S)-1-methyl-2-(morpholin-4-yl)-2-oxoethyl]-2-oxopyrrolidin-3-yl}ethenesulfonamide and these obsd. activity changes have been rationalized using structural studies.
  69. 69
    Young, R. J.; Campbell, M.; Borthwick, A. D.; Brown, D.; Burns-Kurtis, C. L.; Chan, C.; Convery, M. A.; Crowe, M. C.; Dayal, S.; Diallo, H.; Kelly, H. A.; King, N. P.; Kleanthous, S.; Mason, A. M.; Mordaunt, J. E.; Patel, C.; Pateman, A. J.; Senger, S.; Shah, G. P.; Smith, P. W.; Watson, N. S.; Weston, H. E.; Zhou, P. Structure- and property-based design of factor Xa inhibitors: pyrrolidin-2-ones with acyclic alanyl amides as P4 motifs Bioorg. Med. Chem. Lett. 2006, 16, 5953– 5957
    Google Scholar
    69
    Structure- and property-based design of factor Xa inhibitors: Pyrrolidin-2-ones with acyclic alanyl amides as P4 motifs
    Young, Robert J.; Campbell, Matthew; Borthwick, Alan D.; Brown, David; Burns-Kurtis, Cynthia L.; Chan, Chuen; Convery, Maire A.; Crowe, Miriam C.; Dayal, Satish; Diallo, Hawa; Kelly, Henry A.; King, N. Paul; Kleanthous, Savvas; Mason, Andrew M.; Mordaunt, Jackie E.; Patel, Champa; Pateman, Anthony J.; Senger, Stefan; Shah, Gita P.; Smith, Paul W.; Watson, Nigel S.; Weston, Helen E.; Zhou, Ping
    Bioorganic & Medicinal Chemistry Letters (2006), 16 (23), 5953-5957CODEN: BMCLE8; ISSN:0960-894X. (Elsevier Ltd.)
    Nonracemic 1-alaninyl-3-(6-chloro-2-naphthylsulfonylamino)-2-pyrrolidinones such as I are prepd. as factor Xa inhibitors for use as anticoagulants; the hydrophobicities, anticoagulant activities in rats, and pharmacokinetics of some of the compds. are detd. For example, I inhibits factor Xa with a Ki value of 1 nM and a value of 2.5 μM in the prothrombin time assay. 1-Alaninyl-3-(6-chloro-2-naphthylsulfonylamino)-2-pyrrolidinones have poorer pharmacokinetic profiles than the corresponding morpholine-based analogs, with increased plasma clearance and decreased oral bioavailabilities. The structure of I bound to factor Xa is detd. by X-ray crystallog.
  70. 70
    Verdonk, M. L.; Berdini, V.; Hartshorn, M. J.; Mooij, W. T.; Murray, C. W.; Taylor, R. D.; Watson, P. Virtual screening using protein-ligand docking: avoiding artificial enrichment J. Chem. Inf. Comput. Sci. 2004, 44, 793– 806
    Google Scholar
    70
    Virtual screening using protein-ligand docking: Avoiding artificial enrichment
    Verdonk, Marcel L.; Berdini, Valerio; Hartshorn, Michael J.; Mooij, Wijnand T. M.; Murray, Christopher W.; Taylor, Richard D.; Watson, Paul
    Journal of Chemical Information and Computer Sciences (2004), 44 (3), 793-806CODEN: JCISD8; ISSN:0095-2338. (American Chemical Society)
    This study addresses a no. of topical issues around the use of protein-ligand docking in virtual screening. We show that, for the validation of such methods, it is key to use focused libraries (contg. compds. with one-dimensional properties, similar to the actives), rather than "random" or "drug-like" libraries to test the actives against. We also show that, to obtain good enrichments, the docking program needs to produce reliable binding modes. We demonstrate how pharmacophores can be used to guide the dockings and improve enrichments, and we compare the performance of three consensus-ranking protocols against ranking based on individual scoring functions. Finally, we show that protein-ligand docking can be an effective aid in the screening for weak, fragment-like binders, which has rapidly become a popular strategy for hit identification. All results presented are based on carefully constructed virtual screening expts. against four targets, using the protein-ligand docking program GOLD.
  71. 71
    Jacobsson, M.; Karlen, A. Ligand bias of scoring functions in structure-based virtual screening J. Chem. Inf. Model. 2006, 46, 1334– 1343
    Google Scholar
    71
    Ligand Bias of Scoring Functions in Structure-Based Virtual Screening
    Jacobsson, Micael; Karlen, Anders
    Journal of Chemical Information and Modeling (2006), 46 (3), 1334-1343CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
    A total of 945 known actives and roughly 10 000 decoy compds. were docked to eight different targets, and the resulting poses were scored using 10 different scoring functions. Three different score postprocessing methods were evaluated with respect to improvement of the enrichment in virtual screening. The three procedures were (i) multiple active site correction (MASC) as has been proposed by Vigers and Rizzi, (ii) a variation of MASC where corrections terms are predicted from simple mol. descriptors through PLS, PLS MASC, and (iii) size normalization. It was found that MASC did not generally improve the enrichment factors when compared to uncorrected scoring functions. For some combinations of scoring functions and targets, the enrichment was improved, for others not. However, by excluding the std. deviation from the MASC equation and transforming the scores for each target to a mean of 0 and a std. deviation of 1 (unit variance normalization), the performance was improved as compared to the original MASC method for most combinations of targets and scoring functions. Furthermore, when the mol. descriptors were fit to the mean scores over all targets and the resulting PLS models were used to predict mean scores, the enrichment as compared to the raw score was improved more often than by straightforward MASC. A high to intermediate linear correlation between the score and the no. of heavy atoms was found for all scoring functions except FlexX. There seems to be a correlation between the size dependence of a scoring function and the effectiveness of PLS MASC in increasing the enrichment for that scoring function. Finally, normalization by mol. wt. or heavy atom count was sometimes successful in increasing the enrichment. Dividing by the square or cubic root of the mol. wt. or heavy atom count instead was often more successful. These results taken together suggest that ligand bias in scoring functions is a source of false positives in structure-based virtual screening. The no. of false positives caused by ligand bias may be decreased using, for example, the PLS MASC procedure proposed in this study.
  72. 72
    Krovat, E. M.; Steindle, T.; Langer, T. Recent advances in docking and scoring Curr. Comput.-Aided Drug Des. 2005, 1, 93– 102
    Google Scholar
    72
    Recent advances in docking and scoring
    Krovat, E. M.; Steindl, T.; Langer, T.
    Current Computer-Aided Drug Design (2005), 1 (1), 93-102CODEN: CCDDAS; ISSN:1573-4099. (Bentham Science Publishers Ltd.)
    A review on recent advances and new aspects in the field of mol. docking and scoring, and covers multiple applications and case studies. Basic requirements and different algorithms for docking are briefly discussed. Moreover, parameters that influence docking results, combination of different docking algorithms and scoring functions, performance of scoring functions, docking using homol. models, and ligands and protein flexibility are examd. to give an overview of the state-of-the-art methods and a survey of innovative approaches in mol. docking and scoring. Regarding the enormous amt. of literature in this field, an overview is given of several important advances in docking and scoring techniques published within the last two years, i.e. publications ranging from 2002 to 2004.
  73. 73
    Carlson, H. A.; Smith, R. D.; Khazanov, N. A.; Kirchhoff, P. D.; Dunbar, J. B.; Benson, M. L. Differences between high- and low-affinity complexes of enzymes and nonenzymes J. Med. Chem. 2008, 51, 6432– 6441
    Google Scholar
    There is no corresponding record for this reference.
  74. 74
    Gohlke, H.; Klebe, G. Approaches to the description and prediction of the binding affinity of small-molecule ligands to macromolecular receptors Angew. Chem., Int. Ed. Engl. 2002, 41, 2644– 2676
    Google Scholar
    74
    Approaches to the description and prediction of the binding affinity of small-molecule ligands to macromolecular receptors
    Gohlke, Holger; Klebe, Gerhard
    Angewandte Chemie, International Edition (2002), 41 (15), 2644-2676CODEN: ACIEF5; ISSN:1433-7851. (Wiley-VCH Verlag GmbH)
    A review. The influence of a xenobiotic compd. on an organism is usually summarized by the expression biol. activity. If a controlled, therapeutically relevant, and regulatory action is obsd. the compd. has potential as a drug, otherwise its toxicity on the biol. system is of interest. However, what do we understand by the biol. activity. In principle, the overall effect on an organism has to be considered. However, because of the complexity of the interrelated processes involved, as a simplification primarily the "main action" on the organism is taken into consideration. On the mol. level, biol. activity corresponds to the binding of a (lowmol. wt.) compd. to a macromol. receptor, usually a protein. Enzymic reactions or signal-transduction cascades are thereby influenced with respect to their function for the organism. We regard this binding as a process under equil. conditions; thus, binding can be described as an assocn. or dissocn. process. Accordingly, biol. activity is expressed as the affinity of both partners for each other, as a thermodn. equil. quantity. How well do we understand these terms and how well are they theor. predictable today. The holy grail of rational drug design is the prediction of the biol. activity of a compd. The processes involving ligand binding are extremely complicated, both ligand and protein are flexible mols., and the energy inventory between the bound and unbound states must be considered in aq. soln. How sophisticated and reliable are our exptl. approaches to obtaining the necessary insight. The present review summarizes our current understanding of the binding affinity of a small-mol. ligand to a protein. Both theor. and empirical approaches for predicting binding affinity, starting from the three-dimensional structure of a protein-ligand complex, will be described and compared. Exptl. methods, primarily microcalorimetry, will be discussed. As a perspective, our own knowledge-based approach towards affinity prediction and exptl. data on factorizing binding contributions to protein-ligand binding will be presented.
  75. 75
    Davis, A. M.; Teague, S. J. Hydrogen Bonding, Hydrophobic Interactions, and Failure of the Rigid Receptor Hypothesis Angew. Chem., Int. Ed. Engl. 1999, 38, 736– 749
    Google Scholar
    There is no corresponding record for this reference.
  76. 76
    Wildman, S. A.; Crippen, G. M. Prediction of physicochemical parameters by atomic contributions J. Chem. Inf. Comput. Sci. 1999, 39, 868– 873
    Google Scholar
    76
    Prediction of Physicochemical Parameters by Atomic Contributions
    Wildman, Scott A.; Crippen, Gordon M.
    Journal of Chemical Information and Computer Sciences (1999), 39 (5), 868-873CODEN: JCISD8; ISSN:0095-2338. (American Chemical Society)
    We present a new atom type classification system for use in atom-based calcn. of partition coeff. (log P) and molar refractivity (MR) designed in part to address published concerns of previous at. methods. The 68 at. contributions to log P have been detd. by fitting an extensive training set of 9920 mols., with r2 = 0.918 and σ = 0.677. A sep. set of 3412 mols. was used for the detn. of contributions to MR with r2 = 0.997 and σ = 1.43. Both calcns. are shown to have high predictive ability.
  77. 77
    David, L.; Amara, P.; Field, M. J.; Major, F. Parametrization of a force field for metals complexed to biomacromolecules: applications to Fe(II), Cu(II) and Pb(II) J. Comput.- Aided Mol. Des. 2002, 16, 635– 651
    Google Scholar
    77
    Parametrization of a force field for metals complexed to biomacromolecules: applications to Fe(II), Cu(II) and Pb(II)
    David, Laurent; Amara, Patricia; Field, Martin J.; Major, Francois
    Journal of Computer-Aided Molecular Design (2002), 16 (8/9), 635-651CODEN: JCADEQ; ISSN:0920-654X. (Kluwer Academic Publishers)
    Although techniques for the simulation of biomols., such as proteins and RNAs, have greatly advanced in the last decade, modeling complexes of biomols. with metal ions remains problematic. Precise calcns. can be done with quantum mech. methods but these are prohibitive for systems the size of macromols. More qual. modeling can be done with mol. mech. potentials but the parametrization of force fields for metals is often difficult, particularly if the bonding between the metal and the groups in its coordination shell has significant covalent character. In this paper we present a method for deriving bond and bond-angle parameters for metal complexes from exptl. bond and bond-angle distributions obtained from the Cambridge Structural Database. In conjunction with this method, we also introduce a non-std. energy term of gaussian form that allows us to obtain a stable description of the coordination about a metal center during a simulation. The method was evaluated on Fe(II)-porphyrin complexes, on simple Cu(II) ion complexes and a no. of complexes of the Pb(II) ion.
  78. 78
    Irwin, J. J.; Raushel, F. M.; Shoichet, B. K. Virtual screening against metalloenzymes for inhibitors and substrates Biochemistry 2005, 44, 12316– 12328
    Google Scholar
    78
    Virtual Screening against Metalloenzymes for Inhibitors and Substrates
    Irwin, John J.; Raushel, Frank M.; Shoichet, Brian K.
    Biochemistry (2005), 44 (37), 12316-12328CODEN: BICHAW; ISSN:0006-2960. (American Chemical Society)
    Mol. docking uses the three-dimensional structure of a receptor to screen databases of small mols. for potential ligands, often based on energetic complementarity. For many docking scoring functions, which calc. nonbonded interactions, metalloenzymes are challenging because of the partial covalent nature of metal-ligand interactions. To investigate how well mol. docking can identify potential ligands of metalloenzymes using a "std." scoring function, we have docked the MDL Drug Data Report (MDDR), a functionally annotated database of 95 000 small mols., against the X-ray crystal structures of five metalloenzymes. These enzymes included three zinc proteases, the nickel analog of an iron enzyme, and a molybdenum metalloenzyme. The ability of the docking program to retrospectively enrich the annotated ligands as high-scoring hits for each enzyme and to calc. proper geometries was evaluated. In all five systems, the annotated ligands within the MDDR were enriched at least 20 times over random. To test the approach prospectively, a sixth target, the zinc β-lactamase from Bacteroides fragilis, was screened against the fragment-like subset of the ZINC database. We purchased and tested 15 compds. from among the top 50 top-ranked ligands from docking, and found 5 inhibitors with apparent Ki values less than 120 μM, the best of which was 2 μM. A more ambitious test still was predicting actual substrates for a seventh target, a Zn-dependent phosphotriesterase from Pseudomonas diminuta. Screening the Available Chems. Directory (ACD) identified 25 thiophosphate esters as potential substrates within the top 100 ranked compds. Eight of these, all previously uncharacterized for this enzyme, were acquired and tested, and all were confirmed exptl. as substrates. These results suggest that a simple, noncovalent scoring function may be used to identify inhibitors of at least some metalloenzymes.
  79. 79
    Li, X.; Hayik, S. A.; Merz, K. M., Jr. QM/MM X-ray refinement of zinc metalloenzymes J. Inorg. Biochem. 2010, 104, 512– 522
    Google Scholar
    There is no corresponding record for this reference.
  80. 80
    Lipinski, C. A.; Lombardo, F.; Dominy, B. W.; Feeney, P. J. Experimental and Compoutational Approaches to Estimate Solubility and Permeability in Drug Discover and Development Settings Adv. Drug Delivery Rev. 1997, 23, 3– 25
    Google Scholar
    There is no corresponding record for this reference.
  81. 81
    Oprea, T. I. Property distribution of drug-related chemical databases J. Comput.- Aided Mol. Des. 2000, 14, 251– 264
    Google Scholar
    81
    Property distribution of drug-related chemical databases
    Oprea, Tudor I.
    Journal of Computer-Aided Molecular Design (2000), 14 (3), 251-264CODEN: JCADEQ; ISSN:0920-654X. (Kluwer Academic Publishers)
    The process of compd. selection and prioritization is crucial for both combinatorial chem. (CBC) and high throughput screening (HTS). Compd. libraries have to be screened for unwanted chem. structures, as well as for unwanted chem. properties. Property extrema can be eliminated by using property filters, in accordance with their actual distribution. Property distribution was examd. in the following compd. databases: MACCS-II Drug Data Report (MDDR), Current Patents Fast-alert, Comprehensive Medicinal Chem., Physician Desk Ref., New Chem. Entities, and the Available Chem. Directory (ACD). The ACDF and MDDRF subsets were created by removing reactive functionalities from the ACD and MDDR databases, resp. The ACDF subset was further filtered by keeping only mols. with a "drug-like" score below 0.8. The following properties were examd.: mol. wt. (MW), the calcd. octanol/water partition coeff. (CLOGP), the no. of rotatable (RTB) and rigid bonds (RGB), the no. of rings (RNG), and the no. of hydrogen bond donors (HDO) and acceptors (HAC). Of these, MW and CLOGP follow a Gaussian distribution, whereas all other descriptors have an asym. (truncated Gaussian) distribution. Four out of five compds. in ACDF and MDDRF pass the "rule of 5" test, a probability scheme that ests. oral absorption proposed by Lipinski et al. Because property distributions of HDO, HAC, MW and CLOGP (used in the "rule of 5" test) do not differ significantly between these datasets, the "rule of 5" does not distinguish "drugs" from "nondrugs". Therefore, Pareto analyses were performed to examine skewed distributions in all compd. collections. Seventy percent of the "drug-like" compds. were found between the following limits: 0 ≤ HDO ≤ 2, 2 ≤ HAC ≤ 9, 2 ≤ RTB ≤ 8, and 1 ≤ RNG ≤ 4, resp. The no. of launched drugs in MDDR having 0 ≤ HDO ≤ 2 is 4.8 times higher than the no. of drugs having 3 ≤ HDO ≤ 5. Skewed distributions can be exploited to focus on the "drug-like space": 62.68% of ACDF ("nondrug-like") compds. have 0 ≤ RNG ≤ 2, and RGB ≤ 17, while 28.88% of ACDF compds. have 3 ≤ RNG ≤ 13, and 18 ≤ RGB ≤ 56. By contrast, 61.22% of MDDRF compds. have RNG ≥ 3, and RGB ≥ 18, and only 24.73% of MDDRF compds. have 0 ≤ RNG ≤ 2 rings, and RGB ≤ 17. The probability of identifying "drug-like" structures increases with mol. complexity.
  82. 82
    Hunenberger, P. H.; Helms, V.; Narayana, N.; Taylor, S. S.; McCammon, J. A. Determinants of ligand binding to cAMP-dependent protein kinase Biochemistry 1999, 38, 2358– 2366
    Google Scholar
    There is no corresponding record for this reference.
  83. 83
    Brown, S. P.; Muchmore, S. W.; Hajduk, P. J. Healthy skepticism: assessing realistic model performance Drug Discovery Today 2009, 14, 420– 427
    Google Scholar
    83
    Healthy skepticism: assessing realistic model performance
    Brown Scott P; Muchmore Steven W; Hajduk Philip J
    Drug discovery today (2009), 14 (7-8), 420-7 ISSN:.
    Although the development of computational models to aid drug discovery has become an integral part of pharmaceutical research, the application of these models often fails to produce the expected impact on productivity. One reason for this may be that the expected performance of many models is simply not supported by the underlying data, because of often neglected effects of assay and prediction errors on the reliability of the predicted outcome. Another significant challenge to realizing the full potential of computational models is their integration into prospective medicinal chemistry campaigns. This article will analyze the impact of assay and prediction error on model quality, and explore scenarios where computational models can expect to have a significant influence on drug discovery research.
  84. 84
    Czodrowski, P.; Sotriffer, C. A.; Klebe, G. Protonation changes upon ligand binding to trypsin and thrombin: structural interpretation based on pKa calculations and ITC experiments J. Mol. Biol. 2007, 367, 1347– 1356
    Google Scholar
    84
    Protonation Changes upon Ligand Binding to Trypsin and Thrombin: Structural Interpretation Based on pKa Calculations and ITC Experiments
    Czodrowski, Paul; Sotriffer, Christoph A.; Klebe, Gerhard
    Journal of Molecular Biology (2007), 367 (5), 1347-1356CODEN: JMOBAK; ISSN:0022-2836. (Elsevier Ltd.)
    The protonation states of a protein and a ligand can be altered upon complex formation. Such changes can be detected exptl. by isothermal titrn. calorimetry (ITC). For a series of ligands binding to the serine proteases trypsin and thrombin, we previously performed an extensive ITC and crystallog. study and were able to identify protonation changes for four complexes. However, since ITC measures only the overall proton exchange, it does not provide structural insights into the functional groups involved in the proton transfer. Using Poisson-Boltzmann calcns. based on our recently developed PEOE_PB charges, we compute pKa values for all complexes of our former study in order to reveal the residues with altered protonation states. The results indicate that His57, a member of the catalytic triad, is responsible for the most relevant pKa shifts leading to the exptl. detected protonation changes. This finding is in contrast to our previous assumption that the obsd. protonation changes occur at the carboxylic group of the ligands. The newly detected proton acceptor is used for a revised factorization of the ITC data, which is necessary whenever the protonation inventory changes upon complexation. The pK a values of complexes showing no protonation change in the ITC expt. are reliably predicted in most cases, whereas predictions of strongly coupled systems remain problematic.
  85. 85
    Chow, M. A.; McElroy, K. E.; Corbett, K. D.; Berger, J. M.; Kirsch, J. F. Narrowing substrate specificity in a directly evolved enzyme: the A293D mutant of aspartate aminotransferase Biochemistry 2004, 43, 12780– 12787
    Google Scholar
    There is no corresponding record for this reference.
  86. 86
    Goldberg, R. N.; Kishore, N.; Lennen, R. M. Thermodynamic Quantities for the Ionization Reactions of Buffers J. Phys. Chem. Ref. Data 2002, 31, 231– 370
    Google Scholar
    86
    Thermodynamic quantities for the ionization reactions of buffers
    Goldberg, Robert N.; Kishore, Nand; Lennen, Rebecca M.
    Journal of Physical and Chemical Reference Data (2002), 31 (2), 231-370CODEN: JPCRBU; ISSN:0047-2689. (American Institute of Physics)
    This review contains selected values of thermodn. quantities for the aq. ionization reactions of 64 buffers, many of which are used in biol. research. Since the aim is to be able to predict values of the ionization const. at temps. not too far from ambient, the thermodn. quantities which are tabulated are the pK, std. molar Gibbs energy ΔrG°, std. molar enthalpy ΔrH°, and std. molar heat capacity change ΔrCp° for each of the ionization reactions at the temp. T = 298.15 K and the pressure p = 0.1 MPa. The std. state is the hypothetical ideal soln. of unit molality. The chem. name(s) and CAS registry no., structure, empirical formula, and mol. wt. are given for each buffer considered herein. The selection of the values of the thermodn. quantities for each buffer is discussed.

Cited By

Click to copy section linkSection link copied!
Citation Statements
Explore this article's citation statements on scite.ai
powered by  Scite

This article is cited by 124 publications.

  1. Mahmoud A. Alelaimat, Mahmoud A. Al-Sha’er, Haneen A. Basheer. Novel Sulfonamide–Triazine Hybrid Derivatives: Docking, Synthesis, and Biological Evaluation as Anticancer Agents. ACS Omega 2023, 8 (15) , 14247-14263. https://doi.org/10.1021/acsomega.3c01273
  2. Debarati DasGupta, Wallace K. B. Chan, Heather A. Carlson. Computational Identification of Possible Allosteric Sites and Modulators of the SARS-CoV-2 Main Protease. Journal of Chemical Information and Modeling 2022, 62 (3) , 618-626. https://doi.org/10.1021/acs.jcim.1c01223
  3. Nicolas Tielker Lukas Eberlein Oliver Beckstein Stefan Güssregen Bogdan I. Iorga Stefan M. Kast Shuai Liu . Perspective on the SAMPL and D3R Blind Prediction Challenges for Physics-Based Free Energy Methods. , 67-107. https://doi.org/10.1021/bk-2021-1397.ch003
  4. Richard D. Smith, Heather A. Carlson. Identification of Cryptic Binding Sites Using MixMD with Standard and Accelerated Molecular Dynamics. Journal of Chemical Information and Modeling 2021, 61 (3) , 1287-1299. https://doi.org/10.1021/acs.jcim.0c01002
  5. Jonathan Fine, Matthew Muhoberac, Guillaume Fraux, Gaurav Chopra. DUBS: A Framework for Developing Directory of Useful Benchmarking Sets for Virtual Screening. Journal of Chemical Information and Modeling 2020, 60 (9) , 4137-4143. https://doi.org/10.1021/acs.jcim.0c00122
  6. Hussein Hassan-Harrirou, Ce Zhang, Thomas Lemmin. RosENet: Improving Binding Affinity Prediction by Leveraging Molecular Mechanics Energies with an Ensemble of 3D Convolutional Neural Networks. Journal of Chemical Information and Modeling 2020, 60 (6) , 2791-2802. https://doi.org/10.1021/acs.jcim.0c00075
  7. W. Patrick Walters, , Renxiao Wang. New Trends in Virtual Screening. Journal of Chemical Information and Modeling 2019, 59 (9) , 3603-3604. https://doi.org/10.1021/acs.jcim.9b00728
  8. Kelly L. Damm-Ganamet, Nidhi Arora, Stephane Becart, James P. Edwards, Alec D. Lebsack, Heather M. McAllister, Marina I. Nelen, Navin L. Rao, Lori Westover, John J. M. Wiener, Taraneh Mirzadegan. Accelerating Lead Identification by High Throughput Virtual Screening: Prospective Case Studies from the Pharmaceutical Industry. Journal of Chemical Information and Modeling 2019, 59 (5) , 2046-2062. https://doi.org/10.1021/acs.jcim.8b00941
  9. Minyi Su, Qifan Yang, Yu Du, Guoqin Feng, Zhihai Liu, Yan Li, Renxiao Wang. Comparative Assessment of Scoring Functions: The CASF-2016 Update. Journal of Chemical Information and Modeling 2019, 59 (2) , 895-913. https://doi.org/10.1021/acs.jcim.8b00545
  10. Lena Kalinowsky, Julia Weber, Shantheya Balasupramaniam, Knut Baumann, Ewgenij Proschak. A Diverse Benchmark Based on 3D Matched Molecular Pairs for Validating Scoring Functions. ACS Omega 2018, 3 (5) , 5704-5714. https://doi.org/10.1021/acsomega.7b01194
  11. Bing Xie, Trung Hai Nguyen, and David D. L. Minh . Absolute Binding Free Energies between T4 Lysozyme and 141 Small Molecules: Calculations Based on Multiple Rigid Receptor Configurations. Journal of Chemical Theory and Computation 2017, 13 (6) , 2930-2944. https://doi.org/10.1021/acs.jctc.6b01183
  12. Sandeep Sundriyal, Sébastien Moniot, Zimam Mahmud, Shang Yao, Paolo Di Fruscia, Christopher R. Reynolds, David T. Dexter, Michael J. E. Sternberg, Eric W.-F. Lam, Clemens Steegborn, and Matthew J. Fuchter . Thienopyrimidinone Based Sirtuin-2 (SIRT2)-Selective Inhibitors Bind in the Ligand Induced Selectivity Pocket. Journal of Medicinal Chemistry 2017, 60 (5) , 1928-1945. https://doi.org/10.1021/acs.jmedchem.6b01690
  13. Hervé Hogues, Traian Sulea, and Enrico O. Purisima . Evaluation of the Wilma-SIE Virtual Screening Method in Community Structure–Activity Resource 2013 and 2014 Blind Challenges. Journal of Chemical Information and Modeling 2016, 56 (6) , 955-964. https://doi.org/10.1021/acs.jcim.5b00278
  14. Ashutosh Kumar and Kam Y. J. Zhang . Application of Shape Similarity in Pose Selection and Virtual Screening in CSARdock2014 Exercise. Journal of Chemical Information and Modeling 2016, 56 (6) , 965-973. https://doi.org/10.1021/acs.jcim.5b00279
  15. Virginie Y. Martiny, François Martz, Edithe Selwa, and Bogdan I. Iorga . Blind Pose Prediction, Scoring, and Affinity Ranking of the CSAR 2014 Dataset. Journal of Chemical Information and Modeling 2016, 56 (6) , 996-1003. https://doi.org/10.1021/acs.jcim.5b00337
  16. Matthew P. Baumgartner and Carlos J. Camacho . Choosing the Optimal Rigid Receptor for Docking and Scoring in the CSAR 2013/2014 Experiment. Journal of Chemical Information and Modeling 2016, 56 (6) , 1004-1012. https://doi.org/10.1021/acs.jcim.5b00338
  17. Sergei Grudinin, Petr Popov, Emilie Neveu, and Georgy Cheremovskiy . Predicting Binding Poses and Affinities in the CSAR 2013–2014 Docking Exercises Using the Knowledge-Based Convex-PL Potential. Journal of Chemical Information and Modeling 2016, 56 (6) , 1053-1062. https://doi.org/10.1021/acs.jcim.5b00339
  18. Richard D. Smith, Kelly L. Damm-Ganamet, James B. Dunbar, Jr., Aqeel Ahmed, Krishnapriya Chinnaswamy, James E. Delproposto, Ginger M. Kubish, Christine E. Tinberg, Sagar D. Khare, Jiayi Dou, Lindsey Doyle, Jeanne A. Stuckey, David Baker, and Heather A. Carlson . CSAR Benchmark Exercise 2013: Evaluation of Results from a Combined Computational Protein Design, Docking, and Scoring/Ranking Challenge. Journal of Chemical Information and Modeling 2016, 56 (6) , 1022-1031. https://doi.org/10.1021/acs.jcim.5b00387
  19. Philip Prathipati and Kenji Mizuguchi . Integration of Ligand and Structure Based Approaches for CSAR-2014. Journal of Chemical Information and Modeling 2016, 56 (6) , 974-987. https://doi.org/10.1021/acs.jcim.5b00477
  20. Heather A. Carlson, Richard D. Smith, Kelly L. Damm-Ganamet, Jeanne A. Stuckey, Aqeel Ahmed, Maire A. Convery, Donald O. Somers, Michael Kranz, Patricia A. Elkins, Guanglei Cui, Catherine E. Peishoff, Millard H. Lambert, and James B. Dunbar, Jr. . CSAR 2014: A Benchmark Exercise Using Unpublished Data from Pharma. Journal of Chemical Information and Modeling 2016, 56 (6) , 1063-1077. https://doi.org/10.1021/acs.jcim.5b00523
  21. Xiaolei Zhu, Woong-Hee Shin, Hyungrae Kim, and Daisuke Kihara . Combined Approach of Patch-Surfer and PL-PatchSurfer for Protein–Ligand Binding Prediction in CSAR 2013 and 2014. Journal of Chemical Information and Modeling 2016, 56 (6) , 1088-1099. https://doi.org/10.1021/acs.jcim.5b00625
  22. Heather A. Carlson . Lessons Learned over Four Benchmark Exercises from the Community Structure–Activity Resource. Journal of Chemical Information and Modeling 2016, 56 (6) , 951-954. https://doi.org/10.1021/acs.jcim.6b00182
  23. Bernd Kuhn, Wolfgang Guba, Jérôme Hert, David Banner, Caterina Bissantz, Simona Ceccarelli, Wolfgang Haap, Matthias Körner, Andreas Kuglstatter, Christian Lerner, Patrizio Mattei, Werner Neidhart, Emmanuel Pinard, Markus G. Rudolph, Tanja Schulz-Gasch, Thomas Woltering, and Martin Stahl . A Real-World Perspective on Molecular Design. Journal of Medicinal Chemistry 2016, 59 (9) , 4087-4102. https://doi.org/10.1021/acs.jmedchem.5b01875
  24. Jacob D. Durrant, Kathryn E. Carlson, Teresa A. Martin, Tavina L. Offutt, Christopher G. Mayne, John A. Katzenellenbogen, and Rommie E. Amaro . Neural-Network Scoring Functions Identify Structurally Novel Estrogen-Receptor Ligands. Journal of Chemical Information and Modeling 2015, 55 (9) , 1953-1961. https://doi.org/10.1021/acs.jcim.5b00241
  25. Jie Liu and Renxiao Wang . Classification of Current Scoring Functions. Journal of Chemical Information and Modeling 2015, 55 (3) , 475-482. https://doi.org/10.1021/ci500731a
  26. Joffrey Gabel, Jérémy Desaphy, and Didier Rognan . Beware of Machine Learning-Based Scoring Functions—On the Danger of Developing Black Boxes. Journal of Chemical Information and Modeling 2014, 54 (10) , 2807-2815. https://doi.org/10.1021/ci500406k
  27. Yan Li, Zhihai Liu, Jie Li, Li Han, Jie Liu, Zhixiong Zhao, and Renxiao Wang . Comparative Assessment of Scoring Functions on an Updated Benchmark: 1. Compilation of the Test Set. Journal of Chemical Information and Modeling 2014, 54 (6) , 1700-1716. https://doi.org/10.1021/ci500080q
  28. Yan Li, Li Han, Zhihai Liu, and Renxiao Wang . Comparative Assessment of Scoring Functions on an Updated Benchmark: 2. Evaluation Methods and General Results. Journal of Chemical Information and Modeling 2014, 54 (6) , 1717-1736. https://doi.org/10.1021/ci500081m
  29. Pedro J. Ballester, Adrian Schreyer, and Tom L. Blundell . Does a More Precise Chemical Description of Protein–Ligand Complexes Lead to More Accurate Prediction of Binding Affinity?. Journal of Chemical Information and Modeling 2014, 54 (3) , 944-955. https://doi.org/10.1021/ci500091r
  30. Sergey A. Samsonov, Jan-Philip Gehrcke, and M. Teresa Pisabarro . Flexibility and Explicit Solvent in Molecular-Dynamics-Based Docking of Protein–Glycosaminoglycan Systems. Journal of Chemical Information and Modeling 2014, 54 (2) , 582-592. https://doi.org/10.1021/ci4006047
  31. Xiaohua Zhang, Sergio E. Wong, and Felice C. Lightstone . Toward Fully Automated High Performance Computing Drug Discovery: A Massively Parallel Virtual Screening Pipeline for Docking and Molecular Mechanics/Generalized Born Surface Area Rescoring to Improve Enrichment. Journal of Chemical Information and Modeling 2014, 54 (1) , 324-337. https://doi.org/10.1021/ci4005145
  32. Gregory A. Ross, Garrett M. Morris, and Philip C. Biggin . One Size Does Not Fit All: The Limits of Structure-Based Models in Drug Discovery. Journal of Chemical Theory and Computation 2013, 9 (9) , 4266-4274. https://doi.org/10.1021/ct4004228
  33. David Ryan Koes, Matthew P. Baumgartner, and Carlos J. Camacho . Lessons Learned in Empirical Scoring with smina from the CSAR 2011 Benchmarking Exercise. Journal of Chemical Information and Modeling 2013, 53 (8) , 1893-1904. https://doi.org/10.1021/ci300604z
  34. Kelly L. Damm-Ganamet, Richard D. Smith, James B. Dunbar, Jr., Jeanne A. Stuckey, and Heather A. Carlson . CSAR Benchmark Exercise 2011–2012: Evaluation of Results from Docking and Relative Ranking of Blinded Congeneric Series. Journal of Chemical Information and Modeling 2013, 53 (8) , 1853-1870. https://doi.org/10.1021/ci400025f
  35. Sam Z. Grinter, Chengfei Yan, Sheng-You Huang, Lin Jiang, and Xiaoqin Zou . Automated Large-Scale File Preparation, Docking, and Scoring: Evaluation of ITScore and STScore Using the 2012 Community Structure–Activity Resource Benchmark. Journal of Chemical Information and Modeling 2013, 53 (8) , 1905-1914. https://doi.org/10.1021/ci400045v
  36. James B. Dunbar, Jr., Richard D. Smith, Kelly L. Damm-Ganamet, Aqeel Ahmed, Emilio Xavier Esposito, James Delproposto, Krishnapriya Chinnaswamy, You-Na Kang, Ginger Kubish, Jason E. Gestwicki, Jeanne A. Stuckey, and Heather A. Carlson . CSAR Data Set Release 2012: Ligands, Affinities, Complexes, and Docking Decoys. Journal of Chemical Information and Modeling 2013, 53 (8) , 1842-1852. https://doi.org/10.1021/ci4000486
  37. David Zilian and Christoph A. Sotriffer . SFCscoreRF: A Random Forest-Based Scoring Function for Improved Affinity Prediction of Protein–Ligand Complexes. Journal of Chemical Information and Modeling 2013, 53 (8) , 1923-1933. https://doi.org/10.1021/ci400120b
  38. Heather A. Carlson . Check Your Confidence: Size Really Does Matter. Journal of Chemical Information and Modeling 2013, 53 (8) , 1837-1841. https://doi.org/10.1021/ci4004249
  39. Vladimir Chupakhin, Gilles Marcou, Igor Baskin, Alexandre Varnek, and Didier Rognan . Predicting Ligand Binding Modes from Neural Networks Trained on Protein–Ligand Interaction Fingerprints. Journal of Chemical Information and Modeling 2013, 53 (4) , 763-772. https://doi.org/10.1021/ci300200r
  40. Anshuman Dixit and Gennady M. Verkhivker . Integrating Ligand-Based and Protein-Centric Virtual Screening of Kinase Inhibitors Using Ensembles of Multiple Protein Kinase Genes and Conformations. Journal of Chemical Information and Modeling 2012, 52 (10) , 2501-2515. https://doi.org/10.1021/ci3002638
  41. Jamel Meslamani, Jiabo Li, Jon Sutter, Adrian Stevens, Hugues-Olivier Bertrand, and Didier Rognan . Protein–Ligand-Based Pharmacophores: Generation and Utility Assessment in Computational Ligand Profiling. Journal of Chemical Information and Modeling 2012, 52 (4) , 943-955. https://doi.org/10.1021/ci300083r
  42. Mei Li, Ye Cao, Xiaoguang Liu, Hua Ji. Structure-Aware Graph Attention Diffusion Network for Protein–Ligand Binding Affinity Prediction. IEEE Transactions on Neural Networks and Learning Systems 2024, 35 (12) , 18370-18380. https://doi.org/10.1109/TNNLS.2023.3314928
  43. , Pratip K. Chaskar, Rakesh R. Somani, . EXPLORING NOVEL NCEs TARGETING InhA AS PROSPECTIVE KEY COMPOUNDS TO TREAT TUBERCULAR INFECTIONS: A COMPUTATIONAL APPROACH. INDIAN DRUGS 2024, 61 (09) , 21-35. https://doi.org/10.53879/id.61.09.14685
  44. Tobias Harren, Torben Gutermuth, Christoph Grebner, Gerhard Hessler, Matthias Rarey. Modern machine‐learning for binding affinity estimation of protein–ligand complexes: Progress, opportunities, and challenges. WIREs Computational Molecular Science 2024, 14 (3) https://doi.org/10.1002/wcms.1716
  45. Seokhyun Moon, Wonho Zhung, Woo Youn Kim. Toward generalizable structure‐based deep learning models for protein–ligand interaction prediction: Challenges and strategies. WIREs Computational Molecular Science 2024, 14 (1) https://doi.org/10.1002/wcms.1705
  46. Xianjin Xu, Rui Duan, Xiaoqin Zou. Template‐guided method for protein–ligand complex structure prediction: Application to CASP15 protein–ligand studies. Proteins: Structure, Function, and Bioinformatics 2023, 91 (12) , 1829-1836. https://doi.org/10.1002/prot.26535
  47. Nikolai Schapin, Maciej Majewski, Alejandro Varela-Rial, Carlos Arroniz, Gianni De Fabritiis. Machine learning small molecule properties in drug discovery. Artificial Intelligence Chemistry 2023, 1 (2) , 100020. https://doi.org/10.1016/j.aichem.2023.100020
  48. Swapnil Wagle, Richard D. Smith, Anthony J. Dominic, Debarati DasGupta, Sunil Kumar Tripathi, Heather A. Carlson. Sunsetting Binding MOAD with its last data update and the addition of 3D-ligand polypharmacology tools. Scientific Reports 2023, 13 (1) https://doi.org/10.1038/s41598-023-29996-w
  49. Mingwei Pang, Wangqiu He, Xufeng Lu, Yuting She, Liangxu Xie, Ren Kong, Shan Chang. CoDock-Ligand: combined template-based docking and CNN-based scoring in ligand binding prediction. BMC Bioinformatics 2023, 24 (1) https://doi.org/10.1186/s12859-023-05571-y
  50. Renato Araújo da Costa, Andréia do Socorro Silva da Costa, João Augusto Pereira da Rocha, Marlon Ramires da Costa Lima, Elaine Cristina Medeiros da Rocha, Fabiana Cristina de Araújo Nascimento, Anderson José Baia Gomes, José de Arimatéia Rodrigues do Rego, Davi do Socorro Barros Brasil. Exploring Natural Alkaloids from Brazilian Biodiversity as Potential Inhibitors of the Aedes aegypti Juvenile Hormone Enzyme: A Computational Approach for Vector Mosquito Control. Molecules 2023, 28 (19) , 6871. https://doi.org/10.3390/molecules28196871
  51. Sumit Arora, Pallavi Rushiya, Kalpana Tirpude, Nidhi Sapkal, Subhash Yende, Abhay Ittadwar, Sapan Shah. Exploring the phytoconstituents targeting TNF-α as potential lead compounds to treat inflammatory diseases: an in-silico approach. Digital Chinese Medicine 2022, 5 (3) , 264-275. https://doi.org/10.1016/j.dcmed.2022.10.003
  52. Chao Yang, Eric Anthony Chen, Yingkai Zhang. Protein–Ligand Docking in the Machine-Learning Era. Molecules 2022, 27 (14) , 4568. https://doi.org/10.3390/molecules27144568
  53. Rocco Meli, Garrett M. Morris, Philip C. Biggin. Scoring Functions for Protein-Ligand Binding Affinity Prediction Using Structure-based Deep Learning: A Review. Frontiers in Bioinformatics 2022, 2 https://doi.org/10.3389/fbinf.2022.885983
  54. Heba A. Gad, Reham S. Elezaby, Mai Mansour, Rania M. Hathout. Novel Approaches of Solid Lipid Nanoparticles as Drug Carrier. 2022, 107-143. https://doi.org/10.1002/9783527832095.ch5
  55. Juan I. Di Filippo, Claudio N. Cavasotto. Guided structure-based ligand identification and design via artificial intelligence modeling. Expert Opinion on Drug Discovery 2022, 17 (1) , 71-78. https://doi.org/10.1080/17460441.2021.1979514
  56. Kevin Crampon, Alexis Giorkallos, Myrtille Deldossi, Stéphanie Baud, Luiz Angelo Steffenel. Machine-learning methods for ligand–protein molecular docking. Drug Discovery Today 2022, 27 (1) , 151-164. https://doi.org/10.1016/j.drudis.2021.09.007
  57. Mohammed Bule, Nafiseh Jalalimanesh, Zahra Bayrami, Maryam Baeeri, Mohammad Abdollahi. The rise of deep learning and transformations in bioactivity prediction power of molecular modeling tools. Chemical Biology & Drug Design 2021, 98 (5) , 954-967. https://doi.org/10.1111/cbdd.13750
  58. Didier Rognan. Modeling Protein-Ligand Interactions: Are We Ready for Deep Learning?. 2021, 163-173. https://doi.org/10.1016/B978-0-12-801238-3.11521-1
  59. Bruno Junior Neves, Melina Mottin, José Teofilo Moreira-Filho, Bruna Katiele de Paula Sousa, Sabrina Silva Mendonca, Carolina Horta Andrade. Best Practices for Docking-Based Virtual Screening. 2021, 75-98. https://doi.org/10.1016/B978-0-12-822312-3.00001-1
  60. Stefan Holderbach, Lukas Adam, B. Jayaram, Rebecca C. Wade, Goutam Mukherjee. RASPD+: Fast Protein-Ligand Binding Free Energy Prediction Using Simplified Physicochemical Features. Frontiers in Molecular Biosciences 2020, 7 https://doi.org/10.3389/fmolb.2020.601065
  61. G. Poda, V. Tanchuk. Computational Methods for the Discovery of Chemical Probes. 2020, 39-68. https://doi.org/10.1039/9781839160745-00039
  62. Dinesh Kumar Sriramulu, Sangwook Wu, Sun-Gu Lee. Effect of ligand torsion number on the AutoDock mediated prediction of protein-ligand binding affinity. Journal of Industrial and Engineering Chemistry 2020, 83 , 359-365. https://doi.org/10.1016/j.jiec.2019.12.009
  63. Chao Shen, Junjie Ding, Zhe Wang, Dongsheng Cao, Xiaoqin Ding, Tingjun Hou. From machine learning to deep learning: Advances in scoring functions for protein–ligand docking. WIREs Computational Molecular Science 2020, 10 (1) https://doi.org/10.1002/wcms.1429
  64. Maha Thafar, Arwa Bin Raies, Somayah Albaradei, Magbubah Essack, Vladimir B. Bajic. Comparison Study of Computational Prediction Tools for Drug-Target Binding Affinities. Frontiers in Chemistry 2019, 7 https://doi.org/10.3389/fchem.2019.00782
  65. Tingting Shi, Yingwu Yang, Shuheng Huang, Linxin Chen, Zuyin Kuang, Yu Heng, Hu Mei. Molecular image-based convolutional neural network for the prediction of ADMET properties. Chemometrics and Intelligent Laboratory Systems 2019, 194 , 103853. https://doi.org/10.1016/j.chemolab.2019.103853
  66. Pedro H. M. Torres, Ana C. R. Sodero, Paula Jofily, Floriano P. Silva-Jr. Key Topics in Molecular Docking for Drug Design. International Journal of Molecular Sciences 2019, 20 (18) , 4574. https://doi.org/10.3390/ijms20184574
  67. Jeffrey R. Wagner, Christopher P. Churas, Shuai Liu, Robert V. Swift, Michael Chiu, Chenghua Shao, Victoria A. Feher, Stephen K. Burley, Michael K. Gilson, Rommie E. Amaro. Continuous Evaluation of Ligand Protein Predictions: A Weekly Community Challenge for Drug Docking. Structure 2019, 27 (8) , 1326-1335.e4. https://doi.org/10.1016/j.str.2019.05.012
  68. Xianjin Xu, Zhiwei Ma, Rui Duan, Xiaoqin Zou. Predicting protein–ligand binding modes for CELPP and GC3: workflows and insight. Journal of Computer-Aided Molecular Design 2019, 33 (3) , 367-374. https://doi.org/10.1007/s10822-019-00185-0
  69. Shailesh Kumar Panday, Indira Ghosh. In Silico Structure-Based Prediction of Receptor–Ligand Binding Affinity: Current Progress and Challenges. 2019, 109-175. https://doi.org/10.1007/978-3-030-05282-9_5
  70. Zied Gaieb, Conor D. Parks, Michael Chiu, Huanwang Yang, Chenghua Shao, W. Patrick Walters, Millard H. Lambert, Neysa Nevins, Scott D. Bembenek, Michael K. Ameriks, Tara Mirzadegan, Stephen K. Burley, Rommie E. Amaro, Michael K. Gilson. D3R Grand Challenge 3: blind prediction of protein–ligand poses and affinity rankings. Journal of Computer-Aided Molecular Design 2019, 33 (1) , 1-18. https://doi.org/10.1007/s10822-018-0180-4
  71. Joshua Hochuli, Alec Helbling, Tamar Skaist, Matthew Ragoza, David Ryan Koes. Visualizing convolutional neural network protein-ligand scoring. Journal of Molecular Graphics and Modelling 2018, 84 , 96-108. https://doi.org/10.1016/j.jmgm.2018.06.005
  72. Mingyue Zheng, Jihui Zhao, Chen Cui, Zunyun Fu, Xutong Li, Xiaohong Liu, Xiaoyu Ding, Xiaoqin Tan, Fei Li, Xiaomin Luo, Kaixian Chen, Hualiang Jiang. Computational chemical biology and drug design: Facilitating protein structure, function, and modulation studies. Medicinal Research Reviews 2018, 38 (3) , 914-950. https://doi.org/10.1002/med.21483
  73. Yan Li, Minyi Su, Zhihai Liu, Jie Li, Jie Liu, Li Han, Renxiao Wang. Assessing protein–ligand interaction scoring functions with the CASF-2013 benchmark. Nature Protocols 2018, 13 (4) , 666-680. https://doi.org/10.1038/nprot.2017.114
  74. Hossam M. Ashtawy, Nihar R. Mahapatra. Boosted neural networks scoring functions for accurate ligand docking and ranking. Journal of Bioinformatics and Computational Biology 2018, 16 (02) , 1850004. https://doi.org/10.1142/S021972001850004X
  75. Priscila da Silva Figueiredo Celestino Gomes, Franck Da Silva, Guillaume Bret, Didier Rognan. Ranking docking poses by graph matching of protein–ligand interactions: lessons learned from the D3R Grand Challenge 2. Journal of Computer-Aided Molecular Design 2018, 32 (1) , 75-87. https://doi.org/10.1007/s10822-017-0046-1
  76. Maria Kadukova, Sergei Grudinin. Docking of small molecules to farnesoid X receptors using AutoDock Vina with the Convex-PL potential: lessons learned from D3R Grand Challenge 2. Journal of Computer-Aided Molecular Design 2018, 32 (1) , 151-162. https://doi.org/10.1007/s10822-017-0062-1
  77. Bentley M. Wingert, Rick Oerlemans, Carlos J. Camacho. Optimal affinity ranking for automated virtual screening validated in prospective D3R grand challenges. Journal of Computer-Aided Molecular Design 2018, 32 (1) , 287-297. https://doi.org/10.1007/s10822-017-0065-y
  78. Matthew P. Baumgartner, David A. Evans. Lessons learned in induced fit docking and metadynamics in the Drug Design Data Resource Grand Challenge 2. Journal of Computer-Aided Molecular Design 2018, 32 (1) , 45-58. https://doi.org/10.1007/s10822-017-0081-y
  79. Antonia S. J. S. Mey, Jordi Juárez Jiménez, Julien Michel. Impact of domain knowledge on blinded predictions of binding energies by alchemical free energy calculations. Journal of Computer-Aided Molecular Design 2018, 32 (1) , 199-210. https://doi.org/10.1007/s10822-017-0083-9
  80. Zied Gaieb, Shuai Liu, Symon Gathiaka, Michael Chiu, Huanwang Yang, Chenghua Shao, Victoria A. Feher, W. Patrick Walters, Bernd Kuhn, Markus G. Rudolph, Stephen K. Burley, Michael K. Gilson, Rommie E. Amaro. D3R Grand Challenge 2: blind prediction of protein–ligand poses, affinity rankings, and relative binding free energies. Journal of Computer-Aided Molecular Design 2018, 32 (1) , 1-20. https://doi.org/10.1007/s10822-017-0088-4
  81. Jeffrey R. Wagner, Christopher P. Churas, Shuai Liu, Robert V. Swift, Michael Chiu, Chenghua Shao, Victoria A. Feher, Stephen K. Burley, Michael K. Gilson, Rommie E. Amaro. Continuous Evaluation of Ligand Protein Predictions: A Weekly Community Challenge for Drug Docking. SSRN Electronic Journal 2018, 9 https://doi.org/10.2139/ssrn.3291330
  82. Jie Liu, Minyi Su, Zhihai Liu, Jie Li, Yan Li, Renxiao Wang. Enhance the performance of current scoring functions with the aid of 3D protein-ligand interaction fingerprints. BMC Bioinformatics 2017, 18 (1) https://doi.org/10.1186/s12859-017-1750-5
  83. Maria Kadukova, Sergei Grudinin. Convex-PL: a novel knowledge-based potential for protein-ligand interactions deduced from structural databases using convex optimization. Journal of Computer-Aided Molecular Design 2017, 31 (10) , 943-958. https://doi.org/10.1007/s10822-017-0068-8
  84. Xianjin Xu, Chengfei Yan, Xiaoqin Zou. Improving binding mode and binding affinity predictions of docking by ligand-based search of protein conformations: evaluation in D3R grand challenge 2015. Journal of Computer-Aided Molecular Design 2017, 31 (8) , 689-699. https://doi.org/10.1007/s10822-017-0038-1
  85. J. Cole, E. Davis, G. Jones, C.R. Sage. Molecular Docking—A Solved Problem?. 2017, 297-318. https://doi.org/10.1016/B978-0-12-409547-2.12352-2
  86. Zhiqiang Yan, Jin Wang. Scoring Functions of Protein-Ligand Interactions. 2017, 915-940. https://doi.org/10.4018/978-1-5225-0549-5.ch036
  87. Lukas P. Pason, Christoph A. Sotriffer. Empirical Scoring Functions for Affinity Prediction of Protein‐ligand Complexes. Molecular Informatics 2016, 35 (11-12) , 541-548. https://doi.org/10.1002/minf.201600048
  88. S Balaji. Modelling and Simulation of Biological Systems in Medical Applications. 2016, 233-268. https://doi.org/10.1201/9781315368863-11
  89. Ashutosh Kumar, Kam Y. J. Zhang. Prospective evaluation of shape similarity based pose prediction method in D3R Grand Challenge 2015. Journal of Computer-Aided Molecular Design 2016, 30 (9) , 685-693. https://doi.org/10.1007/s10822-016-9931-2
  90. Zhaofeng Ye, Matthew P. Baumgartner, Bentley M. Wingert, Carlos J. Camacho. Optimal strategies for virtual screening of induced-fit and flexible target in the 2015 D3R Grand Challenge. Journal of Computer-Aided Molecular Design 2016, 30 (9) , 695-706. https://doi.org/10.1007/s10822-016-9941-0
  91. Symon Gathiaka, Shuai Liu, Michael Chiu, Huanwang Yang, Jeanne A. Stuckey, You Na Kang, Jim Delproposto, Ginger Kubish, James B. Dunbar, Heather A. Carlson, Stephen K. Burley, W. Patrick Walters, Rommie E. Amaro, Victoria A. Feher, Michael K. Gilson. D3R grand challenge 2015: Evaluation of protein–ligand pose and affinity predictions. Journal of Computer-Aided Molecular Design 2016, 30 (9) , 651-668. https://doi.org/10.1007/s10822-016-9946-8
  92. Jocelyn Sunseri, Matthew Ragoza, Jasmine Collins, David Ryan Koes. A D3R prospective evaluation of machine learning for protein-ligand scoring. Journal of Computer-Aided Molecular Design 2016, 30 (9) , 761-771. https://doi.org/10.1007/s10822-016-9960-x
  93. Sergei Grudinin, Maria Kadukova, Andreas Eisenbarth, Simon Marillet, Frédéric Cazals. Predicting binding poses and affinities for protein - ligand complexes in the 2015 D3R Grand Challenge using a physical model with a statistical parameter estimation. Journal of Computer-Aided Molecular Design 2016, 30 (9) , 791-804. https://doi.org/10.1007/s10822-016-9976-2
  94. Philip Prathipati, Chioko Nagao, Shandar Ahmad, Kenji Mizuguchi. Improved pose and affinity predictions using different protocols tailored on the basis of data availability. Journal of Computer-Aided Molecular Design 2016, 30 (9) , 817-828. https://doi.org/10.1007/s10822-016-9982-4
  95. Vsevolod Yu Tanchuk, Volodymyr O. Tanin, Andriy I. Vovk, Gennady Poda. A New, Improved Hybrid Scoring Function for Molecular Docking and Scoring Based on AutoDock and AutoDock Vina. Chemical Biology & Drug Design 2016, 87 (4) , 618-625. https://doi.org/10.1111/cbdd.12697
  96. Jessica K. Gagnon, Sean M. Law, Charles L. Brooks. Flexible CDOCKER : Development and application of a pseudo‐explicit structure‐based docking method within CHARMM. Journal of Computational Chemistry 2016, 37 (8) , 753-762. https://doi.org/10.1002/jcc.24259
  97. Angélica Nakagawa Lima, Eric Allison Philot, Gustavo Henrique Goulart Trossini, Luis Paulo Barbour Scott, Vinícius Gonçalves Maltarollo, Kathia Maria Honorio. Use of machine learning approaches for novel drug discovery. Expert Opinion on Drug Discovery 2016, 11 (3) , 225-239. https://doi.org/10.1517/17460441.2016.1146250
  98. Mary Regina Boland, Alexandra Jacunski, Tal Lorberbaum, Joseph D. Romano, Robert Moskovitch, Nicholas P. Tatonetti. Systems biology approaches for identifying adverse drug reactions and elucidating their underlying biological mechanisms. WIREs Systems Biology and Medicine 2016, 8 (2) , 104-122. https://doi.org/10.1002/wsbm.1323
  99. Zhiqiang Yan, Jin Wang. Incorporating specificity into optimization: evaluation of SPA using CSAR 2014 and CASF 2013 benchmarks. Journal of Computer-Aided Molecular Design 2016, 30 (3) , 219-227. https://doi.org/10.1007/s10822-016-9897-0
  100. Jung Hsin Lin. Review structure‐ and dynamics‐based computational design of anticancer drugs. Biopolymers 2016, 105 (1) , 2-9. https://doi.org/10.1002/bip.22744
Load all citations
Download PDF
Go to Journal of Chemical Information and Modeling
Get e-Alerts
Get e-Alerts

Journal of Chemical Information and Modeling

Cite this: J. Chem. Inf. Model. 2011, 51, 9, 2115–2131
Click to copy citationCitation copied!
https://doi.org/10.1021/ci200269q
Published August 3, 2011

Copyright © 2011 American Chemical Society. This publication is licensed under these Terms of Use.

Request reuse permissions

Article Views

3384

Altmetric

-

Citations

124
Learn about these metrics

Article Views are the COUNTER-compliant sum of full text article downloads since November 2008 (both PDF and HTML) across all institutions and individuals. These metrics are regularly updated to reflect usage leading up to the last few days.

Citations are the number of other articles citing this article, calculated by Crossref and updated daily. Find more information about Crossref citation counts.

The Altmetric Attention Score is a quantitative measure of the attention that a research article has received online. Clicking on the donut icon will load a page at altmetric.com with additional details about the score and the social media presence for the given article. Find more information on the Altmetric Attention Score and how the score is calculated.

  • Abstract

    High Resolution Image
    Download MS PowerPoint Slide

    Figure 1

    Figure 1. Example of comparing a set of scores, pKd (calculated), to their corresponding experimentally determined affinities. (Top) When fitting a line (black) using least-squares linear regression, the distance in the y direction between each data point and the line is its residual. (Bottom) The residuals for all the data points have a normal distribution around zero. The characteristics are well-defined, including the definition of standard deviation (σ in red, which happens to be 1.4 pKd in this example) and the number of data points with residuals outside ± σ (15.8% in each tail). Higher correlations lead to larger R2 and smaller σ; weaker correlations lead to lower R2 and larger σ, but the distributions remain Gaussian in shape.

    High Resolution Image
    Download MS PowerPoint Slide

    Figure 2

    Figure 2. Crystal structure of FXa bound with a 5 pM ligand (PDB id 2p3t). The ligand is very exposed with few hydrogen bonds to the protein.

    High Resolution Image
    Download MS PowerPoint Slide

    Figure 3

    Figure 3. Least-squares linear regression of the 17 core scoring functions. Black lines are the linear regression fit. Red lines indicate +σ and −σ, the standard deviation of the residuals. Blue points are UNDER complexes which were underscored in ≥12 of the 17 functions. The red points are OVER complexes which were overscored in ≥12 of the 17 functions.

    High Resolution Image
    Download MS PowerPoint Slide

    Figure 4

    Figure 4. Comparison of experimental and calculated values from the nine functions which predicted absolute binding affinity, listed roughly in order of increasing Med |Err| and RMSE. Black lines represent perfect agreement. The red lines indicate +Med |Err| and −Med |Err| from the black line. The blue circles denote complexes for which ≥7 of the 9 methods have consistently underestimated the affinity by at least Med |Err|, while the red circles are those where the affinity was overestimated.

    High Resolution Image
    Download MS PowerPoint Slide

    Figure 5

    Figure 5. Distribution of binding affinities in the GOOD and BAD complexes (left) are compared to those of the NULL case (right). The NULL case is generated by the sets of all complexes with affinities ≤50 nM (high), 50 nM–50 μM (middle), and ≥50 μM (low). This midrange of affinities is highlighted with a wide, gray bar on both figures.

    High Resolution Image
    Download MS PowerPoint Slide

    Figure 6

    Figure 6. Distribution of amino acids in the binding sites of the GOOD and BAD complexes meeting the ≥12 of 17 definition (left) are compared to those of the NULL case (right). The graph in the lower left provides the distribution of all amino acids in the full protein sequences to show that the important trends do not result from inherent differences in composition of the proteins (the same is true of the NULLs, data not shown). Metals and modified residues are denoted as other, “OTH”. Averages and error bars for the amino acid content were determined by bootstrapping.

    High Resolution Image
    Download MS PowerPoint Slide

  • References

    This article references 86 other publications.

    1. 1
      Leach, A. R.; Shoichet, B. K.; Peishoff, C. E. Prediction of protein-ligand interactions. Docking and scoring: successes and gaps J. Med. Chem. 2006, 49, 5851– 5855
      1
      Prediction of protein-ligand interactions. Docking and scoring: Successes and gaps
      Leach, Andrew R.; Shoichet, Brian K.; Peishoff, Catherine E.
      Journal of Medicinal Chemistry (2006), 49 (20), 5851-5855CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
      A review. The current state of docking and scoring technol. are summarized, highlighting the key areas that must be addressed. Benchmarks and measurements to take up the data infrastructure on which new algorithm development depends; and possible industrial, academic, and governmental partnerships that may be needed to fulfill the potential of docking and scoring, are discussed.
    2. 2
      Warren, G. L.; Andrews, C. W.; Capelli, A.-M.; Clarke, B.; LaLonde, J.; Lambert, M. H.; Lindvall, M.; Nevins, N.; Semus, S. F.; Senger, S.; Tedesco, G.; Wall, I. D.; Woolven, J. M.; Peishoff, C. E.; Head, M. S. A critical assessment of docking programs and scoring functions J. Med. Chem. 2006, 49, 5912– 5931
      2
      A Critical Assessment of Docking Programs and Scoring Functions
      Warren, Gregory L.; Andrews, C. Webster; Capelli, Anna-Maria; Clarke, Brian; LaLonde, Judith; Lambert, Millard H.; Lindvall, Mika; Nevins, Neysa; Semus, Simon F.; Senger, Stefan; Tedesco, Giovanna; Wall, Ian D.; Woolven, James M.; Peishoff, Catherine E.; Head, Martha S.
      Journal of Medicinal Chemistry (2006), 49 (20), 5912-5931CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
      Docking is a computational technique that samples conformations of small mols. in protein binding sites; scoring functions are used to assess which of these conformations best complements the protein binding site. An evaluation of 10 docking programs and 37 scoring functions was conducted against eight proteins of seven protein types for three tasks: binding mode prediction, virtual screening for lead identification, and rank-ordering by affinity for lead optimization. All of the docking programs were able to generate ligand conformations similar to crystallog. detd. protein/ligand complex structures for at least one of the targets. However, scoring functions were less successful at distinguishing the crystallog. conformation from the set of docked poses. Docking programs identified active compds. from a pharmaceutically relevant pool of decoy compds.; however, no single program performed well for all of the targets. For prediction of compd. affinity, none of the docking programs or scoring functions made a useful prediction of ligand binding affinity.
    3. 3
      Dunbar, J. B., Jr.; Smith, R. D.; Yang, C. Y.; Ung, P. M.; Lexa, K. W.; Khazanov, N. A.; Stuckey, J. A.; Wang, S.; Carlson, H. A. CSAR Benchmark Exercise of 2010: Selection of the protein-ligand complexes J. Chem. Inf. Model. 2011,  DOI: 10.1021/ci200082t
      There is no corresponding record for this reference.
    4. 4
      Wang, R.; Lu, Y.; Wang, S. Comparative evaluation of 11 scoring functions for molecular docking J. Med. Chem. 2003, 46, 2287– 2303
      4
      Comparative Evaluation of 11 Scoring Functions for Molecular Docking
      Wang, Renxiao; Lu, Yipin; Wang, Shaomeng
      Journal of Medicinal Chemistry (2003), 46 (12), 2287-2303CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
      Eleven popular scoring functions have been tested on 100 protein-ligand complexes to evaluate their abilities to reproduce exptl. detd. structures and binding affinities. They include four scoring functions implemented in the LigFit module in Cerius2 (LigScore, PLP, PMF, and LUDI), four scoring functions implemented in the CScore module in SYBYL (F-Score, G-Score, D-Score, and ChemScore), the scoring function implemented in the AutoDock program, and two stand-alone scoring functions (DrugScore and X-Score). These scoring functions are not tested in the context of a particular docking program. Instead, conformational sampling and scoring are sepd. into two consecutive steps. First, an exhaustive conformational sampling is performed by using the AutoDock program to generate an ensemble of docked conformations for each ligand mol. This conformational ensemble is required to cover the entire conformational space as much as possible rather than to focus on a few energy min. Then, each scoring function is applied to score this conformational ensemble to see if it can identify the exptl. obsd. conformation from all of the other decoys. Among all of the scoring functions under test, six of them, i.e., PLP, F-Score, LigScore, DrugScore, LUDI, and X-Score, yield success rates higher than the AutoDock scoring function. The success rates of these six scoring functions range from 66% to 76% if using root-mean-square deviation ≤2.0 Å as the criterion. Combining any two or three of these six scoring functions into a consensus scoring scheme further improves the success rate to nearly 80% or even higher. However, when applied to reproduce the exptl. detd. binding affinities of the 100 protein-ligand complexes, only X-Score, PLP, DrugScore, and G-Score are able to give correlation coeffs. over 0.50. All of the 11 scoring functions are further inspected by their abilities to construct a descriptive, funnel-shaped energy surface for protein-ligand complexation. The results indicate that X-Score and DrugScore perform better than the other ones at this aspect.
    5. 5
      Muchmore, S. W.; Debe, D. A.; Metz, J. T.; Brown, S. P.; Martin, Y. C.; Hajduk, P. J. Application of belief theory to similarity data fusion for use in analog searching and lead hopping J. Chem. Inf. Model. 2008, 48, 941– 948
      5
      Application of Belief Theory to Similarity Data Fusion for Use in Analog Searching and Lead Hopping
      Muchmore, Steven W.; Debe, Derek A.; Metz, James T.; Brown, Scott P.; Martin, Yvonne C.; Hajduk, Philip J.
      Journal of Chemical Information and Modeling (2008), 48 (5), 941-948CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
      A wide variety of computational algorithms have been developed that strive to capture the chem. similarity between two compds. for use in virtual screening and lead discovery. One limitation of such approaches is that, while a returned similarity value reflects the perceived degree of relatedness between any two compds., there is no direct correlation between this value and the expectation or confidence that any two mols. will in fact be equally active. A lack of a common framework for interpretation of similarity measures also confounds the reliable fusion of information from different algorithms. Here, we present a probabilistic framework for interpreting similarity measures that directly correlates the similarity value to a quant. expectation that two mols. will in fact be equipotent. The approach is based on extensive benchmarking of 10 different similarity methods (MACCS keys, Daylight fingerprints, max. common subgraphs, rapid overlay of chem. structures (ROCS) shape similarity, and six connectivity-based fingerprints) against a database of more than 150 000 compds. with activity data against 23 protein targets. Given this unified and probabilistic framework for interpreting chem. similarity, principles derived from decision theory can then be applied to combine the evidence from different similarity measures in such a way that both capitalizes on the strengths of the individual approaches and maintains a quant. est. of the likelihood that any two mols. will exhibit similar biol. activity.
    6. 6
      Muchmore, S. W.; Edmunds, J. J.; Stewart, K. D.; Hajduk, P. J. Cheminformatic tools for medicinal chemists J. Med. Chem. 2010, 53, 4830– 4841
      There is no corresponding record for this reference.
    7. 7
      Swann, S. L.; Brown, S. P.; Muchmore, S. W.; Patel, H.; Merta, P.; Locklear, J.; Hajduk, P. J. A unified, probabilistic framework for structure- and ligand-based virtual screening J. Med. Chem. 2011, 54, 1223– 1232
      7
      A Unified, Probabilistic Framework for Structure- and Ligand-Based Virtual Screening
      Swann, Steven L.; Brown, Scott P.; Muchmore, Steven W.; Patel, Hetal; Merta, Philip; Locklear, John; Hajduk, Philip J.
      Journal of Medicinal Chemistry (2011), 54 (5), 1223-1232CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
      The authors present a probabilistic framework for interpreting structure-based virtual screening that returns a quant. likelihood of observing bioactivity and can be quant. combined with ligand-based screening methods to yield a cumulative prediction that consistently outperforms any single screening metric. The approach has been developed and validated on more than 30 different protein targets. Transforming structure-based in silico screening results into robust probabilities of activity enables the general fusion of multiple structure- and ligand-based approaches and returns a quant. expectation of success that can be used to prioritize (or deprioritize) further discovery activities. This unified probabilistic framework offers a paradigm shift in how docking and scoring results are interpreted, which can enhance early lead-finding efforts by maximizing the value of in silico computational tools.
    8. 8
      Baber, J. C.; Shirley, W. A.; Gao, Y.; Feher, M. The use of consensus scoring in ligand-based virtual screening J. Chem. Inf. Model. 2006, 46, 277– 288
      8
      The use of consensus scoring in ligand-based virtual screening
      Baber, J. Christian; Shirley, William A.; Gao, Yinghong; Feher, Miklos
      Journal of Chemical Information and Modeling (2006), 46 (1), 277-288CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
      A new consensus approach has been developed for ligand-based virtual screening. It involves combining highly disparate properties in order to improve performance in virtual screening. The properties include structural, 2D pharmacophore and property-based fingerprints, scores derived using BCUT descriptors, and 3D pharmacophore approaches. Different approaches for the combination of all or some of these methods have been tested. Logistic regression and sum ranks were found to be the most advantageous in different pharmaceutical applications. The three major reasons consensus scoring appears to enrich data sets better than single scoring functions are (1) using multiple scoring functions is similar to repeated samplings, in which case the mean is closer to the true value than any single value, (2) due to the better clustering of actives, multiple sampling will recover more actives than inactives, and (3) different methods seem to agree more on the ranking of the actives than on the inactives. Furthermore, consensus results are not only better but are also more consistent across receptor systems.
    9. 9
      Bar-Haim, S.; Aharon, A.; Ben-Moshe, T.; Marantz, Y.; Senderowitz, H. SeleX-CS: a new consensus scoring algorithm for hit discovery and lead optimization J. Chem. Inf. Model. 2009, 49, 623– 633
      9
      SeleX-CS: A New Consensus Scoring Algorithm for Hit Discovery and Lead Optimization
      Bar-Haim, Shay; Aharon, Ayelet; Ben-Moshe, Tal; Marantz, Yael; Senderowitz, Hanoch
      Journal of Chemical Information and Modeling (2009), 49 (3), 623-633CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
      Identifying active compds. (hits) that bind to biol. targets of pharmaceutical relevance is the cornerstone of drug design efforts. Structure based virtual screening, namely, the in silico evaluation of binding energies and geometries between a protein and its putative ligands, has emerged over the past few years as a promising approach in this field. The success of the method relies on the availability of reliable 3-dimensional (3D) structures of the target protein and its candidate ligands (the screening library), a reliable docking method that can fit the different ligands into the protein's binding site, and an accurate scoring function that can rank the resulting binding modes in accord with their binding affinities. This last requirement is arguably the most difficult to meet due to the complexity of the binding process. A potential soln. to this so-called scoring problem is the usage of multiple scoring functions in an approach known as consensus scoring. Several consensus scoring methods were suggested in the literature and have generally demonstrated an improved ranking of screening libraries relative to individual scoring functions. Nevertheless, current consensus scoring strategies suffer from several shortcomings, in particular, strong dependence on the initial parameters and an incomplete treatment of inactive compds. In this work we present a new consensus scoring algorithm (SeleX-Consensus Scoring abbreviated to SeleX-CS) specifically designed to address these limitations. (i) A subset of the initial set of the scoring functions is allowed to form the consensus score, and this subset is optimized via a Monte Carlo/Simulated Annealing procedure. (ii) Rank redundancy between the members of the screening library is removed. (iii) The method explicitly considers the presence of inactive compds. The new algorithm was applied to the ranking of screening libraries targeting two G-protein coupled receptors (GPCR). Excellent enrichment factors were obtained in both cases: For the cannabinoid receptor 1 (CB1), SeleX-CS outperformed the best single score and afforded an enrichment factor of 41 at 1% of the screening library compared with the best single score value of 15 (GOLD_Fitness). For the chemokine receptor type 2 (CCR2) SeleX-CS afforded an enrichment factor of 72 (again at 1% of the screening library) once more outperforming any single score (enrichment factor of 20 by G_SCORE). Moreover, SeleX-CS demonstrated success rates of 67% (CCR2) and 73% (CB1) when applied to ranking an external test set. In both cases, the new algorithm also afforded good derichment of inactive compds. (i.e., the ability to push inactive compds. to the bottom of the ranked library). The method was then extended to rank a lead optimization series targeting the Kv4.3 potassium ion channel, resulting in a Spearman's correlation coeff., ρ = 0.63 (n = 40), between the SeleX-CS-based rank and the actual pKi values. These results suggest that SeleX-CS is a powerful method for ranking screening libraries in the lead discovery phase and also merits consideration as a lead optimization tool.
    10. 10
      Betzi, S.; Suhre, K.; Chetrit, B.; Guerlesquin, F.; Morelli, X. GFscore: a general nonlinear consensus scoring function for high-throughput docking J. Chem. Inf. Model. 2006, 46, 1704– 1712
      10
      GFscore: A General Nonlinear Consensus Scoring Function for High-Throughput Docking
      Betzi, Stephane; Suhre, Karsten; Chetrit, Bernard; Guerlesquin, Francoise; Morelli, Xavier
      Journal of Chemical Information and Modeling (2006), 46 (4), 1704-1712CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
      Most of the recent published works in the field of docking and scoring protein/ligand complexes have focused on ranking true positives resulting from a Virtual Library Screening (VLS) through the use of a specified or consensus linear scoring function. In this work, the authors present a methodol. to speed up the High Throughput Screening (HTS) process, by allowing focused screens or for hit list triaging when a prohibitively large no. of hits is identified in the primary screen, where the authors have extended the principle of consensus scoring in a nonlinear neural network manner. This led us to introduce a nonlinear Generalist scoring Function, GFscore, which was trained to discriminate true positives from false positives in a data set of diverse chem. compds. This original Generalist scoring Function is a combination of the five scoring functions found in the CScore package from Tripos Inc. GFscore eliminates up to 75% of mols., with a confidence rate of 90%. The final result is a Hit Enrichment in the list of mols. to investigate during a research campaign for biol. active compds. where the remaining 25% of mols. would be sent to in vitro screening expts. GFscore is therefore a powerful tool for the biologist, saving both time and money.
    11. 11
      Bissantz, C.; Folkers, G.; Rognan, D. Protein-based virtual screening of chemical databases. 1. Evaluation of different docking/scoring combinations J. Med. Chem. 2000, 43, 4759– 4767
      11
      Protein-Based Virtual Screening of Chemical Databases. 1. Evaluation of Different Docking/Scoring Combinations
      Bissantz, Caterina; Folkers, Gerd; Rognan, Didier
      Journal of Medicinal Chemistry (2000), 43 (25), 4759-4767CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
      Three different database docking programs (Dock, FlexX, Gold) have been used in combination with seven scoring functions (Chemscore, Dock, FlexX, Fresno, Gold, Pmf, Score) to assess the accuracy of virtual screening methods against two protein targets (thymidine kinase, estrogen receptor) of known three-dimensional structure. For both targets, it was generally possible to discriminate about 7 out of 10 true hits from a random database of 990 ligands. The use of consensus lists common to two or three scoring functions clearly enhances hit rates among the top 5% scorers from 10% (single scoring) to 25-40% (double scoring) and up to 65-70% (triple scoring). However, in all tested cases, no clear relationships could be found between docking and ranking accuracies. Moreover, predicting the abs. binding free energy of true hits was not possible whatever docking accuracy was achieved and scoring function used. As the best docking/consensus scoring combination varies with the selected target and the physicochem. of target-ligand interactions, we propose a two-step protocol for screening large databases: (i) screening of a reduced dataset contg. a few known ligands for deriving the optimal docking/consensus scoring scheme, (ii) applying the latter parameters to the screening of the entire database.
    12. 12
      Charifson, P. S.; Corkery, J. J.; Murcko, M. A.; Walters, W. P. Consensus scoring: A method for obtaining improved hit rates from docking databases of three-dimensional structures into proteins J. Med. Chem. 1999, 42, 5100– 5109
      12
      Consensus Scoring: A Method for Obtaining Improved Hit Rates from Docking Databases of Three-Dimensional Structures into Proteins
      Charifson, Paul S.; Corkery, Joseph J.; Murcko, Mark A.; Walters, W. Patrick
      Journal of Medicinal Chemistry (1999), 42 (25), 5100-5109CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
      The authors present the results of an extensive computational study in which the authors show that combining scoring functions in an intersection-based consensus approach results in an enhancement in the ability to discriminate between active and inactive enzyme inhibitors. This is illustrated in the context of docking collections of three-dimensional structures into three different enzymes of pharmaceutical interest: p38 MAP kinase, inosine monophosphate dehydrogenase, and HIV protease. An anal. of two different docking methods and thirteen scoring functions provides insights into which functions perform well, both singly and in combination. The data shows that consensus scoring further provides a dramatic redn. in the no. of false positives identified by individual scoring functions, thus leading to a significant enhancement in hit-rates.
    13. 13
      Clark, R. D.; Strizhev, A.; Leonard, J. M.; Blake, J. F.; Matthew, J. B. Consensus scoring for ligand/protein interactions J. Mol. Graphics Modell. 2002, 20, 281– 295
      13
      Consensus scoring for ligand/protein interactions
      Clark, Robert D.; Strizhev, Alexander; Leonard, Joseph M.; Blake, James F.; Matthew, James B.
      Journal of Molecular Graphics & Modelling (2002), 20 (4), 281-295CODEN: JMGMFI; ISSN:1093-3263. (Elsevier Science Inc.)
      Several different functions have been put forward for evaluating the energetics of ligand binding to proteins. Those employed in the DOCK, GOLD and FlexX docking programs have been esp. widely used, particularly in connection with virtual high-throughput screening (vHTS) projects. Until recently, such evaluation functions were usually considered only in conjunction with the docking programs that relied on them. In such studies, the evaluation function in question actually fills two distinct roles: it serves as the objective function being optimized (fitness function), but is also the scoring function used to compare the candidate docking configurations generated by the program. We have used descriptions available in the open literature to create free-standing scoring functions based on those used in DOCK and GOLD, and have implemented the more recently formulated PMF [J. Med. Chem. 42 (1999) 791] scoring function as well. The performance of these functions was examd. individually for each of several data sets for which both crystal structures and affinities are available, as was the performance of the FlexX scoring function. Various ways of combining individual scores into a consensus score (CScore) were also considered. The individual and consensus scores were also used to try to pick out configurations most similar to those found in crystal structures from among a set of candidate configurations produced by FlexX docking runs. We find that the reliability and interpretability of results can be improved by combining results from all four functions into a CScore.
    14. 14
      Feher, M. Consensus scoring for protein-ligand interactions Drug Discovery Today 2006, 11, 421– 428
      14
      Consensus scoring for protein-ligand interactions
      Feher, Miklos
      Drug Discovery Today (2006), 11 (9 & 10), 421-428CODEN: DDTOFS; ISSN:1359-6446. (Elsevier)
      A review. This article reviews the application of consensus scoring for cases when the target 3D structure is known. Comparing the performance of different methods is not a trivial task, and it appears that consensus scoring usually substantially improves virtual screening performance, contributing to better enrichments. It also seems to improve - albeit less dramatically - the prediction of bound conformations and poses. The prediction of binding energies is still rather inaccurate and although consensus scoring generally improves these predictions, more development is required before it can be used for this purpose in routine lead optimization.
    15. 15
      Garcia-Sosa, A. T.; Sild, S.; Maran, U. Design of multi-binding-site inhibitors, ligand efficiency, and consensus screening of avian influenza H5N1 wild-type neuraminidase and of the oseltamivir-resistant H274Y variant J. Chem. Inf. Model. 2008, 48, 2074– 2080
      There is no corresponding record for this reference.
    16. 16
      Krovat, E. M.; Langer, T. Impact of scoring functions on enrichment in docking-based virtual screening: an application study on renin inhibitors J. Chem. Inf. Comput. Sci. 2004, 44, 1123– 1129
      16
      Impact of Scoring Functions on Enrichment in Docking-Based Virtual Screening: An Application Study on Renin Inhibitors
      Krovat, Eva M.; Langer, Thierry
      Journal of Chemical Information and Computer Sciences (2004), 44 (3), 1123-1129CODEN: JCISD8; ISSN:0095-2338. (American Chemical Society)
      The docking program LigandFit/Cerius2 has been used to perform shape-based virtual screening of databases against the aspartic protease renin, a target of detd. three-dimensional structure. The protein structure was used in the induced fit binding conformation that occurs when renin is bound to the highly active renin inhibitor (IC50 = 2 nM). The scoring was calcd. using several different scoring functions to get insight into the predictability of the magnitude of binding interactions. A database of 1000 diverse and drug-like compds., comprised of 990 members of a virtual database generated by using the iLib diverse software and 10 known active renin inhibitors, was docked flexibly and scored to det. appropriate scoring functions. All seven scoring functions used (LigScore1, LigScore2, PLP1, PLP2, JAIN, PMF, LUDI) were able to retrieve at least 50% of the active compds. within the first 20% (200 mols.) of the entire test database. A hit rate of 90% in the top 1.4% resulted using the quadruple consensus scoring of LigScore2, PLP1, PLP2, and JAIN. Addnl., a focused database was created with the iLib diverse software and used for the same procedure as the test database. Docking and scoring of the 990 focused compds. and the 10 known actives were performed. A hit rate of 100% in the top 8.4% resulted with use of the triple consensus scoring of PLP1, PLP2, and PMF. As expected, a ranking of the known active compds. within the focused database compared to the test database was obsd. Adequate virtual screening conditions were derived empirically. They can be used for proximate docking and scoring application of compds. with putative renin inhibiting potency.
    17. 17
      Oda, A.; Tsuchida, K.; Takakura, T.; Yamaotsu, N.; Hirono, S. Comparison of consensus scoring strategies for evaluating computational models of protein-ligand complexes J. Chem. Inf. Model. 2006, 46, 380– 391
      17
      Comparison of Consensus Scoring Strategies for Evaluating Computational Models of Protein-Ligand Complexes
      Oda, Akifumi; Tsuchida, Keiichi; Takakura, Tadakazu; Yamaotsu, Noriyuki; Hirono, Shuichi
      Journal of Chemical Information and Modeling (2006), 46 (1), 380-391CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
      Here, the comparisons of performance of nine consensus scoring strategies, in which multiple scoring functions were used simultaneously to evaluate candidate structures for a protein-ligand complex, in combination with nine scoring functions (FlexX score, GOLD score, PMF score, DOCK score, ChemScore, DrugScore, PLP, ScreenScore, and X-Score), were carried out. The systematic naming of consensus scoring strategies was also proposed. The authors' results demonstrate that choosing the most appropriate type of consensus score is essential for model selection in computational docking; although the vote-by-no. strategy was an effective selection method, the no.-by-no. and rank-by-no. strategies were more appropriate when computational tractability was taken into account. By incorporating these consensus scores into the FlexX program, reasonable complex models can be obtained more efficiently than those selected by independent FlexX scores. These strategies might also improve the scoring of other docking programs, and more-effective structure-based drug design should result from these improvements.
    18. 18
      Omigari, K.; Mitomo, D.; Kubota, S.; Nakamura, K.; Fukunishi, Y. A method to enhance the hit ratio by a combination of structure-based drug screening and ligand-based screening Adv. Appl. Bioinf. Chem. 2008, 1, 19– 28
      There is no corresponding record for this reference.
    19. 19
      Paul, N.; Rognan, D. ConsDock: A new program for the consensus analysis of protein-ligand interactions Proteins: Struct., Funct., Bioinf. 2002, 47, 521– 533
      There is no corresponding record for this reference.
    20. 20
      Renner, S.; Derksen, S.; Radestock, S.; Morchen, F. Maximum common binding modes (MCBM): consensus docking scoring using multiple ligand information and interaction fingerprints J. Chem. Inf. Model. 2008, 48, 319– 332
      20
      Maximum Common Binding Modes (MCBM): Consensus Docking Scoring Using Multiple Ligand Information and Interaction Fingerprints
      Renner, Steffen; Derksen, Swetlana; Radestock, Sebastian; Moerchen, Fabian
      Journal of Chemical Information and Modeling (2008), 48 (2), 319-332CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
      Improving the scoring functions for small mol.-protein docking is a highly challenging task in current computational drug design. Here we present a novel consensus scoring concept for the prediction of binding modes for multiple known active ligands. Similar ligands are generally believed to bind to their receptor in a similar fashion. The presumption of our approach was that the true binding modes of similar ligands should be more similar to each other compared to false pos. binding modes. The no. of conserved (consensus) interactions between similar ligands was used as a docking score. Patterns of interactions were modeled using ligand receptor interaction fingerprints. Our approach was evaluated for four different data sets of known cocrystal structures (CDK-2, dihydrofolate reductase, HIV-1 protease, and thrombin). Docking poses were generated with FlexX and rescored by our approach. For comparison the CScore scoring functions from Sybyl were used, and consensus scores were calcd. thereof. Our approach performed better than individual scoring functions and was comparable to consensus scoring. Anal. of the distribution of docking poses by self-organizing maps (SOM) and interaction fingerprints confirmed that clusters of docking poses composed of multiple ligands were preferentially obsd. near the native binding mode. Being conceptually unrelated to commonly used docking scoring functions our approach provides a powerful method to complement and improve computational docking expts.
    21. 21
      Teramoto, R.; Fukunishi, H. Structure-based virtual screening with supervised consensus scoring: evaluation of pose prediction and enrichment factors J. Chem. Inf. Model. 2008, 48, 747– 754
      21
      Structure-Based Virtual Screening with Supervised Consensus Scoring: Evaluation of Pose Prediction and Enrichment Factors
      Teramoto, Reiji; Fukunishi, Hiroaki
      Journal of Chemical Information and Modeling (2008), 48 (4), 747-754CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
      Since the evaluation of ligand conformations is a crucial aspect of structure-based virtual screening, scoring functions play significant roles in it. However, it is known that a scoring function does not always work well for all target proteins. When one cannot know which scoring function works best against a target protein a priori, there is no std. scoring method to know it even if 3D structure of a target protein-ligand complex is available. Therefore, development of the method to achieve high enrichments from given scoring functions and 3D structure of protein-ligand complex is a crucial and challenging task. To address this problem, we applied SCS (supervised consensus scoring), which employs a rough linear correlation between the binding free energy and the root-mean-square deviation (rmsd) of a native ligand conformations and incorporates protein-ligand binding process with docked ligand conformations using supervised learning, to virtual screening. We evaluated both the docking poses and enrichments of SCS and five scoring functions (F-Score, G-Score, D-Score, ChemScore, and PMF) for three different target proteins: thymidine kinase (TK), thrombin (thrombin), and peroxisome proliferator-activated receptor gamma (PPARγ). Our enrichment studies show that SCS is competitive or superior to a best single scoring function at the top ranks of screened database. We found that the enrichments of SCS could be limited by a best scoring function, because SCS is obtained on the basis of the five individual scoring functions. Therefore, it is concluded that SCS works very successfully from our results. Moreover, from docking pose anal., we revealed the connection between enrichment and av. centroid distance of top-scored docking poses. Since SCS requires only one 3D structure of protein-ligand complex, SCS will be useful for identifying new ligands.
    22. 22
      Teramoto, R.; Fukunishi, H. Consensus scoring with feature selection for structure-based virtual screening J. Chem. Inf. Model. 2008, 48, 288– 295
      22
      Consensus Scoring with Feature Selection for Structure-Based Virtual Screening
      Teramoto, Reiji; Fukunishi, Hiroaki
      Journal of Chemical Information and Modeling (2008), 48 (2), 288-295CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
      The evaluation of ligand conformations is a crucial aspect of structure-based virtual screening, and scoring functions play significant roles in it. While consensus scoring (CS) generally improves enrichment by compensating for the deficiencies of each scoring function, the strategy of how individual scoring functions are selected remains a challenging task when few known active compds. are available. To address this problem, the authors propose feature selection-based consensus scoring (FSCS), which performs supervised feature selection with docked native ligand conformations to select complementary scoring functions. The authors evaluated the enrichments of five scoring functions (F-Score, D-Score, PMF, G-Score, and ChemScore), FSCS, and RCS (rank-by-rank consensus scoring) for four different target proteins: acetylcholine esterase (AChE), thrombin, phosphodiesterase 5 (PDE5), and peroxisome proliferator-activated receptor gamma (PPARγ). The results indicated that FSCS was able to select the complementary scoring functions and enhance ligand enrichments and that it outperformed RCS and the individual scoring functions for all target proteins. They also indicated that the performances of the single scoring functions were strongly dependent on the target protein. An esp. favorable result with implications for practical drug screening is that FSCS performs well even if only one 3D structure of the protein-ligand complex is known. Moreover, the authors found that one can infer which scoring functions significantly enrich active compds. by using feature selection before actual docking and that the selected scoring functions are complementary.
    23. 23
      Wang, R.; Wang, S. How does consensus scoring work for virtual library screening? An idealized computer experiment J. Chem. Inf. Comput. Sci. 2001, 41, 1422– 1426
      23
      How Does Consensus Scoring Work for Virtual Library Screening? An Idealized Computer Experiment
      Wang, Renxiao; Wang, Shaomeng
      Journal of Chemical Information and Computer Sciences (2001), 41 (5), 1422-1426CODEN: JCISD8; ISSN:0095-2338. (American Chemical Society)
      It has been reported recently that consensus scoring, which combines multiple scoring functions in binding affinity estn., leads to higher hit-rates in virtual library screening studies. This method seems quite independent to the target receptor, the docking program, or even the scoring functions under investigation. Here we present an idealized computer expt. to explore how consensus scoring works. A hypothetical set of 5000 compds. is used to represent a chem. library under screening. The binding affinities of all its member compds. are assigned by mimicking a real situation. Based on the assumption that the error of a scoring function is a random no. in a normal distribution, the predicted binding affinities were generated by adding such a random no. to the "obsd." binding affinities. The relation between the hit-rates and the no. of scoring functions employed in scoring was then investigated. The performance of several typical ranking strategies for a consensus scoring procedure was also explored. Our results demonstrate that consensus scoring outperforms any single scoring for a simple statistical reason: the mean value of repeated samplings tends to be closer to the true value. Our results also suggest that a moderate no. of scoring functions, three or four, are sufficient for the purpose of consensus scoring. As for the ranking strategy, both the rank-by-no. and the rank-by-rank strategy work more effectively than the rank-by-vote strategy.
    24. 24
      Yang, J. M.; Chen, Y. F.; Shen, T. W.; Kristal, B. S.; Hsu, D. F. Consensus scoring criteria for improving enrichment in virtual screening J. Chem. Inf. Model. 2005, 45, 1134– 1146
      24
      Consensus Scoring Criteria for Improving Enrichment in Virtual Screening
      Yang, Jinn-Moon; Chen, Yen-Fu; Shen, Tsai-Wei; Kristal, Bruce S.; Hsu, D. Frank
      Journal of Chemical Information and Modeling (2005), 45 (4), 1134-1146CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
      Motivation: Virtual screening of mol. compd. libraries is a potentially powerful and inexpensive method for the discovery of novel lead compds. for drug development. The major weakness of virtual screening-the inability to consistently identify true positives (leads)-is likely due to our incomplete understanding of the chem. involved in ligand binding and the subsequently imprecise scoring algorithms. It has been demonstrated that combining multiple scoring functions (consensus scoring) improves the enrichment of true positives. Previous efforts at consensus scoring have largely focused on empirical results, but they have yet to provide a theor. anal. that gives insight into real features of combinations and data fusion for virtual screening. Results: The authors demonstrate that combining multiple scoring functions improves the enrichment of true positives only if (a) each of the individual scoring functions has relatively high performance and (b) the individual scoring functions are distinctive. Notably, these two prediction variables are previously established criteria for the performance of data fusion approaches using either rank or score combinations. This work, thus, establishes a potential theor. basis for the probable success of data fusion approaches to improve yields in in silico screening expts. Furthermore, it is similarly established that the second criterion (b) can, in at least some cases, be functionally defined as the area between the rank vs. score plots generated by the two (or more) algorithms. Because rank-score plots are independent of the performance of the individual scoring function, this establishes a second theor. defined approach to detg. the likely success of combining data from different predictive algorithms. This approach is, thus, useful in practical settings in the virtual screening process when the performance of at least two individual scoring functions (such as in criterion a) can be estd. as having a high likelihood of having high performance, even if no training sets are available. The authors provide initial validation of this theor. approach using data from five scoring systems with two evolutionary docking algorithms on four targets, thymidine kinase, human dihydrofolate reductase, and estrogen receptors of antagonists and agonists. Our procedure is computationally efficient, able to adapt to different situations, and scalable to a large no. of compds. as well as to a greater no. of combinations. Results of the expt. show a fairly significant improvement (vs. single algorithms) in several measures of scoring quality, specifically "goodness-of-hit" scores, false pos. rates, and "enrichment". This approach (available online at http://gemdock.life. nctu.edu.tw/dock/download.php) has practical utility for cases where the basic tools are known or believed to be generally applicable, but where specific training sets are absent.
    25. 25
      Hogg, R. V.; Tanis, E. A. Probability and Statistical Inference; Prentice Hall College Division: Englewood Cliffs, NJ, 2001, pp 402– 411.
      There is no corresponding record for this reference.
    26. 26
      Books of Abstracts; 240th American Chemical Society National Meeting, Boston, MA, August 22–28, 2010; ACS: Washington, D.C., 2010.
      There is no corresponding record for this reference.
    27. 27
      Benson, M. L.; Smith, R. D.; Khazanov, N. A.; Dimcheff, B.; Beaver, J.; Dresslar, P.; Nerothin, J.; Carlson, H. A. Binding MOAD, a high-quality protein-ligand database Nucleic Acids Res. 2008, 36, D674– D678
      There is no corresponding record for this reference.
    28. 28
      Hu, L.; Benson, M. L.; Smith, R. D.; Lerner, M. G.; Carlson, H. A. Binding MOAD (Mother Of All Databases) Proteins: Struct., Funct., Bioinf. 2005, 60, 333– 340
      28
      Binding MOAD (Mother of All Databases)
      Hu, Liegi; Benson, Mark L.; Smith, Richard D.; Lerner, Michael G.; Carlson, Heather A.
      Proteins: Structure, Function, and Bioinformatics (2005), 60 (3), 333-340CODEN: PSFBAF ISSN:. (Wiley-Liss, Inc.)
      Binding MOAD (Mother of All Databases) is the largest collection of high-quality, protein-ligand complexes available from the Protein Data Bank. At this time, Binding MOAD contains 5331 protein-ligand complexes comprised of 1780 unique protein families and 2630 unique ligands. We have searched the crystallog. papers for all 5000 + structures and compiled binding data for 1375 (26%) of the protein-ligand complexes. The binding-affinity data ranges 13 orders of magnitude. This is the largest collection of binding data reported to date in the literature. We have also addressed the issue of redundancy in the data. To create a nonredundant dataset, one protein from each of the 1780 protein families was chosen as a representative. Representatives were chosen by tightest binding, best resoln., etc. For the 1780 "best" complexes that comprise the nonredundant version of Binding MOAD, 475 (27%) have binding data. This significant collection of protein-ligand complexes will be very useful in elucidating the biophys. patterns of mol. recognition and enzymic regulation. The complexes with binding-affinity data will help in the development of improved scoring functions and structure-based drug discovery techniques. The dataset can be accessed at http://www.BindingMOAD.org.
    29. 29
      Wang, R.; Fang, X.; Lu, Y.; Wang, S. The PDBbind database: collection of binding affinities for protein-ligand complexes with known three-dimensional structures J. Med. Chem. 2004, 47, 2977– 2980
      29
      The PDBbind database: Collection of binding affinities for protein-ligand complexes with known three-dimensional structures
      Wang, Renxiao; Fang, Xueliang; Lu, Yipin; Wang, Shaomeng
      Journal of Medicinal Chemistry (2004), 47 (12), 2977-2980CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
      We have screened the entire Protein Data Bank (Release No. 103, Jan. 2003) and identified 5671 protein-ligand complexes out of 19 621 exptl. structures. A systematic examn. of the primary refs. of these entries has led to a collection of binding affinity data (Kd, Ki, and IC50) for a total of 1359 complexes. The outcomes of this project have been organized into a Web-accessible database named the PDBbind database.
    30. 30
      Wang, R.; Fang, X.; Lu, Y.; Yang, C.-Y.; Wang, S. The PDBbind database: methodologies and updates J. Med. Chem. 2005, 48, 4111– 4119
      30
      The PDBbind Database: Methodologies and Updates
      Wang, Renxiao; Fang, Xueliang; Lu, Yipin; Yang, Chao-Yie; Wang, Shaomeng
      Journal of Medicinal Chemistry (2005), 48 (12), 4111-4119CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
      The authors have developed the PDBbind database to provide a comprehensive collection of binding affinities for the protein-ligand complexes in the Protein Data Bank (PDB). This paper gives a full description of the latest version, i.e., version 2003, which is an update to our recently reported work. Out of 23 790 entries in the PDB release No.107 (Jan. 2004), 5897 entries were identified as protein-ligand complexes that meet our definition. Exptl. detd. binding affinities (Kd, Ki, and IC50) for 1622 of these were retrieved from the refs. assocd. with these complexes. A total of 900 complexes were selected to form a "refined set", which is of particular value as a std. data set for docking and scoring studies. All of the final data, including binding affinity data, ref. citations, and processed structural files, have been incorporated into the PDBbind database accessible online at http:// www.pdbbind.org/.
    31. 31
      Morris, G. M.; Goodsell, D. S.; Huey, R.; Olson, A. J. Distributed automated docking of flexible ligands to proteins: parallel applications of AutoDock 2.4 J. Comput.-Aided Mol. Des. 1996, 10, 293– 304
      31
      Distributed automated docking of flexible ligands to proteins: parallel applications of AutoDock 2.4
      Morris, Garrett M.; Goodsell, David S.; Huey, Ruth; Olson, Arthur J.
      Journal of Computer-Aided Molecular Design (1996), 10 (4), 293-304CODEN: JCADEQ; ISSN:0920-654X. (ESCOM)
      AutoDock 2.4 predicts the bound conformations of a small, flexible ligand to a nonflexible macromol. target of known structure. The technique combines simulated annealing for conformation searching with a rapid grid-based method of energy evaluation based on the AMBER force field. AutoDock has been optimized in performance without sacrificing accuracy; it incorporates many enhancements and addns., including an intuitive interface. We have developed a set of tools for launching and analyzing many independent docking jobs in parallel on a heterogeneous network of UNIX-based workstations. This paper describes the current release, and the results of a suite of diverse test systems. We also present the results of a systematic investigation in to the effects of varying simulated-annealing parameters on mol. docking. We show that even for ligands with a large no. of degrees of freedom, root-mean-square deviations of less than 1 Å from the crystallog. conformation are obtained for the lowest-energy dockings, although fewer dockings find the crystallog. conformation when there are more degrees of freedom.
    32. 32
      Trott, O.; Olson, A. J. Software News and Update AutoDock Vina: Improving the Speed and Accuracy of Docking with a New Scoring Function, Efficient Optimization and Multithreading J. Comput. Chem. 2010, 31, 455– 461
      There is no corresponding record for this reference.
    33. 33
      Shoichet, B. K.; Bodian, D. L.; Kuntz, I. D. Molecular Docking Using Shape Descriptors J. Comput. Chem. 1992, 13, 380– 397
      33
      Molecular docking using shape descriptors
      Shoichet, Brian K.; Bodian, Dale L.; Kuntz, Irwin D.
      Journal of Computational Chemistry (1992), 13 (3), 380-97CODEN: JCCHDD; ISSN:0192-8651.
      Mol. docking explores the binding modes of two interacting mols. The technique is increasingly popular for studying protein-ligand interactions and for drug design. A fundamental problem with mol. docking is that orientation space is very large and grows combinatorially with the no. of degrees of freedom of the interacting mols. Here, algorithms are described and evaluated that improve the efficiency and accuracy of a shape-based docking method. Mol. organization and sampling techniques are used to remove the exponential time dependence on mol. size in docking calcns. The new techniques allow one to study systems that were prohibitively large for the original method. The new algorithms are tested in 10 different protein-ligand systems, including systems, including 7 systems where the ligand is itself a protein. In all cases, the new algorithms successfully reproduce the exptl. detd. configurations of the ligand in the protein.
    34. 34
      Gohlke, H.; Hendlich, M.; Klebe, G. Knowledge-based scoring function to predict protein-ligand interactions J. Mol. Biol. 2000, 295, 337– 356
      34
      Knowledge-based Scoring Function to Predict Protein-Ligand Interactions
      Gohlke, Holger; Hendlich, Manfred; Klebe, Gerhard
      Journal of Molecular Biology (2000), 295 (2), 337-356CODEN: JMOBAK; ISSN:0022-2836. (Academic Press)
      The development and validation of a new knowledge-based scoring function (DrugScore) to describe the binding geometry of ligands in proteins is presented. It discriminates efficiently between well-docked ligand binding modes (root-mean-square deviation <2.0 Å with respect to a crystallog. detd. ref. complex) and those largely deviating from the native structure, e.g. generated by computer docking programs. Structural information is extd. from crystallog. detd. protein-ligand complexes using ReLiBase and converted into distance-dependent pair-preferences and solvent-accessible surface (SAS) dependent singlet preferences for protein and ligand atoms. Definition of an appropriate ref. state and accounting for inaccuracies inherently present in exptl. data is required to achieve good predictive power. The sum of the pair preferences and the singlet preferences is calcd. based on the 3D structure of protein-ligand binding modes generated by docking tools. For two test sets of 91 and 68 protein-ligand complexes, taken from the Protein Data Bank (PDB), the calcd. score recognizes poses generated by FlexX deviating <2 Å from the crystal structure on rank 1 in three quarters of all possible cases. Compared to FlexX, this is a substantial improvement. For ligand geometries generated by DOCK, DrugScore is superior to the "chem. scoring" implemented into this tool, while comparable results are obtained using the "energy scoring" in DOCK. None of the presently known scoring functions achieves comparable power to ext. binding modes in agreement with expt. It is fast to compute, regards implicitly solvation and entropy contributions and produces correctly the geometry of directional interactions. Small deviations in the 3D structure are tolerated and, since only contacts to non-hydrogen atoms are regarded, it is independent from assumptions of protonation states. (c) 2000 Academic Press.
    35. 35
      Zsoldos, Z.; Reid, D.; Simon, A.; Sadjad, S. B.; Johnson, A. P. eHiTS: a new fast, exhaustive flexible ligand docking system J. Mol. Graph. Model. 2007, 26, 198– 212
      35
      eHiTS: A new fast, exhaustive flexible ligand docking system
      Zsoldos, Zsolt; Reid, Darryl; Simon, Aniko; Sadjad, Sayyed Bashir; Johnson, A. Peter
      Journal of Molecular Graphics & Modelling (2007), 26 (1), 198-212CODEN: JMGMFI; ISSN:1093-3263. (Elsevier Inc.)
      The flexible ligand docking problem is divided into two subproblems: pose/conformation search and scoring function. For successful virtual screening the search algorithm must be fast and able to find the optimal binding pose and conformation of the ligand. Statistical anal. of exptl. data of bound ligand conformations is presented with conclusions about the sampling requirements for docking algorithms. EHiTS is an exhaustive flexible-docking method that systematically covers the part of the conformational and positional search space that avoids severe steric clashes, producing highly accurate docking poses at a speed practical for virtual high-throughput screening. The customizable scoring function of eHiTS combines novel terms (based on local surface point contact evaluation) with traditional empirical and statistical approaches. Validation results of eHiTS are presented and compared to three other docking software on a set of 91 PDB structures that are common to the validation sets published for the other programs.
    36. 36
      FRED; version 2.2.5; OpenEye Scientific Software, Inc.: Santa FRED, NM 87508, 2009.
      There is no corresponding record for this reference.
    37. 37
      Friesner, R. A.; Banks, J. L.; Murphy, R. B.; Halgren, T. A.; Klicic, J. J.; Mainz, D. T.; Repasky, M. P.; Knoll, E. H.; Shelley, M.; Perry, J. K.; Shaw, D. E.; Francis, P.; Shenkin, P. S. Glide: a new approach for rapid, accurate docking and scoring. 1. Method and assessment of docking accuracy J. Med. Chem. 2004, 47, 1739– 1749
      37
      Glide: A new approach for rapid, accurate docking and scoring. 1. method and assessment of docking accuracy
      Friesner, Richard A.; Banks, Jay L.; Murphy, Robert B.; Halgren, Thomas A.; Klicic, Jasna J.; Mainz, Daniel T.; Repasky, Matthew P.; Knoll, Eric H.; Shelley, Mee; Perry, Jason K.; Shaw, David E.; Francis, Perry; Shenkin, Peter S.
      Journal of Medicinal Chemistry (2004), 47 (7), 1739-1749CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
      A review. Unlike other methods for docking ligands to the rigid 3D structure of a known protein receptor, Glide approximates a complete systematic search of the conformational, orientational, and positional space of the docked ligand. In this search, an initial rough positioning and scoring phase that dramatically narrows the search space is followed by torsionally flexible energy optimization on an OPLS-AA nonbonded potential grid for a few hundred surviving candidate poses. The very best candidates are further refined via a Monte Carlo sampling of pose conformation; in some cases, this is crucial to obtaining an accurate docked pose. Selection of the best docked pose uses a model energy function that combines empirical and force-field-based terms. Docking accuracy is assessed by redocking ligands from 282 cocrystd. PDB complexes starting from conformationally optimized ligand geometries that bear no memory of the correctly docked pose. Errors in geometry for the top-ranked pose are less than 1 Å in nearly half of the cases and are greater than 2 Å in only about one-third of them. Comparisons to published data on rms deviations show that Glide is nearly twice as accurate as GOLD and more than twice as accurate as FlexX for ligands having up to 20 rotatable bonds. Glide is also found to be more accurate than the recently described Surflex method.
    38. 38
      Verdonk, M. L.; Cole, J. C.; Hartshorn, M. J.; Murray, C. W.; Taylor, R. D. Improved protein-ligand docking using GOLD Proteins: Struct., Funct., Bioinf. 2003, 52, 609– 623
      There is no corresponding record for this reference.
    39. 39
      Kramer, C.; Gedeck, P. Global free energy scoring functions based on distance-dependent atom-type pair descriptors J.Chem. Inf. Model. 2011, 51, 707– 720
      39
      Global free energy scoring functions based on distance-dependent atom-type pair descriptors
      Kramer, Christian; Gedeck, Peter
      Journal of Chemical Information and Modeling (2011), 51 (3), 707-720CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
      Scoring functions for protein-ligand docking have received much attention in the past two decades. In many cases, remarkable success has been demonstrated in predicting the correct geometry of interaction. On independent test sets, however, the predicted binding energies or scores correlate only slightly with the obsd. free energies of binding. In this study, we analyze how well free energies of binding can be predicted on the basis of crystal structures using traditional QSAR techniques in a proteochemometric approach. We introduce a new set of protein-ligand interaction descriptors on the basis of distance-binned Crippen-like atom type pairs. A subset of the publicly available PDBbind09-CN refined set (MW < 900 g/mol, #P < 2, ndon + nacc < 20; N = 1387) is being used as data set. It is demonstrated how simple, yet surprisingly good, scoring functions can be generated for the whole diverse database (R2out-of-bag = 0.48, Rp = 0.69, RMSE = 1.44, MUE = 1.14) and individual protein family subsets. This performance is significantly better than the performance of almost all other scoring functions published that have been validated on a test set as large and diverse as the PDBbind refined set. We also find that on some protein families surprisingly good scoring functions can be obtained using simple ligand-only descriptors like logS, logP, and mol. wt. The ligand-descriptor based scoring function equals or even outperforms commonly used scoring functions, highlighting the need for better scoring functions. We demonstrate how the obsd. performance depends on the validation strategy, and we outline a general validation protocol for future free energy scoring functions.
    40. 40
      Huang, S.-Y.; Zou, X. An iterative knowledge-based scoring function to predict protein-ligand interactions: I. Derivation of interaction potentials J. Comput. Chem. 2006, 27, 1866– 1875
      40
      An iterative knowledge-based scoring function to predict protein-ligand interactions: I. Derivation of interaction potentials
      Huang, Sheng-You; Zou, Xiaoqin
      Journal of Computational Chemistry (2006), 27 (15), 1866-1875CODEN: JCCHDD; ISSN:0192-8651. (John Wiley & Sons, Inc.)
      Using a novel iterative method, the authors have developed a knowledge-based scoring function (ITScore) to predict protein-ligand interactions. The pair potentials for ITScore were derived from a training set of 786 protein-ligand complex structures in the Protein Data Bank. Twenty-six atom types were used based on the atom type category of the SYBYL software. The iterative method circumvents the long-standing ref. state problem in the derivation of knowledge-based scoring functions. The basic idea is to improve pair potentials by iteration until they correctly discriminate exptl. detd. binding modes from decoy ligand poses for the ligand-protein complexes in the training set. The iterative method is efficient and normally converges within 20 iterative steps. The scoring function based on the derived potentials was tested on a diverse set of 140 protein-ligand complexes for affinity prediction, yielding a high correlation coeff. of 0.74. Because ITScore uses SYBYL-defined atom types, this scoring function is easy to use for mol. files prepd. by SYBYL or converted by software such as BABEL.
    41. 41
      Stroganov, O. V.; Novikov, F. N.; Stroylov, V. S.; Kulkov, V.; Chilov, G. G. Lead finder: an approach to improve accuracy of protein-ligand docking, binding energy estimation, and virtual screening J. Chem. Inf. Model. 2008, 48, 2371– 2385
      41
      Lead Finder: An Approach To Improve Accuracy of Protein-Ligand Docking, Binding Energy Estimation, and Virtual Screening
      Stroganov, Oleg V.; Novikov, Fedor N.; Stroylov, Viktor S.; Kulkov, Val; Chilov, Ghermes G.
      Journal of Chemical Information and Modeling (2008), 48 (12), 2371-2385CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
      An innovative mol. docking algorithm and three specialized high accuracy scoring functions are introduced in the Lead Finder docking software. Lead Finder's algorithm for ligand docking combines the classical genetic algorithm with various local optimization procedures and resourceful exploitation of the knowledge generated during docking process. Lead Finder's scoring functions are based on a mol. mechanics functional which explicitly accounts for different types of energy contributions scaled with empiric coeffs. to produce three scoring functions tailored for (a) accurate binding energy predictions; (b) correct energy-ranking of docked ligand poses; and (c) correct rank-ordering of active and inactive compds. in virtual screening expts. The predicted values of the free energy of protein-ligand binding were benchmarked against a set of exptl. measured binding energies for 330 diverse protein-ligand complexes yielding rmsd of 1.50 kcal/mol. The accuracy of ligand docking was assessed on a set of 407 structures, which included almost all published test sets of the following programs: FlexX, Glide SP, Glide XP, Gold, LigandFit, MolDock, and Surflex. Rmsd of 2 Å or less was obsd. for 80-96% of the structures in the test sets (80.0% on the Glide XP and FlexX test sets, 96.0% on the Surflex and MolDock test sets). The ability of Lead Finder to distinguish between active and inactive compds. during virtual screening expts. was benchmarked against 34 therapeutically relevant protein targets. Impressive enrichment factors were obtained for almost all of the targets with the av. area under receiver operator curve being equal to 0.92.
    42. 42
      Build_model; version 2.0.1 build 07.30; MolTech Ltd.: 2008–2011.
      There is no corresponding record for this reference.
    43. 43
      Yin, S.; Biedermannova, L.; Vondrasek, J.; Dokholyan, N. V. MedusaScore: an accurate force field-based scoring function for virtual drug screening J. Chem. Inf. Model. 2008, 48, 1656– 1662
      43
      MedusaScore: An Accurate Force Field-Based Scoring Function for Virtual Drug Screening
      Yin, Shuangye; Biedermannova, Lada; Vondrasek, Jiri; Dokholyan, Nikolay V.
      Journal of Chemical Information and Modeling (2008), 48 (8), 1656-1662CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
      Virtual screening is becoming an important tool for drug discovery. However, the application of virtual screening has been limited by the lack of accurate scoring functions. Here, we present a novel scoring function, MedusaScore, for evaluating protein-ligand binding. MedusaScore is based on models of phys. interactions that include van der Waals, solvation, and hydrogen bonding energies. To ensure the best transferability of the scoring function, we do not use any protein-ligand exptl. data for parameter training. We then test the MedusaScore for docking decoy recognition and binding affinity prediction and find superior performance compared to other widely used scoring functions. Statistical anal. indicates that one source of inaccuracy of MedusaScore may arise from the unaccounted entropic loss upon ligand binding, which suggests avenues of approach for further MedusaScore improvement.
    44. 44
      Molecular Operating Environment (MOE); version 2010.10; Chemical Computing Group: Montreal, C.N., 2010.
      There is no corresponding record for this reference.
    45. 45
      Goto, J.; Kataoka, R.; Muta, H.; Hirayama, N. ASEDock-docking based on alpha spheres and excluded volumes J. Chem. Inf. Model. 2008, 48, 583– 590
      45
      ASEDock - docking based on alpha spheres and excluded volumes
      Goto, Junichi; Kataoka, Ryoichi; Muta, Hajime; Hirayama, Noriaki
      Journal of Chemical Information and Modeling (2008), 48 (3), 583-590CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
      ASEDock is a novel docking program based on a shape similarity assessment between a concave portion (i.e., concavity) on a protein and the ligand. We have introduced two novel concepts into ASEDock. One is an ASE model, which is defined by the combination of alpha spheres generated at a concavity in a protein and the excluded vols. around the concavity. The other is an ASE score, which evaluates the shape similarity between the ligand and the ASE model. The ASE score selects and refines the initial pose by maximizing the overlap between the alpha spheres and the ligand, and minimizing the overlap between the excluded vol. and the ligand. Because the ASE score makes good use of the Gaussian-type function for evaluating and optimizing the overlap between the ligand and the site model, it can pose a ligand onto the docking site relatively faster and more effectively than using potential energy functions. The posing stage through the use of the ASE score is followed by full atomistic energy minimization. Because the posing algorithm of ASEDock is free from any bias except for shape, it is a very robust docking method. A validation study using 59 high-quality X-ray structures of the complexes between drug-like mols. and the target proteins has demonstrated that ASEDock can faithfully reproduce exptl. detd. docking modes of various druglike mols. in their target proteins. Almost 80% of the structures were reconstructed within the estd. exptl. error. The success rate of ∼98% was attained based on the docking criterion of the root-mean-square deviation (RMSD) of non-hydrogen atoms (≤2.0 Å). The markedly high success of ASEDock in redocking expts. clearly indicates that the most important factor governing the docking process is shape complementarity.
    46. 46
      Yang, C.-Y.; Wang, R.; Wang, S. M-score: a knowledge-based potential scoring function accounting for protein atom mobility J. Med. Chem. 2006, 49, 5903– 5911
      46
      M-Score: A Knowledge-Based Potential Scoring Function Accounting for Protein Atom Mobility
      Yang, Chao-Yie; Wang, Renxiao; Wang, Shaomeng
      Journal of Medicinal Chemistry (2006), 49 (20), 5903-5911CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
      A knowledge-based potential scoring function, named M-Score, has been developed based upon 2331 high-resoln. crystal structures of protein-ligand complexes. M-Score considers the mobility of protein atoms, describing the location of each protein atom by a Gaussian distribution instead of a fixed position based upon the isotropic B-factors. This leads to an increase in the no. of atom-pairs in the construction of knowledge-based potentials and a smoothing effect on the pairwise distribution functions. M-Score was validated using 896 complexes which were not included in the 2331 data set and whose exptl. detd. binding affinities were available. The overall linear correlation coeff. (r) between the calcd. scores and exptl. detd. binding affinities (pKi or pKd) for these 896 complexes is -0.49. Evaluation of M-Score against 17 protein families showed that we obtained good to excellent correlations for six protein families, modest correlations for four protein families, and poor correlations for the remaining seven protein families.
    47. 47
      Rahaman, O.; Estrada, T.; Doran, D.; Taufer, M.; Brooks, C.; Armen, R. Evaluation of Several Two-step Scoring Functions Based on Linear Interaction Energy, Effective Ligand Size, and Empirical Pair Potentials for Prediction of Protein-Ligand Binding Geometry and Free Energy J. Chem. Inf. Model. 2011,  DOI: 10.1021/ci1003009
      There is no corresponding record for this reference.
    48. 48
      Naim, M.; Bhat, S.; Rankin, K. N.; Dennis, S.; Chowdhury, S. F.; Siddiqi, I.; Drabik, P.; Sulea, T.; Bayly, C. I.; Jakalian, A.; Purisima, E. O. Solvated interaction energy (SIE) for scoring protein-ligand binding affinities. 1. Exploring the parameter space J. Chem. Inf. Model. 2007, 47, 122– 133
      48
      Solvated interaction energy (SIE) for scoring protein-ligand binding affinities. 1. Exploring the parameter space
      Naim Marwen; Bhat Sathesh; Rankin Kathryn N; Dennis Sheldon; Chowdhury Shafinaz F; Siddiqi Imran; Drabik Piotr; Sulea Traian; Bayly Christopher I; Jakalian Araz; Purisima Enrico O
      Journal of chemical information and modeling (2007), 47 (1), 122-33 ISSN:1549-9596.
      We present a binding free energy function that consists of force field terms supplemented by solvation terms. We used this function to calibrate the solvation model along with the binding interaction terms in a self-consistent manner. The motivation for this approach was that the solute dielectric-constant dependence of calculated hydration gas-to-water transfer free energies is markedly different from that of binding free energies (J. Comput. Chem. 2003, 24, 954). Hence, we sought to calibrate directly the solvation terms in the context of a binding calculation. The five parameters of the model were systematically scanned to best reproduce the absolute binding free energies for a set of 99 protein-ligand complexes. We obtained a mean unsigned error of 1.29 kcal/mol for the predicted absolute binding affinity in a parameter space that was fairly shallow near the optimum. The lowest errors were obtained with solute dielectric values of Din = 20 or higher and scaling of the intermolecular van der Waals interaction energy by factors ranging from 0.03 to 0.15. The high apparent Din and strong van der Waals scaling may reflect the anticorrelation of the change in solvated potential energy and configurational entropy, that is, enthalpy-entropy compensation in ligand binding (Biophys. J. 2004, 87, 3035-3049). Five variations of preparing the protein-ligand data set were explored in order to examine the effect of energy refinement and the presence of bound water on the calculated results. We find that retaining water in the final protein structure used for calculating the binding free energy is not necessary to obtain good results; that is the continuum solvation model is sufficient. Virtual screening enrichment studies on estrogen receptor and thymidine kinase showed a good ability of the binding free energy function to recover true hits in a collection of decoys.
    49. 49
      Wang, R.; Lai, L.; Wang, S. Further development and validation of empirical scoring functions for structure-based binding affinity prediction J. Comput. Aid. Mol. Des. 2002, 16, 11– 26
      49
      Further development and validation of empirical scoring functions for structure-based binding affinity prediction
      Wang, Renxiao; Lai, Luhua; Wang, Shaomeng
      Journal of Computer-Aided Molecular Design (2002), 16 (1), 11-26CODEN: JCADEQ; ISSN:0920-654X. (Kluwer Academic Publishers)
      New empirical scoring functions have been developed to est. the binding affinity of a given protein-ligand complex with known three-dimensional structure. These scoring functions include terms accounting for van der Waals interaction, hydrogen bonding, deformation penalty, and hydrophobic effect. A special feature is that three different algorithms have been implemented to calc. the hydrophobic effect term, which results in three parallel scoring functions. All three scoring functions are calibrated through multivariate regression anal. of a set of 200 protein-ligand complexes and they reproduce the binding free energies of the entire training set with std. deviations of 2.2 kcal/mol, 2.1 kcal/mol, and 2.0 kcal/mol, resp. These three scoring functions are further combined into a consensus scoring function, X-CSCORE. When tested on an independent set of 30 protein-ligand complexes, X-CSCORE is able to predict their binding free energies with a std. deviation of 2.2 kcal/mol. The potential application of X-CSCORE to mol. docking is also investigated. Our results show that this consensus scoring function improves the docking accuracy considerably when compared to the conventional force field computation used for mol. docking.
    50. 50
      R: A Language and Environment for Statistical Computing; Team, R.; D. C.; version 2.9.2; R Project for Statistical Computing: Vienna, Austria, 2009.
      There is no corresponding record for this reference.
    51. 51
      Bonett, D. G.; Wright, T. A. Sample Size Requirements for Estimating Pearson, Kendall and Spearman Correlatons Psychometrika 2000, 65, 23– 28
      There is no corresponding record for this reference.
    52. 52
      JMP; version 9.0.0; SAS institute Inc.: Cary, N.C.: 2010.
      There is no corresponding record for this reference.
    53. 53
      Davis, I. W.; Leaver-Fay, A.; Chen, V. B.; Block, J. N.; Kapral, G. J.; Wang, X.; Murray, L. W.; Arendall, W. B., 3rd; Snoeyink, J.; Richardson, J. S.; Richardson, D. C. MolProbity: all-atom contacts and structure validation for proteins and nucleic acids Nucleic Acids Res. 2007, 35, W375– W383
      53
      MolProbity: all-atom contacts and structure validation for proteins and nucleic acids
      Davis Ian W; Leaver-Fay Andrew; Chen Vincent B; Block Jeremy N; Kapral Gary J; Wang Xueyi; Murray Laura W; Arendall W Bryan 3rd; Snoeyink Jack; Richardson Jane S; Richardson David C
      Nucleic acids research (2007), 35 (Web Server issue), W375-83 ISSN:.
      MolProbity is a general-purpose web server offering quality validation for 3D structures of proteins, nucleic acids and complexes. It provides detailed all-atom contact analysis of any steric problems within the molecules as well as updated dihedral-angle diagnostics, and it can calculate and display the H-bond and van der Waals contacts in the interfaces between components. An integral step in the process is the addition and full optimization of all hydrogen atoms, both polar and nonpolar. New analysis functions have been added for RNA, for interfaces, and for NMR ensembles. Additionally, both the web site and major component programs have been rewritten to improve speed, convenience, clarity and integration with other resources. MolProbity results are reported in multiple forms: as overall numeric scores, as lists or charts of local problems, as downloadable PDB and graphics files, and most notably as informative, manipulable 3D kinemage graphics shown online in the KiNG viewer. This service is available free to all users at http://molprobity.biochem.duke.edu.
    54. 54
      Vriend, G. WHAT IF: a molecular modeling and drug design program J. Mol. Graph. 1990, 8, 52– 56
      54
      WHAT IF: a molecular modeling and drug design program
      Vriend, G.
      Journal of Molecular Graphics (1990), 8 (1), 52-6, 29CODEN: JMGRDV; ISSN:0263-7855.
      A FORTRAN 77 computer program has been written to aid with macromol. modeling and drug design. Called WHAT IF, it provides an intelligent and flexible environment for displaying, manipulating, and analyzing small mols., proteins, nucleic acids, and their interactions. A relational protein structure database is incorporated to be queried. The program is suitable for most common crystallog. work. The menu-driven operation of WHAT IF, combined with the use of default values wherever user input is required, makes it very easy to use for a novice user while keeping full flexibility for more sophisticated studies. Although there are not too many unique features in WHAT IF, the fact that everything is integrated in one program makes if a unique tool for many purposes.
    55. 55
      Cruickshank, D. W. Remarks about protein structure precision Acta Crystallogr. D 1999, 55, 583– 601
      55
      Remarks about protein structure precision
      Cruickshank D W
      Acta crystallographica. Section D, Biological crystallography (1999), 55 (Pt 3), 583-601 ISSN:0907-4449.
      Full-matrix least squares is taken as the basis for an examination of protein structure precision. A two-atom protein model is used to compare the precisions of unrestrained and restrained refinements. In this model, restrained refinement determines a bond length which is the weighted mean of the unrestrained diffraction-only length and the geometric dictionary length. Data of 0.94 A resolution for the 237-residue protein concanavalin A are used in unrestrained and restrained full-matrix inversions to provide standard uncertainties sigma(r) for positions and sigma(l) for bond lengths. sigma(r) is as small as 0.01 A for atoms with low Debye B values but increases strongly with B. The results emphasize the distinction between unrestrained and restrained refinements and between sigma(r) and sigma(l). Other full-matrix inversions are reported. Such inversions require massive calculations. Several approximate methods are examined and compared critically. These include a Fourier map formula [Cruickshank (1949). Acta Cryst. 2, 65-82], Luzzati plots [Luzzati (1952). Acta Cryst. 5, 802-810] and a new diffraction-component precision index (DPI). The DPI estimate of sigma(r, Bavg) is given by a simple formula. It uses R or Rfree and is based on a very rough approximation to the least-squares method. Many examples show its usefulness as a precision comparator for high- and low-resolution structures. The effect of restraints as resolution varies is examined. More regular use of full-matrix inversion is urged to establish positional precision and hence the precision of non-dictionary distances in both high- and low-resolution structures. Failing this, parameter blocks for representative residues and their neighbours should be inverted to gain a general idea of sigma(r) as a function of B. The whole discussion is subject to some caveats about the effects of disordered regions in the crystal.
    56. 56
      Smith, R. D.; Hu, L.; Falkner, J. A.; Benson, M. L.; Nerothin, J. P.; Carlson, H. A. Exploring protein-ligand recognition with Binding MOAD J. Mol. Graphics Modell. 2006, 24, 414– 425
      56
      Exploring protein-ligand recognition with Binding MOAD
      Smith, Richard D.; Hu, Liegi; Falkner, Jayson A.; Benson, Mark L.; Nerothin, Jason P.; Carlson, Heather A.
      Journal of Molecular Graphics & Modelling (2006), 24 (6), 414-425CODEN: JMGMFI; ISSN:1093-3263. (Elsevier Inc.)
      We have recently announced the largest database of protein-ligand complexes, Binding MOAD (Mother of All Databases). After the August 2004 update, Binding MOAD contains 6816 complexes. There are 2220 protein families and 3316 unique ligands. After searching 6000+ crystallog. papers, we have obtained binding data for 1793 (27%) of the complexes. We have also created a non-redundant set of complexes with only one complex from each protein family; in that set, 630 (28%) of the unique complexes have binding data. Here, we present information about the data provided at the Binding MOAD website. We also present the results of mining Binding MOAD to map the degree of solvent exposure for binding sites. We have detd. that most cavities and ligands (70-85%) are well buried in the complexes. This fits with the common paradigm that a large degree of contact between the ligand and protein is significant in mol. recognition. GoCAV and the GoCAVviewer are the tools we created for this study. To share our data and make our online dataset more useful to other research groups, we have integrated the viewer into the Binding MOAD website (www.BindingMOAD.org).
    57. 57
      Wallin, R.; Hutson, S. M. Warfarin and the vitamin K-dependent gamma-carboxylation system Trends Mol. Med. 2004, 10, 299– 302
      There is no corresponding record for this reference.
    58. 58
      Liebeschuetz, J. W.; Jones, S. D.; Morgan, P. J.; Murray, C. W.; Rimmer, A. D.; Roscoe, J. M.; Waszkowycz, B.; Welsh, P. M.; Wylie, W. A.; Young, S. C.; Martin, H.; Mahler, J.; Brady, L.; Wilkinson, K. PRO_SELECT: combining structure-based drug design and array-based chemistry for rapid lead discovery. 2. The development of a series of highly potent and selective factor Xa inhibitors J. Med. Chem. 2002, 45, 1221– 1232
      There is no corresponding record for this reference.
    59. 59
      Huang, N.; Shoichet, B. K.; Irwin, J. J. Benchmarking sets for molecular docking J. Med. Chem. 2006, 49, 6789– 6801
      59
      Benchmarking Sets for Molecular Docking
      Huang, Niu; Shoichet, Brian K.; Irwin, John J.
      Journal of Medicinal Chemistry (2006), 49 (23), 6789-6801CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
      Ligand enrichment among top-ranking hits is a key metric of mol. docking. To avoid bias, decoys should resemble ligands phys., so that enrichment is not simply a sepn. of gross features, yet be chem. distinct from them, so that they are unlikely to be binders. We have assembled a directory of useful decoys (DUD), with 2950 ligands for 40 different targets. Every ligand has 36 decoy mols. that are phys. similar but topol. distinct, leading to a database of 98 266 compds. For most targets, enrichment was at least half a log better with uncorrected databases such as the MDDR than with DUD, evidence of bias in the former. These calcns. also allowed 40×40 cross-docking, where the enrichments of each ligand set could be compared for all 40 targets, enabling a specificity metric for the docking screens. DUD is freely available online as a benchmarking set for docking at http://blaster.docking.org/dud/.
    60. 60
      Murcia, M.; Ortiz, A. R. Virtual screening with flexible docking and COMBINE-based models. Application to a series of factor Xa inhibitors J. Med. Chem. 2004, 47, 805– 820
      There is no corresponding record for this reference.
    61. 61
      Nazare, M.; Will, D. W.; Matter, H.; Schreuder, H.; Ritter, K.; Urmann, M.; Essrich, M.; Bauer, A.; Wagner, M.; Czech, J.; Lorenz, M.; Laux, V.; Wehner, V. Probing the subpockets of factor Xa reveals two binding modes for inhibitors based on a 2-carboxyindole scaffold: a study combining structure-activity relationship and X-ray crystallography J. Med. Chem. 2005, 48, 4511– 4525
      61
      Probing the Subpockets of Factor Xa Reveals Two Binding Modes for Inhibitors Based on a 2-Carboxyindole Scaffold: A Study Combining Structure-Activity Relationship and X-ray Crystallography
      Nazare, Marc; Will, David W.; Matter, Hans; Schreuder, Herman; Ritter, Kurt; Urmann, Matthias; Essrich, Melanie; Bauer, Armin; Wagner, Michael; Czech, Joerg; Lorenz, Martin; Laux, Volker; Wehner, Volkmar
      Journal of Medicinal Chemistry (2005), 48 (14), 4511-4525CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
      Structure-activity relationships within a series of highly potent 2-carboxyindole-based factor Xa inhibitors incorporating a neutral P1 ligand are described with particular emphasis on the structural requirements for addressing subpockets of the factor Xa enzyme. Interactions with the subpockets were probed by systematic substitution of the 2-carboxyindole scaffold, in combination with privileged P1 and P4 substituents. Combining the most favorable substituents at the indole nucleus led to the discovery of a remarkably potent factor Xa inhibitor displaying a Ki value of 0.07 nM. X-ray crystallog. of inhibitors bound to factor Xa revealed substituent-dependent switching of the inhibitor binding mode and provided a rationale for the SAR obtained. These results underscore the key role played by the P1 ligand not only in detg. the binding affinity of the inhibitor by direct interaction but also in modifying the binding mode of the whole scaffold, resulting in a nonlinear SAR.
    62. 62
      Pinto, D. J.; Orwat, M. J.; Koch, S.; Rossi, K. A.; Alexander, R. S.; Smallwood, A.; Wong, P. C.; Rendina, A. R.; Luettgen, J. M.; Knabb, R. M.; He, K.; Xin, B.; Wexler, R. R.; Lam, P. Y. Discovery of 1-(4-methoxyphenyl)-7-oxo-6-(4-(2-oxopiperidin-1-yl)phenyl)-4,5,6,7-tetrah ydro-1H-pyrazolo[3,4-c]pyridine-3-carboxamide (apixaban, BMS-562247), a highly potent, selective, efficacious, and orally bioavailable inhibitor of blood coagulation factor Xa J. Med. Chem. 2007, 50, 5339– 5356
      There is no corresponding record for this reference.
    63. 63
      Qiao, J. X.; Chang, C. H.; Cheney, D. L.; Morin, P. E.; Wang, G. Z.; King, S. R.; Wang, T. C.; Rendina, A. R.; Luettgen, J. M.; Knabb, R. M.; Wexler, R. R.; Lam, P. Y. SAR and X-ray structures of enantiopure 1,2-cis-(1R,2S)-cyclopentyldiamine and cyclohexyldiamine derivatives as inhibitors of coagulation Factor Xa Bioorg. Med. Chem. Lett. 2007, 17, 4419– 4427
      There is no corresponding record for this reference.
    64. 64
      Qiao, J. X.; Cheng, X.; Smallheer, J. M.; Galemmo, R. A.; Drummond, S.; Pinto, D. J.; Cheney, D. L.; He, K.; Wong, P. C.; Luettgen, J. M.; Knabb, R. M.; Wexler, R. R.; Lam, P. Y. Pyrazole-based factor Xa inhibitors containing N-arylpiperidinyl P4 residues Bioorg. Med. Chem. Lett. 2007, 17, 1432– 1437
      There is no corresponding record for this reference.
    65. 65
      Senger, S.; Convery, M. A.; Chan, C.; Watson, N. S. Arylsulfonamides: a study of the relationship between activity and conformational preferences for a series of factor Xa inhibitors Bioorg. Med. Chem. Lett. 2006, 16, 5731– 5735
      There is no corresponding record for this reference.
    66. 66
      Watson, N. S.; Brown, D.; Campbell, M.; Chan, C.; Chaudry, L.; Convery, M. A.; Fenwick, R.; Hamblin, J. N.; Haslam, C.; Kelly, H. A.; King, N. P.; Kurtis, C. L.; Leach, A. R.; Manchee, G. R.; Mason, A. M.; Mitchell, C.; Patel, C.; Patel, V. K.; Senger, S.; Shah, G. P.; Weston, H. E.; Whitworth, C.; Young, R. J. Design and synthesis of orally active pyrrolidin-2-one-based factor Xa inhibitors Bioorg. Med. Chem. Lett. 2006, 16, 3784– 3788
      There is no corresponding record for this reference.
    67. 67
      Ye, B.; Arnaiz, D. O.; Chou, Y. L.; Griedel, B. D.; Karanjawala, R.; Lee, W.; Morrissey, M. M.; Sacchi, K. L.; Sakata, S. T.; Shaw, K. J.; Wu, S. C.; Zhao, Z.; Adler, M.; Cheeseman, S.; Dole, W. P.; Ewing, J.; Fitch, R.; Lentz, D.; Liang, A.; Light, D.; Morser, J.; Post, J.; Rumennik, G.; Subramanyam, B.; Sullivan, M. E.; Vergona, R.; Walters, J.; Wang, Y. X.; White, K. A.; Whitlow, M.; Kochanny, M. J. Thiophene-anthranilamides as highly potent and orally available factor Xa inhibitors J. Med. Chem. 2007, 50, 2967– 2980
      67
      Thiophene-Anthranilamides as Highly Potent and Orally Available Factor Xa Inhibitors
      Ye, Bin; Arnaiz, Damian O.; Chou, Yuo-Ling; Griedel, Brian D.; Karanjawala, Rushad; Lee, Wheeseong; Morrissey, Michael M.; Sacchi, Karna L.; Sakata, Steven T.; Shaw, Kenneth J.; Wu, Shung C.; Zhao, Zuchun; Adler, Marc; Cheeseman, Sarah; Dole, William P.; Ewing, Janice; Fitch, Richard; Lentz, Dao; Liang, Amy; Light, David; Morser, John; Post, Joseph; Rumennik, Galina; Subramanyam, Babu; Sullivan, Mark E.; Vergona, Ron; Walters, Janette; Wang, Yi-Xin; White, Kathy A.; Whitlow, Marc; Kochanny, Monica J.
      Journal of Medicinal Chemistry (2007), 50 (13), 2967-2980CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
      There remains a high unmet medical need for a safe oral therapy for thrombotic disorders. The serine protease factor Xa (fXa), with its central role in the coagulation cascade, is among the more promising targets for anticoagulant therapy and has been the subject of intensive drug discovery efforts. Investigation of a hit from high-throughput screening identified a series of thiophene-substituted anthranilamides as potent nonamidine fXa inhibitors. Lead optimization by incorporation of hydrophilic groups led to the discovery of compds. with picomolar inhibitory potency and micromolar in vitro anticoagulant activity. Based on their high potency, selectivity, oral pharmacokinetics, and efficacy in a rat venous stasis model of thrombosis, compds. ZK 814048 (10b), ZK 810388 (13a), and ZK 813039 (17m) were advanced into development.
    68. 68
      Young, R. J.; Brown, D.; Burns-Kurtis, C. L.; Chan, C.; Convery, M. A.; Hubbard, J. A.; Kelly, H. A.; Pateman, A. J.; Patikis, A.; Senger, S.; Shah, G. P.; Toomey, J. R.; Watson, N. S.; Zhou, P. Selective and dual action orally active inhibitors of thrombin and factor Xa Bioorg. Med. Chem. Lett. 2007, 17, 2927– 2930
      68
      Selective and dual action orally active inhibitors of thrombin and factor Xa
      Young, Robert J.; Brown, David; Burns-Kurtis, Cynthia L.; Chan, Chuen; Convery, Maire A.; Hubbard, Julia A.; Kelly, Henry A.; Pateman, Anthony J.; Patikis, Angela; Senger, Stefan; Shah, Gita P.; Toomey, John R.; Watson, Nigel S.; Zhou, Ping
      Bioorganic & Medicinal Chemistry Letters (2007), 17 (10), 2927-2930CODEN: BMCLE8; ISSN:0960-894X. (Elsevier Ltd.)
      The synthetic entry to new classes of dual fXa/thrombin and selective thrombin inhibitors with significant oral bioavailability is described. This was achieved through minor modifications to the sulfonamide group in our potent and selective fXa inhibitor (E)-2-(5-chlorothien-2-yl)-N-{(3S)-1-[(1S)-1-methyl-2-(morpholin-4-yl)-2-oxoethyl]-2-oxopyrrolidin-3-yl}ethenesulfonamide and these obsd. activity changes have been rationalized using structural studies.
    69. 69
      Young, R. J.; Campbell, M.; Borthwick, A. D.; Brown, D.; Burns-Kurtis, C. L.; Chan, C.; Convery, M. A.; Crowe, M. C.; Dayal, S.; Diallo, H.; Kelly, H. A.; King, N. P.; Kleanthous, S.; Mason, A. M.; Mordaunt, J. E.; Patel, C.; Pateman, A. J.; Senger, S.; Shah, G. P.; Smith, P. W.; Watson, N. S.; Weston, H. E.; Zhou, P. Structure- and property-based design of factor Xa inhibitors: pyrrolidin-2-ones with acyclic alanyl amides as P4 motifs Bioorg. Med. Chem. Lett. 2006, 16, 5953– 5957
      69
      Structure- and property-based design of factor Xa inhibitors: Pyrrolidin-2-ones with acyclic alanyl amides as P4 motifs
      Young, Robert J.; Campbell, Matthew; Borthwick, Alan D.; Brown, David; Burns-Kurtis, Cynthia L.; Chan, Chuen; Convery, Maire A.; Crowe, Miriam C.; Dayal, Satish; Diallo, Hawa; Kelly, Henry A.; King, N. Paul; Kleanthous, Savvas; Mason, Andrew M.; Mordaunt, Jackie E.; Patel, Champa; Pateman, Anthony J.; Senger, Stefan; Shah, Gita P.; Smith, Paul W.; Watson, Nigel S.; Weston, Helen E.; Zhou, Ping
      Bioorganic & Medicinal Chemistry Letters (2006), 16 (23), 5953-5957CODEN: BMCLE8; ISSN:0960-894X. (Elsevier Ltd.)
      Nonracemic 1-alaninyl-3-(6-chloro-2-naphthylsulfonylamino)-2-pyrrolidinones such as I are prepd. as factor Xa inhibitors for use as anticoagulants; the hydrophobicities, anticoagulant activities in rats, and pharmacokinetics of some of the compds. are detd. For example, I inhibits factor Xa with a Ki value of 1 nM and a value of 2.5 μM in the prothrombin time assay. 1-Alaninyl-3-(6-chloro-2-naphthylsulfonylamino)-2-pyrrolidinones have poorer pharmacokinetic profiles than the corresponding morpholine-based analogs, with increased plasma clearance and decreased oral bioavailabilities. The structure of I bound to factor Xa is detd. by X-ray crystallog.
    70. 70
      Verdonk, M. L.; Berdini, V.; Hartshorn, M. J.; Mooij, W. T.; Murray, C. W.; Taylor, R. D.; Watson, P. Virtual screening using protein-ligand docking: avoiding artificial enrichment J. Chem. Inf. Comput. Sci. 2004, 44, 793– 806
      70
      Virtual screening using protein-ligand docking: Avoiding artificial enrichment
      Verdonk, Marcel L.; Berdini, Valerio; Hartshorn, Michael J.; Mooij, Wijnand T. M.; Murray, Christopher W.; Taylor, Richard D.; Watson, Paul
      Journal of Chemical Information and Computer Sciences (2004), 44 (3), 793-806CODEN: JCISD8; ISSN:0095-2338. (American Chemical Society)
      This study addresses a no. of topical issues around the use of protein-ligand docking in virtual screening. We show that, for the validation of such methods, it is key to use focused libraries (contg. compds. with one-dimensional properties, similar to the actives), rather than "random" or "drug-like" libraries to test the actives against. We also show that, to obtain good enrichments, the docking program needs to produce reliable binding modes. We demonstrate how pharmacophores can be used to guide the dockings and improve enrichments, and we compare the performance of three consensus-ranking protocols against ranking based on individual scoring functions. Finally, we show that protein-ligand docking can be an effective aid in the screening for weak, fragment-like binders, which has rapidly become a popular strategy for hit identification. All results presented are based on carefully constructed virtual screening expts. against four targets, using the protein-ligand docking program GOLD.
    71. 71
      Jacobsson, M.; Karlen, A. Ligand bias of scoring functions in structure-based virtual screening J. Chem. Inf. Model. 2006, 46, 1334– 1343
      71
      Ligand Bias of Scoring Functions in Structure-Based Virtual Screening
      Jacobsson, Micael; Karlen, Anders
      Journal of Chemical Information and Modeling (2006), 46 (3), 1334-1343CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
      A total of 945 known actives and roughly 10 000 decoy compds. were docked to eight different targets, and the resulting poses were scored using 10 different scoring functions. Three different score postprocessing methods were evaluated with respect to improvement of the enrichment in virtual screening. The three procedures were (i) multiple active site correction (MASC) as has been proposed by Vigers and Rizzi, (ii) a variation of MASC where corrections terms are predicted from simple mol. descriptors through PLS, PLS MASC, and (iii) size normalization. It was found that MASC did not generally improve the enrichment factors when compared to uncorrected scoring functions. For some combinations of scoring functions and targets, the enrichment was improved, for others not. However, by excluding the std. deviation from the MASC equation and transforming the scores for each target to a mean of 0 and a std. deviation of 1 (unit variance normalization), the performance was improved as compared to the original MASC method for most combinations of targets and scoring functions. Furthermore, when the mol. descriptors were fit to the mean scores over all targets and the resulting PLS models were used to predict mean scores, the enrichment as compared to the raw score was improved more often than by straightforward MASC. A high to intermediate linear correlation between the score and the no. of heavy atoms was found for all scoring functions except FlexX. There seems to be a correlation between the size dependence of a scoring function and the effectiveness of PLS MASC in increasing the enrichment for that scoring function. Finally, normalization by mol. wt. or heavy atom count was sometimes successful in increasing the enrichment. Dividing by the square or cubic root of the mol. wt. or heavy atom count instead was often more successful. These results taken together suggest that ligand bias in scoring functions is a source of false positives in structure-based virtual screening. The no. of false positives caused by ligand bias may be decreased using, for example, the PLS MASC procedure proposed in this study.
    72. 72
      Krovat, E. M.; Steindle, T.; Langer, T. Recent advances in docking and scoring Curr. Comput.-Aided Drug Des. 2005, 1, 93– 102
      72
      Recent advances in docking and scoring
      Krovat, E. M.; Steindl, T.; Langer, T.
      Current Computer-Aided Drug Design (2005), 1 (1), 93-102CODEN: CCDDAS; ISSN:1573-4099. (Bentham Science Publishers Ltd.)
      A review on recent advances and new aspects in the field of mol. docking and scoring, and covers multiple applications and case studies. Basic requirements and different algorithms for docking are briefly discussed. Moreover, parameters that influence docking results, combination of different docking algorithms and scoring functions, performance of scoring functions, docking using homol. models, and ligands and protein flexibility are examd. to give an overview of the state-of-the-art methods and a survey of innovative approaches in mol. docking and scoring. Regarding the enormous amt. of literature in this field, an overview is given of several important advances in docking and scoring techniques published within the last two years, i.e. publications ranging from 2002 to 2004.
    73. 73
      Carlson, H. A.; Smith, R. D.; Khazanov, N. A.; Kirchhoff, P. D.; Dunbar, J. B.; Benson, M. L. Differences between high- and low-affinity complexes of enzymes and nonenzymes J. Med. Chem. 2008, 51, 6432– 6441
      There is no corresponding record for this reference.
    74. 74
      Gohlke, H.; Klebe, G. Approaches to the description and prediction of the binding affinity of small-molecule ligands to macromolecular receptors Angew. Chem., Int. Ed. Engl. 2002, 41, 2644– 2676
      74
      Approaches to the description and prediction of the binding affinity of small-molecule ligands to macromolecular receptors
      Gohlke, Holger; Klebe, Gerhard
      Angewandte Chemie, International Edition (2002), 41 (15), 2644-2676CODEN: ACIEF5; ISSN:1433-7851. (Wiley-VCH Verlag GmbH)
      A review. The influence of a xenobiotic compd. on an organism is usually summarized by the expression biol. activity. If a controlled, therapeutically relevant, and regulatory action is obsd. the compd. has potential as a drug, otherwise its toxicity on the biol. system is of interest. However, what do we understand by the biol. activity. In principle, the overall effect on an organism has to be considered. However, because of the complexity of the interrelated processes involved, as a simplification primarily the "main action" on the organism is taken into consideration. On the mol. level, biol. activity corresponds to the binding of a (lowmol. wt.) compd. to a macromol. receptor, usually a protein. Enzymic reactions or signal-transduction cascades are thereby influenced with respect to their function for the organism. We regard this binding as a process under equil. conditions; thus, binding can be described as an assocn. or dissocn. process. Accordingly, biol. activity is expressed as the affinity of both partners for each other, as a thermodn. equil. quantity. How well do we understand these terms and how well are they theor. predictable today. The holy grail of rational drug design is the prediction of the biol. activity of a compd. The processes involving ligand binding are extremely complicated, both ligand and protein are flexible mols., and the energy inventory between the bound and unbound states must be considered in aq. soln. How sophisticated and reliable are our exptl. approaches to obtaining the necessary insight. The present review summarizes our current understanding of the binding affinity of a small-mol. ligand to a protein. Both theor. and empirical approaches for predicting binding affinity, starting from the three-dimensional structure of a protein-ligand complex, will be described and compared. Exptl. methods, primarily microcalorimetry, will be discussed. As a perspective, our own knowledge-based approach towards affinity prediction and exptl. data on factorizing binding contributions to protein-ligand binding will be presented.
    75. 75
      Davis, A. M.; Teague, S. J. Hydrogen Bonding, Hydrophobic Interactions, and Failure of the Rigid Receptor Hypothesis Angew. Chem., Int. Ed. Engl. 1999, 38, 736– 749
      There is no corresponding record for this reference.
    76. 76
      Wildman, S. A.; Crippen, G. M. Prediction of physicochemical parameters by atomic contributions J. Chem. Inf. Comput. Sci. 1999, 39, 868– 873
      76
      Prediction of Physicochemical Parameters by Atomic Contributions
      Wildman, Scott A.; Crippen, Gordon M.
      Journal of Chemical Information and Computer Sciences (1999), 39 (5), 868-873CODEN: JCISD8; ISSN:0095-2338. (American Chemical Society)
      We present a new atom type classification system for use in atom-based calcn. of partition coeff. (log P) and molar refractivity (MR) designed in part to address published concerns of previous at. methods. The 68 at. contributions to log P have been detd. by fitting an extensive training set of 9920 mols., with r2 = 0.918 and σ = 0.677. A sep. set of 3412 mols. was used for the detn. of contributions to MR with r2 = 0.997 and σ = 1.43. Both calcns. are shown to have high predictive ability.
    77. 77
      David, L.; Amara, P.; Field, M. J.; Major, F. Parametrization of a force field for metals complexed to biomacromolecules: applications to Fe(II), Cu(II) and Pb(II) J. Comput.- Aided Mol. Des. 2002, 16, 635– 651
      77
      Parametrization of a force field for metals complexed to biomacromolecules: applications to Fe(II), Cu(II) and Pb(II)
      David, Laurent; Amara, Patricia; Field, Martin J.; Major, Francois
      Journal of Computer-Aided Molecular Design (2002), 16 (8/9), 635-651CODEN: JCADEQ; ISSN:0920-654X. (Kluwer Academic Publishers)
      Although techniques for the simulation of biomols., such as proteins and RNAs, have greatly advanced in the last decade, modeling complexes of biomols. with metal ions remains problematic. Precise calcns. can be done with quantum mech. methods but these are prohibitive for systems the size of macromols. More qual. modeling can be done with mol. mech. potentials but the parametrization of force fields for metals is often difficult, particularly if the bonding between the metal and the groups in its coordination shell has significant covalent character. In this paper we present a method for deriving bond and bond-angle parameters for metal complexes from exptl. bond and bond-angle distributions obtained from the Cambridge Structural Database. In conjunction with this method, we also introduce a non-std. energy term of gaussian form that allows us to obtain a stable description of the coordination about a metal center during a simulation. The method was evaluated on Fe(II)-porphyrin complexes, on simple Cu(II) ion complexes and a no. of complexes of the Pb(II) ion.
    78. 78
      Irwin, J. J.; Raushel, F. M.; Shoichet, B. K. Virtual screening against metalloenzymes for inhibitors and substrates Biochemistry 2005, 44, 12316– 12328
      78
      Virtual Screening against Metalloenzymes for Inhibitors and Substrates
      Irwin, John J.; Raushel, Frank M.; Shoichet, Brian K.
      Biochemistry (2005), 44 (37), 12316-12328CODEN: BICHAW; ISSN:0006-2960. (American Chemical Society)
      Mol. docking uses the three-dimensional structure of a receptor to screen databases of small mols. for potential ligands, often based on energetic complementarity. For many docking scoring functions, which calc. nonbonded interactions, metalloenzymes are challenging because of the partial covalent nature of metal-ligand interactions. To investigate how well mol. docking can identify potential ligands of metalloenzymes using a "std." scoring function, we have docked the MDL Drug Data Report (MDDR), a functionally annotated database of 95 000 small mols., against the X-ray crystal structures of five metalloenzymes. These enzymes included three zinc proteases, the nickel analog of an iron enzyme, and a molybdenum metalloenzyme. The ability of the docking program to retrospectively enrich the annotated ligands as high-scoring hits for each enzyme and to calc. proper geometries was evaluated. In all five systems, the annotated ligands within the MDDR were enriched at least 20 times over random. To test the approach prospectively, a sixth target, the zinc β-lactamase from Bacteroides fragilis, was screened against the fragment-like subset of the ZINC database. We purchased and tested 15 compds. from among the top 50 top-ranked ligands from docking, and found 5 inhibitors with apparent Ki values less than 120 μM, the best of which was 2 μM. A more ambitious test still was predicting actual substrates for a seventh target, a Zn-dependent phosphotriesterase from Pseudomonas diminuta. Screening the Available Chems. Directory (ACD) identified 25 thiophosphate esters as potential substrates within the top 100 ranked compds. Eight of these, all previously uncharacterized for this enzyme, were acquired and tested, and all were confirmed exptl. as substrates. These results suggest that a simple, noncovalent scoring function may be used to identify inhibitors of at least some metalloenzymes.
    79. 79
      Li, X.; Hayik, S. A.; Merz, K. M., Jr. QM/MM X-ray refinement of zinc metalloenzymes J. Inorg. Biochem. 2010, 104, 512– 522
      There is no corresponding record for this reference.
    80. 80
      Lipinski, C. A.; Lombardo, F.; Dominy, B. W.; Feeney, P. J. Experimental and Compoutational Approaches to Estimate Solubility and Permeability in Drug Discover and Development Settings Adv. Drug Delivery Rev. 1997, 23, 3– 25
      There is no corresponding record for this reference.
    81. 81
      Oprea, T. I. Property distribution of drug-related chemical databases J. Comput.- Aided Mol. Des. 2000, 14, 251– 264
      81
      Property distribution of drug-related chemical databases
      Oprea, Tudor I.
      Journal of Computer-Aided Molecular Design (2000), 14 (3), 251-264CODEN: JCADEQ; ISSN:0920-654X. (Kluwer Academic Publishers)
      The process of compd. selection and prioritization is crucial for both combinatorial chem. (CBC) and high throughput screening (HTS). Compd. libraries have to be screened for unwanted chem. structures, as well as for unwanted chem. properties. Property extrema can be eliminated by using property filters, in accordance with their actual distribution. Property distribution was examd. in the following compd. databases: MACCS-II Drug Data Report (MDDR), Current Patents Fast-alert, Comprehensive Medicinal Chem., Physician Desk Ref., New Chem. Entities, and the Available Chem. Directory (ACD). The ACDF and MDDRF subsets were created by removing reactive functionalities from the ACD and MDDR databases, resp. The ACDF subset was further filtered by keeping only mols. with a "drug-like" score below 0.8. The following properties were examd.: mol. wt. (MW), the calcd. octanol/water partition coeff. (CLOGP), the no. of rotatable (RTB) and rigid bonds (RGB), the no. of rings (RNG), and the no. of hydrogen bond donors (HDO) and acceptors (HAC). Of these, MW and CLOGP follow a Gaussian distribution, whereas all other descriptors have an asym. (truncated Gaussian) distribution. Four out of five compds. in ACDF and MDDRF pass the "rule of 5" test, a probability scheme that ests. oral absorption proposed by Lipinski et al. Because property distributions of HDO, HAC, MW and CLOGP (used in the "rule of 5" test) do not differ significantly between these datasets, the "rule of 5" does not distinguish "drugs" from "nondrugs". Therefore, Pareto analyses were performed to examine skewed distributions in all compd. collections. Seventy percent of the "drug-like" compds. were found between the following limits: 0 ≤ HDO ≤ 2, 2 ≤ HAC ≤ 9, 2 ≤ RTB ≤ 8, and 1 ≤ RNG ≤ 4, resp. The no. of launched drugs in MDDR having 0 ≤ HDO ≤ 2 is 4.8 times higher than the no. of drugs having 3 ≤ HDO ≤ 5. Skewed distributions can be exploited to focus on the "drug-like space": 62.68% of ACDF ("nondrug-like") compds. have 0 ≤ RNG ≤ 2, and RGB ≤ 17, while 28.88% of ACDF compds. have 3 ≤ RNG ≤ 13, and 18 ≤ RGB ≤ 56. By contrast, 61.22% of MDDRF compds. have RNG ≥ 3, and RGB ≥ 18, and only 24.73% of MDDRF compds. have 0 ≤ RNG ≤ 2 rings, and RGB ≤ 17. The probability of identifying "drug-like" structures increases with mol. complexity.
    82. 82
      Hunenberger, P. H.; Helms, V.; Narayana, N.; Taylor, S. S.; McCammon, J. A. Determinants of ligand binding to cAMP-dependent protein kinase Biochemistry 1999, 38, 2358– 2366
      There is no corresponding record for this reference.
    83. 83
      Brown, S. P.; Muchmore, S. W.; Hajduk, P. J. Healthy skepticism: assessing realistic model performance Drug Discovery Today 2009, 14, 420– 427
      83
      Healthy skepticism: assessing realistic model performance
      Brown Scott P; Muchmore Steven W; Hajduk Philip J
      Drug discovery today (2009), 14 (7-8), 420-7 ISSN:.
      Although the development of computational models to aid drug discovery has become an integral part of pharmaceutical research, the application of these models often fails to produce the expected impact on productivity. One reason for this may be that the expected performance of many models is simply not supported by the underlying data, because of often neglected effects of assay and prediction errors on the reliability of the predicted outcome. Another significant challenge to realizing the full potential of computational models is their integration into prospective medicinal chemistry campaigns. This article will analyze the impact of assay and prediction error on model quality, and explore scenarios where computational models can expect to have a significant influence on drug discovery research.
    84. 84
      Czodrowski, P.; Sotriffer, C. A.; Klebe, G. Protonation changes upon ligand binding to trypsin and thrombin: structural interpretation based on pKa calculations and ITC experiments J. Mol. Biol. 2007, 367, 1347– 1356
      84
      Protonation Changes upon Ligand Binding to Trypsin and Thrombin: Structural Interpretation Based on pKa Calculations and ITC Experiments
      Czodrowski, Paul; Sotriffer, Christoph A.; Klebe, Gerhard
      Journal of Molecular Biology (2007), 367 (5), 1347-1356CODEN: JMOBAK; ISSN:0022-2836. (Elsevier Ltd.)
      The protonation states of a protein and a ligand can be altered upon complex formation. Such changes can be detected exptl. by isothermal titrn. calorimetry (ITC). For a series of ligands binding to the serine proteases trypsin and thrombin, we previously performed an extensive ITC and crystallog. study and were able to identify protonation changes for four complexes. However, since ITC measures only the overall proton exchange, it does not provide structural insights into the functional groups involved in the proton transfer. Using Poisson-Boltzmann calcns. based on our recently developed PEOE_PB charges, we compute pKa values for all complexes of our former study in order to reveal the residues with altered protonation states. The results indicate that His57, a member of the catalytic triad, is responsible for the most relevant pKa shifts leading to the exptl. detected protonation changes. This finding is in contrast to our previous assumption that the obsd. protonation changes occur at the carboxylic group of the ligands. The newly detected proton acceptor is used for a revised factorization of the ITC data, which is necessary whenever the protonation inventory changes upon complexation. The pK a values of complexes showing no protonation change in the ITC expt. are reliably predicted in most cases, whereas predictions of strongly coupled systems remain problematic.
    85. 85
      Chow, M. A.; McElroy, K. E.; Corbett, K. D.; Berger, J. M.; Kirsch, J. F. Narrowing substrate specificity in a directly evolved enzyme: the A293D mutant of aspartate aminotransferase Biochemistry 2004, 43, 12780– 12787
      There is no corresponding record for this reference.
    86. 86
      Goldberg, R. N.; Kishore, N.; Lennen, R. M. Thermodynamic Quantities for the Ionization Reactions of Buffers J. Phys. Chem. Ref. Data 2002, 31, 231– 370
      86
      Thermodynamic quantities for the ionization reactions of buffers
      Goldberg, Robert N.; Kishore, Nand; Lennen, Rebecca M.
      Journal of Physical and Chemical Reference Data (2002), 31 (2), 231-370CODEN: JPCRBU; ISSN:0047-2689. (American Institute of Physics)
      This review contains selected values of thermodn. quantities for the aq. ionization reactions of 64 buffers, many of which are used in biol. research. Since the aim is to be able to predict values of the ionization const. at temps. not too far from ambient, the thermodn. quantities which are tabulated are the pK, std. molar Gibbs energy ΔrG°, std. molar enthalpy ΔrH°, and std. molar heat capacity change ΔrCp° for each of the ionization reactions at the temp. T = 298.15 K and the pressure p = 0.1 MPa. The std. state is the hypothetical ideal soln. of unit molality. The chem. name(s) and CAS registry no., structure, empirical formula, and mol. wt. are given for each buffer considered herein. The selection of the values of the thermodn. quantities for each buffer is discussed.

  • Supporting Information

    Supporting Information

    Table of 10 proteins with ligand series and the performance of each of the 17 core codes on relative ranking, a discussion of methods and metrics for identifying the GOOD complexes, and a complete listing of GOOD and BAD complexes. This material is available free of charge via the Internet at http://pubs.acs.org.


    Terms & Conditions

    Most electronic Supporting Information files are available without a subscription to ACS Web Editions. Such files may be downloaded by article for research use (if there is a public use license linked to the relevant article, that license may permit other uses). Permission may be obtained from ACS for other uses through requests via the RightsLink permission system: http://pubs.acs.org/page/copyright/permissions.html.