Download Hi-Res ImageDownload to MS-PowerPointCite This:J. Chem. Inf. Model. 2017, 57, 8, 1832-1846

FAME 2: Simple and Effective Machine Learning Model of Cytochrome P450 Regioselectivity

View Author Information

† Faculty of Mathematics, Informatics and Natural Sciences, Department of Computer Science, Center for Bioinformatics, Universität Hamburg, Hamburg, 20146, Germany

‡ CZ-OPENSCREEN: National Infrastructure for Chemical Biology, Laboratory of Informatics and Chemistry, Faculty of Chemical Technology, University of Chemistry and Technology Prague, 166 28 Prague 6, Czech Republic

*E-mail: [email protected]. Tel.: +49 (0)40 42838 7303 (J.K.).

Cite this: J. Chem. Inf. Model. 2017, 57, 8, 1832–1846

Publication Date (Web):August 7, 2017

https://doi.org/10.1021/acs.jcim.7b00250

Request reuse permissions

Article Views

5354

Altmetric

Citations

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.

PDF (4 MB)

Get e-Alerts

Supporting Info (1)»Supporting Information Supporting Information

SUBJECTS:

Go to Journal of Chemical Information and Modeling

Get e-Alerts

Abstract

We report on the further development of FAst MEtabolizer (FAME; J. Chem. Inf. Model.2013, 53, 2896–2907), a collection of random forest models for the prediction of sites of metabolism (SoMs) of xenobiotics. A broad set of descriptors was explored, from simple 2D descriptors such as those used in FAME, to quantum chemical descriptors employed in some of the most accurate models for SoM prediction currently available. In line with the original FAME approach, our objective was to keep things simple and to come up with accurate and robust models that are based on a small number of 2D descriptors. We found that circular descriptions of atoms and their environments with such descriptors in combination with an extremely randomized trees algorithm can yield models that perform equally well compared to more complex approaches. Thorough evaluation experiments on an independent test set showed that the best of these models obtained a Matthews correlation coefficient, area under the receiver operating characteristic curve, and Top-2 accuracy of 0.57, 0.91 and 94.1%, respectively. Models for the prediction of isoform-specific regioselectivity of CYP 3A4, 2D6, and 2C9 were also developed and showed competitive performance. The best models have been integrated into a newly developed software package (FAME 2), which is available free of charge from the authors.

Introduction

ARTICLE SECTIONS

Jump To

The metabolic system has evolved as a primary line of defense of organisms against potentially harmful xenobiotics. Biotransformation can yield metabolites that differ substantially from their parent compounds with respect to biological and physicochemical properties. (1) For example, only about 3% of all drug metabolites are confirmed to maintain pharmacological activity, and even though the primary function of the metabolic system is to detoxify substances, at least 7% of metabolites are known as toxic or reactive. (2) Therefore, understanding metabolism is essential for the development of effective and safe compounds, in particular drugs, agrochemicals and cosmetics.

The ability to predict and identify metabolically labile atom positions in a molecule (i.e., sites of metabolism, SoMs) can aid in the derivation of likely metabolites and, consequently, strategies for the optimization of the metabolic properties of small molecules. However, experimental identification of SoMs or metabolites is an expensive and time-consuming task. Therefore, being able to accurately predict SoMs without the need to carry out laboratory experiments would greatly reduce costs and time needed to evaluate each compound. (3)

In the past few years, in silico methods for SoM prediction have shown their ability to carry out such predictions with good accuracy, (4) and multiple approaches have been presented and reviewed in the literature. (3, 5-9)

In principle, a SoM prediction methodology should account for three main factors: (i) reactivity of atom positions in a molecule, (ii) pharmacophoric and shape constraints imposed by the binding site of the enzyme, and (iii) the accessibility of potential sites from the catalytic center. In the case of cytochrome P450 enzymes (CYPs), CYP 3A4 in particular, reactivity often plays a more significant role than the two other factors because several CYP isoforms have large and flexible binding sites and can accommodate a plethora of substrates. (10) However, taking accessibility and pharmacophoric constraints into account as well can often increase accuracy and has more impact when modeling metabolism of less flexible and promiscuous isoforms such as CYP 2D6. (11)

In silico methods for SoM prediction can be roughly divided into three classes: (i) rule-based, (ii) structure-based, and (iii) ligand-based. Rule-based methods attempt to derive likely metabolites and SoMs by applying a dictionary of biotransformation rules (usually compiled by human experts) to molecular fragments. (12-21) In relation to rule-based methods, structure-based approaches are on the opposite side of the spectrum. Their objective is to focus on the substrate–protein interaction in more detail and not to abstract it away by looking at the transformations and their outcomes alone. (8, 22) Therefore, very different information can be obtained from structure-based modeling. The most significant advantage is that more insight is given into the orientation of a ligand at the binding site, which can provide additional benefits when optimizing the metabolic properties of compounds. In principle, it is possible to evaluate whether a compound is a plausible ligand or not, and how ligand stereochemistry influences the outcome of the biotransformation. However, this more detailed information comes at a higher computational cost and requires expert input. Moreover, it is often difficult to accurately capture the flexibility of the CYP isoforms. (9) In fact, the observed conformational space of the binding sites of CYPs observed in X-ray structures does not account for a substantial number of known substrates. (3) Structure-based approaches also suffer from typical docking problems. This means that the scoring functions are usually not sufficiently accurate to predict the relative binding free energy. (23) Additionally, water is not always explicitly modeled or modeled at all. Thus, some important hydrogen bonds that stabilize the substrate–enzyme complex or important effects related to entropy can be easily missed. Overall, structure-based approaches can be directly interpreted and provide a lot of information, but accurate modeling requires high quality crystal structures covering the relevant conformational space, sophisticated scoring methods, more computation time and considerable human effort during the preparation and analysis of the data. Therefore, structure-based approaches are more suited for a thorough investigation of a particular substrate–enzyme pair of interest than for a general screen of compounds.

Ligand-based methods do not explicitly take any structural information on substrate-enzyme complexes into account but rely heavily on available experimental data on SoMs and metabolites. They can have higher throughput and still rival structure-based methods in performance. It can also be argued that, with a large amount of high-quality data and a reliable set of descriptors, both flexibility of the enzymes and reactivity of the substrates can be captured with comparable accuracy from empirical data, albeit with the important exception of metabolic stereoselectivity. A lot of methods have been developed over the years, (3) and most of the modern approaches employ machine learning and atom-level descriptors to model CYP regioselectivity. One of the first such methods, published by Sheridan et al., (24) was a reasonably predictive random forest model which relied on simple 2D topological descriptors that encoded local chemical environments and atom accessibility. This method was more accurate than some mechanistic approaches that explicitly modeled reactivity but did not take accessibility into account. This outcome showed that accessibility is an important factor, even for flexible isoforms such as CYP 3A4, and that a relatively simple empirical approach can be quite successful when enough data are available.

SMARTCyp is a freely available software package (25) and web service (26) for SoM prediction which does not rely on empirical data but on precomputed activation energies from density functional theory (DFT) calculations of model compounds. These energies were used to derive a lookup table which assigns an activation energy for a particular fragment in the query molecule as a measure of a site’s reactivity. Accessibility of different sites was also included via a simple descriptor that was used to correct the final reactivity score. (11) This approach was able to predict a correct SoM within the top two highest ranking positions for 76% and 83% of the molecules of two validation sets. (11) Later, an additional accessibility descriptor was added, which resulted in a modest increase in performance. (27)

Two methods that have made heavy use of machine learning algorithms are RS-predictor (28-30) and its successor, XenoSite. (31, 32) Both methods used quite a large number of different descriptors to capture the reactivity and accessibility of potential SoMs. RS-predictor was further improved by inclusion of SMARTCyp reactivities to model reactivity more accurately. (29) XenoSite is the final evolutionary step of RS-predictor so far and employs a neural network as a machine learning algorithm. (31, 32) Using a leave-one-out cross-validation procedure, XenoSite was found to give at least one correct SoM prediction in the top two ranking positions for 89% of the molecules. A major contribution to SMARTCyp, RS-predictor and XenoSite was a large publicly available (“Zaretzki”) data set, which includes SoM data for multiple CYP isoforms and which has been used to develop and test many other approaches as well.

A rather simple but effective machine learning method was implemented by Kirchmair et al. in FAME. (33) It relies on just seven descriptors based on the 2D representation of a molecule and uses a random forest model to facilitate predictions. Despite its simplicity, this approach was able to identify the correct SoM in the top two ranking positions for up to 81% of molecules from an external validation set, and it also allows the prediction of biotransformations catalyzed by non-CYP enzymes.

A progressive method in terms of estimation of activation energies was recently presented by Tyzack et al. (34) This method does not rely on a library of precomputed fragments to estimate the relative activation energy of different sites like SMARTCyp but performs effective semiempirical quantum chemical calculations of the whole substrate. For most CYP isoforms, this method was able to give a correct prediction in the top two ranked positions for more than 85% of the molecules in an external validation set.

Another recent contribution to the world of ligand-based approaches is a method published by He et al. (35) In this case, several models were created for each CYP-mediated reaction. The method relies on 56 quantum chemical descriptors, 29 physicochemical properties and 27 topological descriptors. Several methods from the machine learning toolbox were tested and the built models were shown to perform well on an external test set, with nearly 88% of the reactions predicted correctly.

More recently, Finkelmann et al. (36) reported on a new SoM predictor based on ab initio calculations that were used to describe electronic properties of molecules in the open source data set published with XenoSite. A scheme was introduced to encode partial charge distributions into atomic descriptors capable of not only describing potential SoMs themselves but also the properties of neighboring atoms. In this way, each SoM was described in a larger context rather than being treated as an independent sample. It was demonstrated that using such descriptors had a positive influence on machine learning model performance. The models presented in the study were able to predict a correct SoM within the first two ranked atom positions for 91% of the compounds in leave-one-out cross-validation experiments.

In this work, we present an accurate SoM prediction methodology further developed from the simple machine learning approach originally implemented in FAME. (33) Rather than using the random forest machine learning algorithm used in the original method, in this work extremely randomized trees (37) are applied. Using a revised version of the Zaretzki data set, we investigate the influence of various molecular descriptors (both topological and quantum chemical) on the accuracy of this ligand-based approach. We also describe a relatively simple 2D method to capture atomic environments of potential SoMs and show that models built using such descriptors have comparable or better performance than other more complex approaches reported in the literature. Finally, a freely available software package is introduced in order to make it possible for researchers to readily use the models presented in this work.

Results

ARTICLE SECTIONS

Jump To

All models were trained and tested on a revised version of the Zaretzki data set (see Methods for detail). The Zaretzki data set has been successfully applied in several other studies (27, 36, 38-40) and, thus, is well-established in the field. In total, the prepared data contained information on 15,233 atoms, a sufficient number of data points for machine learning. It should, however, be noted that these are not entirely independent samples. Prior to any experiments, this data set was randomly split on molecules into a training set and an independent test set using an 80:20 ratio (Figure 1). This resulted in a training set that contained data on 542 compounds and a test set of 136 molecules (Table 1).

Figure 1. Overview of the data preparation, model building, and model evaluation workflow.

Table 1. Number of Compounds in Each Dataset Used in the Current Study

CYP isoforma	totalb	training setc	test setd	uncorr test sete
global	678	542	136	71
3A4	473	378	95	52
2D6	269	215	54	32
2C9	225	180	45	32

The code of the isoform for which a specific model was built. The joined global model which includes all isoforms in the Zaretzki data set is marked as “global”.

Total number of compounds that had SoMs annotated for the particular isoform in the data set.

Compounds randomly selected for training and hyperparameter optimization.

Compounds randomly selected for testing.

Compounds that remained in the uncorrelated test set after the similarity filter was applied.

Model Evaluation Measures

In order to evaluate the performance of the developed models, an appropriate set of evaluation measures needed to be selected. We used three different metrics: (i) the Matthews correlation coefficient (MCC), (ii) area under the ROC curve (AUC), and (iii) Top-k. The MCC was used as the main classification quality metric in this study. The MCC is a balanced measure which not only takes into account true and false positives but also negatives. It is calculated according to

(1)where TP is the number of true positive, TN the number of true negative, FP the number of false positive and FN the number of false negative predictions. Because the MCC includes all items of the confusion matrix, it is regarded as a rather robust and balanced measure which is suitable for data sets with imbalanced classes. (41) Therefore, it was used as the main quality measure in this work and as a scoring function during model parameter optimization. In addition to the MCC, we also used the AUC, which provides insight into how well the models are able to rank the putative sites according to the probabilities given by the ensemble of decision trees.

Related to the AUC is the third quality measure reported in this study, the Top-k metric, which is commonly used for model evaluation in metabolism regioselectivity prediction. The Top-k measure provides a means to evaluate performance in terms of molecules rather than individual atoms. When calculating the measure, the atoms in the molecule are sorted according to the ranking obtained from the model and if at least one known SoM is discovered within the top k positions in the sorted list, the molecule is marked as correctly predicted. The percentage of the correctly predicted compounds in the data set is then reported. The most commonly used values of k are 2 and 3.

Model Performance as a Function of Descriptor Types and Model Parameters

Different types of descriptors were explored for model development: (i) CDK atomic descriptors in analogy to those used in FAME (33) (“CDK”), (ii) atom type fingerprints (“ATF”) used in various approaches, (42-44) (iii) quantum chemical (“QC”) descriptors, and (iv) combinations thereof (Table 2). Additionally, these descriptors were used to derive circular representations of atom environments with a depth of up to 6 bonds (Figure 2), giving rise to a total of 32 different models.

Table 2. Various Descriptor Types Investigated in this Study

class	groupa	description
2Db	CDK	This group comprises 15 basic 2D descriptors implemented in CDK. Those are the original 15 descriptors considered in FAME, and they encode various properties of individual atoms in the molecule (Table S1).
	circCDK	This contains circular descriptors derived from the 15 CDK descriptors. In other words, it is equal to the CDK set enriched by the circular versions of the CDK descriptors of bond depth 1–6.
	CDK + ATF	This is a combination of the 15 basic CDK descriptors and the circular atom type fingerprints. The fingerprints represent occurrence counts of various atom types within a certain distance from the considered atom.
	circCDK + ATF	This group includes atom type circular fingerprints in addition to all descriptors from the circCDK set.
QC-enhancedc	CDK + QC	This is the simplest quantum chemistry (QC) enhanced set. It represents the complete set of noncircular atomic descriptors used in the current study (Table S1 and Table 3).
	CDK + circQC	This combination comprises the 15 basic CDK descriptors that are combined with circular versions of the 10 quantum chemical descriptors. In other words, this set contains the CDK + QC set and the circular versions of the QC descriptors for bond depths of 1–6.
	circCDK + ATF + circQC	These represent the combination of all descriptors investigated in the current study.

Codes representing the descriptor groups explored in the current study.

Simple 2D descriptor sets that only use topological information and their circular counterparts.

Descriptor sets enhanced by quantum chemical descriptors.

Table 3. Quantum Chemical Descriptors Calculated by MOPAC (45, 46)

name	description
piS(r)	atom self-polarizability (47, 48)
De(r)	electrophilic delocalizabilities of an atom (47, 49)
Dn(r)	nucleophilic delocalizabilities of an atom (47, 49)
s-Pop	population of the s-orbital (i.e., formal electron density of the s-orbital)
p-Pop	population of the p-orbitals (i.e., formal electron density of the p-orbitals)
NumOfElecs	number of valence electrons localized on the atom (i.e., electron density)
NetCharge	net charge (i.e., formal number of valence electrons–NumOfElecs)
valence	sum of bonds for an atom (i.e., molecular orbital valency) (50)
mull_charge	partial charge of an atom determined by Mulliken (51, 52)
mull_pop	electron population of an atom determined by Mulliken (51, 52)

Figure 2. Example showing how circular atom type fingerprints and descriptors of up to bond depth 3 were calculated for a single atom. In this example, neighboring atoms up to three bonds away from the sp³ hybridized oxygen were encoded. Two examples of data matrix parts corresponding to neighbors encoded in bond depths 0 and 3 are shown. In both the fingerprints and descriptors, atoms were grouped by their atom type (highlighted in blue in the column names) in each bond depth (highlighted in orange). In the case of the fingerprints, the occurrences of each atom type were noted, while for real-valued descriptors, the average value of the basic descriptor among the grouped atoms was calculated and assigned to the corresponding atom type. Therefore, for the *sigmaElectronegativity* descriptor in this example, the third layer would contain 8.35 (8.35/1 = 8.35) for the sp³ hybridized nitrogen type and 8.31 ((8.31 + 8.31)/2 = 8.31) for the aromatic carbon type. Consequently, what can be derived from this matrix fragment is that there are two aromatic carbons (which are topologically identical) within three bonds of the oxygen atom that have a *sigmaElectronegativity* equal to 8.31 each.

Optimized Model Parameters

For each model, the decision_threshold, max_features, class_weight, and maximum number of features allowed in the model (Table 4) were optimized using a 10-fold cross-validated grid search on the training data set (a subset of the original data set that contained data on 542 molecules; see Methods for details). For this optimization, the MCC was used as a scoring function and the model with the highest average MCC over all folds was selected as optimal.

Table 4. Hyperparameter Grid

model parameter	explored values
decision_threshold	0.2, 0.4, 0.5, 0.6
class_weight	balanced, balanced_subsample, none
max_features	sqrt, 0.3, 0.6, 0.9
max_features_ANOVA	100, 150, 200, 250, 300, 350, 400

Decision Threshold and Class Weights

Adjusting the decision threshold provides a means to affect the trade-off between false-positives and false-negatives. This adjustment can remove the bias of the models toward the overrepresented class (non-SoMs) by lowering the percentage of estimators in the ensemble that are needed to make a decision that an atom is a SoM. The same strategy was also employed by Finkelmann et al. (36) to balance their models.

In addition to the decision threshold, the class balancing parameter of the classifier was optimized. This parameter adjusts the weight of the SoM and non-SoM classes during model building to compensate for the imbalance of the two classes in the training set. The weights are set to be inversely proportional to the number of samples available for each particular class in the training data.

A clear preference of decision thresholds lower than the default value of 0.5 was observed almost uniformly across all models (Table S2). The most commonly selected value was 0.4, but even values as low as 0.2 were sometimes selected. Especially low decision threshold values were observed for the models built using atom type fingerprints as the only circular descriptor (“CDK + ATF”). The CDK + ATF model was also the only one that made use of class balancing uniformly across all experiments. Therefore, one could argue that such models were probably more susceptible to class imbalance than others and that it was thus necessary to compensate for the influence of the negative class by lowering the decision threshold and adjusting the weights of the classes during model building.

Maximum Number of Considered Features per Split

The optimal maximum percentage of features allowed to be sampled at each tree branch varied, but the most commonly selected value of the max_features parameter was 0.9, meaning that at most 90% percent of features were sampled (Table S2). Usually, the more features were available in total, the higher was the percentage of features required to be sampled when looking for a split. This was probably due to the fact that circular descriptors introduced more noise to the data and, therefore, a more extensive sampling of features was needed at each step. However, the CDK + ATF model was again an exception here. It seems that sampling more than 30% of features per split mostly did not provide any benefit in this case. This might have been a result of encoding less noisy information that covers only shallow relationships in the data (just atom type occurrences). There were also much fewer features overall in these models, which might have further contributed to the observed trend.

ANOVA F-Test

Prior to modeling, the number of all features was reduced using a univariate ANOVA F-test on each variable. No more than 400 features per model were allowed in our experiments because a greater number of features increases model complexity and, thus, the variance and risk of overfitting.

The median number of features per model was 275, and it was observed that the maximum number of features selected for modeling using the ANOVA F-test increased with the maximum bond depth of circular descriptors (Table S2). This behavior could be expected since encoding a wider neighborhood of an atom must mean that more detailed information is available about each potential SoM.

Influence of Descriptors on Model Performance During Training

Good models were obtained with each of the seven explored descriptor sets. The MCC values obtained for the models during cross-validation ranged from 0.51 to 0.59, whereas the Top-2 prediction rates ranged from 84.1 to 88.4% (all results provided in Table S3, including AUC values). The best models included the circCDK model at a bond depth of 6 (MCC 0.59; Top-2 86.9%) and the circCDK + ATF + circQC model at a bond depth of 2 (MCC 0.58; Top-2 88.4%).

Models based on 2D descriptors gradually improved in MCC with increasing bond depth of circular descriptors during cross-validation while QC-enhanced models did not show any growth after bond depth 1 (Figure 3a). Other metrics generally correlated well with the MCC. The only exception was the CDK + circQC model, for which there was no apparent correlation between the MCC, Top-2, and Top-3. However, there was a good correlation between the MCC and the AUC for all models (Pearson correlation coefficient of 0.92). Therefore, the further discussion will mostly focus on the MCC and only highlight other metrics where necessary.

Figure 3. Visualization of the relationship between maximum bond depth of circular descriptors and model performance on the (a) training set (measured as the mean MCC across the 10 folds in cross-validation of the optimized model), (b) independent test set, (c) uncorrelated test set, and (d) full Zaretzki data set (measured as the mean MCC across the 10 folds in cross-validation of the optimized model). Models that do not use circular descriptors are shown to have a bond depth equal to 0.

The faster increase in accuracy and the early performance plateau of the QC-enhanced models could be interpreted as a result of having the 3D conformation of the molecule available and the fact that QC calculations give a much more precise description of the electron density around each atom. If the 3D conformation is available and the electron density is, therefore, more accurately described, more detailed properties of each atom can be calculated and more detailed information about its neighborhood is already implicitly included in the description of the atom itself. Hence, encoding a wider neighborhood of a potential SoM did not result in substantial improvements when quantum chemical descriptors were used, but it was necessary in the case of the 2D descriptors to reach comparable performance. Therefore, the QC approach, albeit computationally more costly, may have an advantage in that good model performance can be obtained with fewer features.

One more important conclusion that can be drawn from Figure 3a is that when sufficient area around a potential SoM is encoded using a purely 2D methodology, the resulting models have comparable performance to those enhanced by quantum chemical calculations.

Model Performance on an Independent Test Set

In order to assess the generalization of the models and their ability to perform in practice, an external validation set of 136 compounds was randomly drawn from the whole data set prior to any modeling and cross-validation experiments. The resulting scores obtained by each model on this independent test set (Figure 3b and Table 5) suggest similar trends to those observed during cross-validation (Figure 3a and Table S3), with the exception of lower performance improvement of the circular QC-enhanced models relative to the basic CDK + QC methodology. Also in contrast to the cross-validation results, the combined circCDK + ATF models seem to show greater performance improvement in comparison to the basic CDK method but with no clear increase in MCC beyond bond depth 1.

Table 5. Model Performance on an Independent Test Seta

^{Table a}

Model performance is indicated by a color gradient, ranging from dark red (worst performance among all models) via white to dark green (best performance among all models).

^{Table b}

Model code according to the descriptor set (Table 2) and maximum bond depth (indicated by a number in brackets) used.

^{Table c}

Value of the given performance metric calculated using predictions of 136 compounds in the independent test set.

For a more robust comparison of the performance of different models, 100 random subsamples comprised of 50 molecules each were selected (with replacement) from the original test set and treated as separate external sets. Therefore, each subsample represented a possible classification problem of fixed size but varying difficulty.

In order to determine which models performed statistically differently from others, the MCC scores obtained on the subsamples were subjected to both parametric (repeated measures ANOVA as an omnibus test and paired t tests with p-values corrected according to Holm (53) for posthoc analysis) and nonparametric (Friedman test (54-56) as an omnibus test and the Nemenyi test for posthoc analysis) statistical tests. We concluded that the use of parametric tests was possible in our case since neither visual observation of Q–Q plots nor the Shapiro–Wilk test (57) highlighted any issues with the underlying normality assumption. The data only violated the sphericity condition of the repeated measures ANOVA test (determined by the Mauchly test (58)). Therefore, the correction methods of Greenhouse–Geisser (59) and Huynh–Feldt (60) were employed to obtain the final p-values.

A significant statistical difference between at least two models was confirmed by both the repeated measures ANOVA with the aforementioned corrections and the Friedman omnibus tests (all p-values were lower than 2.2 × 10^–16). There was also a good agreement between the parametric and nonparametric posthoc tests regarding the rejection of the underlying null hypotheses (Figure 4). The two approaches differed in 52 out of all 496 pairwise comparisons conducted. This could be attributed to the Holm correction which is more conservative than the Nemenyi test. However, these differences do not influence the main results of our study.

Figure 4. Consensus matrix combining results from parametric and nonparametric posthoc tests. The pairs of models for which the null hypothesis (the equality of means for t test or similar performance ranking for Nemenyi test) was rejected at an α level of 0.05 by both the parametric and nonparametric method are highlighted in teal.

As shown in Figure 5, most models were positively affected by the bond depth of the circular descriptors. This effect is most apparent for the models based on 2D descriptors for which the performance gradually increased with bond depth, and most 2D models with six layers significantly outperformed their simpler counterparts. The only exception was the circCDK + ATF model with bond depth 1 that showed comparable performance to the more complex models.

Figure 5. Point plot which represents the expected performance of each model by indicating 95% confidence intervals for mean MCC. The confidence intervals were obtained from a statistical analysis of predictions of 100 random subsamples (50 molecules each, sampled with replacement) of the original test set.

The QC-enhanced models were only modestly improved by circular descriptors, with only six of them statistically significantly outperforming the simplest CDK + QC model. Additionally, no QC-enhanced model showed significantly better performance than any of the top scoring 2D models (see the circCDK + ATF models with bond depth 1, 5, and 6 and the circCDK models with bond depth 5 and 6, for instance).

Therefore, in terms of performance, there was no model that showed a significantly better success rate than all others. However, it should be noted that these results favor the use of 2D approaches in practice since they are considerably easier and faster to both use and implement. Simpler approaches are also likely to be robust and easier to interpret. Therefore, the circCDK + ATF model with maximum bond depth of 1 seems to offer a good trade-off between accuracy and robustness. Even though the wider atomic environment was not considered in this case, the model was still able to perform well on data not used to develop the model. This is probably due to the fact that only the most important information needed to describe the reactive center, that is the types, counts, and continuous properties of the immediate neighbors, was encoded. Thus, the circCDK + ATF model with maximum bond depth of 1 seems to be a good choice for use in practice.

The subsampling experiments outlined above were also conducted using the random forest algorithm rather than extra trees. In these random forest experiments, the same parameters, data samples and random states were used as already outlined above. The only difference was in the used algorithm and that only 2D models were trained. In vast majority of the cases, using the random forest algorithm did not show any advantage over extra trees and there were a few cases where its average performance over the subsamples was slightly lower (compare Figure S1 and Figure 5).

Considering the Top-2 score, there was a positive difference in performance between the test set and the cross-validation average. The average difference between the cross-validation means and the test set scores was +6.5%, with a standard deviation of 1.3%. Though the performance was generally better on the test set in terms of the Top-2 metric, this was not the case for the MCC, for which the performance mostly remained the same with a difference of only −0.02 (standard deviation 0.01). This difference might have occurred because the whole training set of 542 compounds was used to train the final models rather than just 90% of the training set during cross-validation. Since Top-2 only depends on the first two predictions and is only calculated on a per-molecule basis, it might have been more sensitive to this effect.

Model Performance in a Structurally Uncorrelated Test Set

To investigate the structural correlation between the test set and the training set, an analysis of molecular similarity was conducted using MACCS keys as structural fingerprints and Tanimoto similarity. It was found that the median maximum similarity between the compounds from the testing set and the training set was 0.78, which hints at possible structural correlation of some compounds in the test set with compounds in the training set. This was also confirmed by plotting the histogram of maximum similarities shown in Figure 6a. In total, there were 65 compounds in the test set with a similarity to the closest compound in the training set higher than or equal to 0.8.

Figure 6. Histogram of maximum Tanimoto similarities (a) between compounds in the training set and the independent test set (136 molecules) and (b) between compounds within the uncorrelated test set (71 molecules).

Therefore, a second (structurally uncorrelated) test set was constructed as a subset of the original test data. This second test set comprised of 71 compounds that showed maximum similarity to any of the compounds in the training data lower than 0.80. This second test set had median maximum similarity of 0.70 relative to the training set, and the mean maximum similarity within this test set was 0.55 with a standard deviation of 0.17 (Figure 6b). Therefore, the second test set contained molecules that were much less similar to the training data and also represented a rather diverse set of compounds.

In the case of the uncorrelated test set, the MCC changed by −0.09 on average relative to the mean from cross-validation, but the Top-2 remained very high with an average difference of +4% (Table S4). Therefore, even in the uncorrelated test set, the Top-2 performance had generally improved relative to the mean from cross-validation.

The fact that Top-2 did not decrease for the independent and the uncorrelated test sets might also be related to the nature of the metric itself. Top-2 results in a correctly predicted molecule whenever at least one true SoM is within the top two ranked atoms, which not only is a rather loose criterion but also fails to capture the effect of false positives and, to some extent, false negative predictions (unlike the MCC). Therefore, this result might also hint at the unsuitability of the Top-2 metric to correctly reflect overall classification performance, for which the MCC is much more robust and reliable.

However, it should also be noted that the observed Top-2 values show that the best models were able to correctly predict at least one true SoM within the two top ranked positions in more than 90% of the molecules in both test sets. Since false negative results are usually more prevalent in SoM predictions (because of the imbalance of the classes), this could mean that the most obvious SoMs were still correctly identified by the models regardless of structural similarity of the test compounds to the training set.

Just like in the cross-validation experiments, the MCC obtained for models based on 2D descriptors generally improved with increasing bond depth (Figure 3c). However, this was not the case for the circCDK + ATF model, for which the performance seems to have decreased slightly overall.

For QC-enhanced models, no general improvement in the MCC with increasing bond depth was observed. Most of these models scored in the same range as the basic CDK + QC model. Moreover, the fact that the simple circCDK + ATF model with bond depth 1 also scored highly on the uncorrelated test set suggests that this model is fairly robust and will probably perform well in practical applications.

Comparison with Other Methods

Two of the latest and most accurate models for SoM prediction are Xenosite (31) and the models reported by Finkelmann et al. (36) Both approaches are based on the Zaretzki data set. While in the original publications the performance of XenoSite was evaluated by leave-one-out experiments using the Top-2 measure, the performance of the Finkelmann models was assessed using 10-fold cross-validation across the Zaretzki data set (and, in addition, was tested on a set of 25 undisclosed compounds from Bayer drug discovery projects). We followed both evaluation protocols as closely as possible to allow direct comparison of these models with ours. In order to compare the performance of our models with those presented by Finkelmann et al., we performed 10-fold cross-validation with hyper-parameter optimization on the full Zaretzki data set. For comparison of our models with Xenosite, we conducted experiments with leave-one-out cross-validation.

Performance Comparison with the Finkelmann Models

In general, the performance of our models and the Finkelmann models was comparable. Their final model (i.e., the best out of 10 random reruns) obtained an average MCC of 0.63 across all folds with a standard deviation of 0.06. They also devised a simpler model with lower variance by employing ANOVA F-test feature selection. This model performed with an average MCC of 0.59 and a standard deviation of 0.03. The result is comparable with the majority of the models developed in the current study.

Although our favorite simple model (the circCDK + ATF model with a bond depth of 1) obtained a lower average MCC score of 0.56 with a standard deviation of 0.03, its six layer version reached a performance comparable to the Finkelmann models, with an MCC of 0.59 and a standard deviation of 0.03 (Table S5 and Figure 3d). However, it should be noted that the mean score from the cross-validation was used to optimize the hyper-parameters of the models and, thus, a slight overtraining might have occurred which is more likely to happen for more complex models. Therefore, we still encourage the use of the simplest circCDK + ATF model in practice since it is expected to be less susceptible to such phenomena (see previous sections).

These findings indicate that simple 2D approaches can be just as effective in SoM prediction as more complex models based on descriptors derived from semiempirical or even ab initio quantum chemical calculations. Therefore, at the moment, there might be no need to work with a 3D representation of the molecule and calculations can be done much faster. Thanks to their simplicity, the resulting 2D models are also expected to be more robust and less prone to overfitting. The simplest circCDK + ATF model showed especially good generalization ability since it showed comparable or better performance than more complex QC-enhanced models on both test sets (Figure 3b and c) and in the resampling experiments (Figures 4 and 5). Additionally, the simple 2D models developed herein can be used without relying on a number of (often proprietary) software packages and applications, making usability and distribution to the scientific community easy and straightforward.

Performance Comparison with Xenosite

Most of the models presented herein showed comparable performance to both XenoSite and the models presented by Finkelmann et al. (36) that reached leave-one-out Top-2 accuracy of 89.4% and 90.9%, respectively. The average Top-2 accuracy of our models was between 86.3% for the simplest CDK model and 90.3% for the CDK + QC model (Table S6). The best 2D models were the circCDK and circCDK + ATF models with maximum Top-2 equal to 89.4%. The favored circCDK + ATF model with bond depth 1 again scored slightly lower with Top-2 of 88.5%, but this difference is, in our opinion, almost negligible.

It should also be noted that we report results from one run with a fixed random seed, while Finkelmann et al. (36) and Zaretzki et al. (31) reported the best model out of multiple random reruns. Additionally, the two other methods used the full set of descriptors while we always limited their number to no more than 400 using an ANOVA F-test prior to modeling, since we used the same set of optimal hyper-parameters as determined by 10-fold cross-validation on the whole data set of 678 compounds.

Isoform-Specific Models

The same workflow outlined in Figure 1 was employed to obtain isoform-specific models for CYP 3A4, 2D6, and 2C9. The original data set did not contain SoM annotation for all isoforms and all compounds. Therefore, an appropriate subset of the original data was identified for each isoform (Table 1 for composition of each isoform-specific set).

The prediction of the SoMs of CYP 3A4 substrates proved to be more challenging than for the other two CYP isoforms. The best-performing model during cross-validation was the CDK + circQC model with circular descriptors of up to bond depth 3 with an MCC of 0.55 (Table 6). The best performance on the uncorrelated test set was obtained with circCDK and circCDK + ATF models with a maximum circular descriptor bond depth of 2. Both models reached MCC values as high as 0.49.

Table 6. Model Performance of Isoform-Specific Modelsa

^{Table a}

Model performance is indicated by a color gradient, ranging from dark red (worst performance among all models) via white to dark green (best performance among all models).

^{Table b}

Model code according to the descriptor set (Table 2) and maximum bond depth (indicated by a number in brackets) used.

^{Table c}

Isoform code.

^{Table d}

Average value of the given performance metric across ten cross-validation folds on the training set.

^{Table e}

Value of the given performance metric as calculated using predictions on the independent test set.

^{Table f}

Value of the given performance metric as calculated using predictions on the uncorrelated test set.

Models built for CYP 2D6 had the best performance among the three different isoforms, with an average MCC of 0.55. The best model in cross-validation was the CDK + circQC model with a maximum circular descriptor bond depth of 2 that obtained an MCC of 0.59. However, the CYP 2D6 models also showed the biggest difference in MCC between the cross-validation and the uncorrelated test set (−0.09 on average). The best performing models on the uncorrelated test set were the circCDK model with circular descriptors up to bond depth 4 and the CDK + ATF model with a maximum bond depth of 6.

The CYP 2C9 models showed the lowest mean model performance during the 10-fold cross-validation on the training set (mean MCC of 0.50 across the averages from folds) but the best performance on the uncorrelated test set. The best-performing models during the cross-validation were the CDK + circQC models with maximum circular bond depths of 2 and 6 and an MCC of 0.54. The best model in the uncorrelated test set was the CDK + ATF model with maximum circular descriptor bond depth of 5, which reached an MCC as high as 0.59. Therefore, even though the models did not perform as well during cross-validation, they were still able to generalize well.

It should be pointed out that also in the case of the isoforms, the best performing models in the uncorrelated test set were based on 2D descriptors. This outcome suggests that using a 2D method to encode molecular properties is indeed a robust approach that results in models with good generalization ability. However, it should also be noted that the isoform-specific models often showed lower overall performance in comparison to the global model, which is most likely related to the lack of training data (particularly in the case of the CYP 2D6 and 2C9 isoforms). Therefore, it is expected that such models could greatly improve if more experimental data were available.

The choice of the best model also becomes more difficult in this case, since with less data in the test sets the variability of the score increases, making a comparison more difficult. Nevertheless, for the CYP 3A4 isoform, the circCDK and circCDK + ATF models with a bond depth of 2 seem to have been most successful on both test sets. The success of simpler models like these might again be attributed to their robustness and simplicity, but also to the fact that for CYP 3A4 reactivity is a more important factor than the shape of the substrate and, thus, the immediate neighbors of the SoM are more influential. This conclusion also seems to be reinforced with the results obtained for CYP 2C9 and 2D6 for which the best performing model on both test sets was the CDK + ATF model with a bond depth of 6. Since these isoforms are expected to be more selective regarding their substrates, encoding a wider area may be necessary to include the influence of the other interacting atoms.

Computation Time

Average execution time for the predictor was measured for the full data set of 678 compounds. A workstation equipped with an Intel Core i5 processor (i5-6200U), 8 GB of RAM, and a Linux operating system was used to make predictions using the circCDK + ATF model with a maximum bond depth of 6, since this was the most complex 2D model. The median execution time per compound was 265 ms, with only 24 compounds taking longer than 5 s to compute. Therefore, very fast predictions were possible for the vast majority of compounds.

Software Package Description

A Java software package named FAME 2 was developed to make the 2D models that are likely to be most successful in practice available for anyone to use. FAME 2 contains the 2D model that showed the best average performance during test set resampling (circCDK + ATF with bond depth 6) and the simplest 2D models that were not found to perform significantly differently than the best model (the circCDK model with bond depth 4 and the circCDK + ATF model with bond depth 1). These final global metabolism models were developed using the whole data set and 10-fold cross-validation to optimize the parameters. The software uses the simplest model (circCDK + ATF with bond depth 1) by default, but it also enables the user to easily select any of the two other models. The software package is available from the authors free of charge.

FAME 2 uses a slightly modified version of the visualization developed by Patrik Rydberg and implemented in SMARTCyp. (25) The predictions are generated as a simple HTML page which displays the structure of the compound with the predicted SoMs highlighted with yellow circles (Figure 7). The software also displays the probabilities generated by the extra trees model (the percentage of trees that predicted a given atom as a SoM) and highlights the atoms that scored above the decision threshold of the current model. Some metadata and helpful information as well as the model used for the predictions are also displayed.

Figure 7. Example of prediction visualization by FAME 2.

Conclusions

ARTICLE SECTIONS

Jump To

Current models for SoM prediction often rely on complex sets of (quantum) chemical descriptors and accurate representation of substrate conformations. In this work we explored the use of extremely randomized trees for modeling the regioselectivity of cytochrome P450 enzymes and showed that simple, circular descriptions of atoms and their neighborhood with 2D descriptors can yield models that are robust and just as accurate as more complex approaches. This conclusion is supported by a thorough analysis of model performance during which we also postulate that the simplest one layer model based on 2D circular CDK descriptors and atom type fingerprints (model “circCDK + ATF [1]”) is most suitable to use in practice since it showed good (or comparable) generalization ability in comparison to other models (MCC of 0.57 on an independent test set).

The method presented here is simple and effective, and we distribute the models that we found most likely to perform well in practice in the form of an easy-to-use software package named FAME 2. This will also allow researchers to use these models with confidential data as no transmission of such to external sources is required (as in the case of the many available web services).

It appears that with the data on enzyme regioselectivity available at present we are approaching a plateau with respect to the accuracy and applicability of models. It is expected that with the availability of additional data in the future the performance of SoM predictors can be further improved and their applicability extended.

Methods

ARTICLE SECTIONS

Jump To

Data Set

The study is based on a revised version of the Zaretzki data set (31) published in ref 61. The main difference between the revised and the original version of the data set is a more accurate annotation of stereochemical information for chiral molecules. The general workflow of the experiments in this study is outlined in Figure 1. All stereoisomers of molecules with at least one undefined stereocenter were enumerated with the flipper option in OMEGA (62, 63) (v2.5.1.4), and the conformation of each isomer optimized with OMEGA (default settings, with maxconfs = 1 and canonOrder = true in order to keep SoM annotation consistent with the atom sequence). From all possible stereoisomers generated this way, one was selected at random and used throughout all experiments. The preprocessed molecules were then used to calculate various atomic descriptors (see Descriptors).

The final data set contained information about 678 molecules. Two molecules of the original set (31) could not be processed due to missing parameters in MOPAC for the boron atom in bortezomib and a failure of CDK to correctly assign the atom type of the nitrogen atom in the nitro group of imidacloprid.

Topologically symmetrical atoms were considered to be equivalent when purely 2D descriptors were used for modeling. For the calculation of quantum chemical descriptors, symmetrical atoms were all assigned one probability value when making the predictions, which was the maximum probability calculated by the model for potential SoMs in one particular symmetry class.

Finally, the molecules were randomly assigned into a training and testing set. After the split, the test set contained 136 molecules (20% of the original data), while the remaining 542 compounds were designated for training. The initial test set was further reduced to 71 compounds by applying a similarity filter that only retained the molecules with maximum structural similarity to any of the compounds in the training set lower than 0.8. MACCS keys and Tanimoto similarity as implemented in RDKit (64) were used for compound representation and similarity calculation.

Descriptors

Various sets of atomic descriptors were investigated in this work (see Table 2 for an overview of descriptor types and abbreviations). The most basic set of descriptors was comprised of the same set of 15 atomic descriptors investigated in FAME (33) and implemented in CDK (65, 66) (Table S1). In addition to these simple 2D descriptors, ten quantum chemical descriptors were extracted from geometry optimization calculations by MOPAC (45, 46) with the semiempirical Austin Model 1 (AM1), which is based on the Neglect of Differential Diatomic Overlap (NDDO) with the VSTO-3G basis set (standard basis set in MOPAC). The quantum chemical descriptors in Table 3 were used (generated by the keywords AM1, XYZ, MMOK, VECTORS, BONDS, PI, PRECISE, ENPART, EF, MULLIK, and SUPER).

Furthermore, circular atom type fingerprints were calculated for each potential SoM as described by Tyzack et al. (44) In general, the fingerprints encode how many atoms of a particular type are within a certain topological distance (i.e., number of bonds) from a putative SoM. The AtomType descriptor was used to group the neighboring atoms by their types. Therefore, each layer of the fingerprint represents occurrences of Sybyl atom types that lie a certain number of bonds away from the putative SoM in question (Figure 2). In our experiments, fingerprints of up to bond depth 6 were used.

A more detailed set of circular descriptors was also calculated using both the CDK descriptors and the quantum chemical descriptors described above. These descriptors were generated in a similar fashion as the fingerprints, but the actual descriptor values (instead of simple atom type occurrence counts) were encoded in each layer (Figure 2). To be more precise, the atoms in each layer were grouped by atom types and the values of a particular descriptor calculated for all atoms in a group were averaged and assigned to the corresponding atom type. The rest of the atom types then had an unspecified value attached in this layer. This process was repeated for every atomic descriptor and all atoms up to a distance of six bonds from a possible SoM were considered as neighbors and encoded in this manner.

Descriptor Selection

In order to reduce model complexity, a univariate ANOVA F-test was conducted before each model was built. The maximum number of best features to retain was optimized using a cross-validated grid search on the training data using the MCC (described in the Results section) as a scoring function. Additionally, all variables with less than three unique values (including an unspecified value) were also removed prior to modeling.

Value Imputation

In order to replace the unspecified values in circular descriptors, a simple imputation strategy was implemented according to

(2)where i_d is the imputed value of descriptor d while

is the mean of all numerical values of descriptor d and s_d is their standard deviation. The purpose of the factor 10 in the formula was to ensure (with reasonable confidence) that the imputed values lie outside the expected range of the known numeric values. The imputed values for the test set were equal to those determined from the training set according to eq 2.

Model Building

An extra trees classifier implementation from the machine learning Python package scikit-learn (v0.18) (67) was used to build all models in the present study. In all experiments, bootstrap sampling and the out-of-bag error estimate were used during training. Each model ensemble contained 500 estimators. All other hyperparameters were kept at their default values during training, but a 10-fold cross-validated grid search was conducted to optimize some of the most important ones. The grid shown in Table 4 was used for parameter optimization.

The decision_threshold parameter denotes the minimum probability needed to decide whether an atom is categorized as a SoM or non-SoM when the atom positions in a predicted molecule are classified. The max_features parameter is the maximum number of available features to consider when looking for an optimum split. The value “sqrt” means that the number was set as the square root of all available features. The class_weight parameter adjusts the weight of the two atom classes in the data set (SoM vs non-SoM) during model building to compensate for the imbalance in the training set. The weights are inversely proportional to class frequencies and can be either set using global frequencies in the whole training data (“balanced”) or frequencies in the bootstrap sample of each tree (“balanced_subsample”). The value of “none” means that class balancing was turned off (all classes had the same weight). The max_features_ANOVA parameter is the maximum number of best features to select for modeling according to the ANOVA F-test. The original number of features was retained if it did not exceed 100.

Supporting Information

ARTICLE SECTIONS

Jump To

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jcim.7b00250.

Additional figures and tables with calculated CDK descriptors, hyperparameter optimization results, model validation results, and performance of the random forest algorithm (PDF)

ci7b00250_si_001.pdf (684.23 kb)

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

ARTICLE SECTIONS

Jump To

Corresponding Author
- Johannes Kirchmair - Faculty of Mathematics, Informatics and Natural Sciences, Department of Computer Science, Center for Bioinformatics, Universität Hamburg, Hamburg, 20146, Germany; http://orcid.org/0000-0003-2667-5877; Email: [email protected]
Authors
- Martin Šícho - Faculty of Mathematics, Informatics and Natural Sciences, Department of Computer Science, Center for Bioinformatics, Universität Hamburg, Hamburg, 20146, Germany; CZ-OPENSCREEN: National Infrastructure for Chemical Biology, Laboratory of Informatics and Chemistry, Faculty of Chemical Technology, University of Chemistry and Technology Prague, 166 28 Prague 6, Czech Republic; http://orcid.org/0000-0002-8771-1731
- Christina de Bruyn Kops - Faculty of Mathematics, Informatics and Natural Sciences, Department of Computer Science, Center for Bioinformatics, Universität Hamburg, Hamburg, 20146, Germany; http://orcid.org/0000-0001-8890-2137
- Conrad Stork - Faculty of Mathematics, Informatics and Natural Sciences, Department of Computer Science, Center for Bioinformatics, Universität Hamburg, Hamburg, 20146, Germany; http://orcid.org/0000-0002-5499-742X
- Daniel Svozil - CZ-OPENSCREEN: National Infrastructure for Chemical Biology, Laboratory of Informatics and Chemistry, Faculty of Chemical Technology, University of Chemistry and Technology Prague, 166 28 Prague 6, Czech Republic; http://orcid.org/0000-0003-2577-5163
Funding
This work was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation)—project number KI 2085/1-1—and by the Ministry of Education of the Czech Republic—project numbers MSMT No 20-SVV/2017, NPU I-LO1220, and LM2015063. M.S. was supported by the Erasmus+ Programme of the European Commission.
Notes
The authors declare no competing financial interest.

Acknowledgment

ARTICLE SECTIONS

Jump To

Access to computing and storage facilities owned by parties and projects contributing to the National Grid Infrastructure MetaCentrum provided under the programme “Projects of Large Research, Development, and Innovations Infrastructures” (CESNET LM2015042), is greatly appreciated. OpenEye Scientific Software is acknowledged for providing licenses for OMEGA via the Free Academic Licensing Program.

List of Abbreviations

ARTICLE SECTIONS

Jump To

AM1	Austin Model 1
ATF	atom type fingerprints
AUC	area under the ROC curve
CYP	cytochrome P450 enzyme
DFT	density functional theory
MCC	Matthews correlation coefficient
NDDO	Neglect of Differential Diatomic Overlap
QC	quantum chemistry
SOM	site of metabolism

References

ARTICLE SECTIONS

Jump To

This article references 67 other publications.

1
Kirchmair, J.; Howlett, A.; Peironcely, J. E.; Murrell, D. S.; Williamson, M. J.; Adams, S. E.; Hankemeier, T.; van Buren, L.; Duchateau, G.; Klaffke, W.; Glen, R. C. How Do Metabolites Differ from Their Parent Molecules and How Are They Excreted? J. Chem. Inf. Model. 2013, 53, 354– 367 DOI: 10.1021/ci300487z
[ACS Full Text ], [CAS], Google Scholar
1
How Do Metabolites Differ from Their Parent Molecules and How Are They Excreted?
Kirchmair, Johannes; Howlett, Andrew; Peironcely, Julio E.; Murrell, Daniel S.; Williamson, Mark J.; Adams, Samuel E.; Hankemeier, Thomas; van Buren, Leo; Duchateau, Guus; Klaffke, Werner; Glen, Robert C.
Journal of Chemical Information and Modeling (2013), 53 (2), 354-367CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
Understanding which physicochem. properties, or property distributions, are favorable for successful design and development of drugs, nutritional supplements, cosmetics, and agrochems. is of great importance. In this study the authors have analyzed mols. from three distinct chem. spaces (i) approved drugs, (ii) human metabolites, and (iii) traditional Chinese medicine (TCM) to investigate four aspects detg. the disposition of small org. mols. First, the authors examd. the physicochem. properties of these three classes of mols. and identified characteristic features resulting from their distinctive biol. functions. For example, human metabolites and TCM mols. can be larger and more hydrophobic than drugs, which makes them less likely to cross membranes. The authors then quantified the shifts in physicochem. property space induced by metab. from a holistic perspective by analyzing a data set of several thousand exptl. obsd. metabolic trees. Results show how the metabolic system aims to retain nutrients/micronutrients while facilitating a rapid elimination of xenobiotics. In the third part the authors compared these global shifts with the contributions made by individual metabolic reactions. For better resoln., all reactions were classified into phase I and phase II biotransformations. Interestingly, not all metabolic reactions lead to more hydrophilic mols. The authors were able to identify biotransformations leading to an increase of logP by more than one log unit, which could be used for the design of drugs with enhanced efficacy. The study closes with the anal. of the physicochem. properties of metabolites found in the bile, feces, and urine. Metabolites in the bile can be large and are often neg. charged. Mols. with mol. wt. >500 Da are rarely found in the urine, and most of these large mols. are charged phase II conjugates.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3sXhsVemtrY%253D&md5=ea563b68312736c7736b40939d7620a6
2
Testa, B.; Pedretti, A.; Vistoli, G. Reactions and Enzymes in the Metabolism of Drugs and Other Xenobiotics Drug Discovery Today 2012, 17, 549– 560 DOI: 10.1016/j.drudis.2012.01.017
[Crossref], [PubMed], [CAS], Google Scholar
2
Reactions and enzymes in the metabolism of drugs and other xenobiotics
Testa, Bernard; Pedretti, Alessandro; Vistoli, Giulio
Drug Discovery Today (2012), 17 (11-12), 549-560CODEN: DDTOFS; ISSN:1359-6446. (Elsevier Ltd.)
In this article, we offer an overview of the compared quant. importance of biotransformation reactions in the metab. of drugs and other xenobiotics, based on a meta-anal. of current research interests. Also, we assess the relative significance the enzyme (super)families or categories catalyzing these reactions. We put the facts unveiled by the anal. into a drug discovery context and draw some implications. The results confirm the primary role of cytochrome P 450-catalyzed oxidns. and UDP-glucuronosyl-catalyzed glucuronidations, but they also document the marked significance of several other reactions. Thus, there is a need for several drug discovery scientists to better grasp the variety of drug metab. reactions and enzymes and their consequences.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC38XnslOiur8%253D&md5=e11b5b1b7ea18e8bc817bff83efad456
3
Kirchmair, J.; Göller, A. H.; Lang, D.; Kunze, J.; Testa, B.; Wilson, I. D.; Glen, R. C.; Schneider, G. Predicting Drug Metabolism: Experiment and/or Computation? Nat. Rev. Drug Discovery 2015, 14, 387– 404 DOI: 10.1038/nrd4581
[Crossref], [PubMed], [CAS], Google Scholar
3
Predicting drug metabolism: experiment and/or computation?
Kirchmair, Johannes; Goeller, Andreas H.; Lang, Dieter; Kunze, Jens; Testa, Bernard; Wilson, Ian D.; Glen, Robert C.; Schneider, Gisbert
Nature Reviews Drug Discovery (2015), 14 (6), 387-404CODEN: NRDDAG; ISSN:1474-1776. (Nature Publishing Group)
A review. Drug metab. can produce metabolites with physicochem. and pharmacol. properties that differ substantially from those of the parent drug, and consequently has important implications for both drug safety and efficacy. To reduce the risk of costly clin.-stage attrition due to the metabolic characteristics of drug candidates, there is a need for efficient and reliable ways to predict drug metab. in vitro, in silico and in vivo. In this Perspective, we provide an overview of the state of the art of exptl. and computational approaches for investigating drug metab. We highlight the scope and limitations of these methods, and indicate strategies to harvest the synergies that result from combining measurement and prediction of drug metab.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC2MXntFCju7k%253D&md5=231060e7860047948cffd071b8a57197
4
Campagna-Slater, V.; Pottel, J.; Therrien, E.; Cantin, L.-D.; Moitessier, N. Development of a Computational Tool to Rival Experts in the Prediction of Sites of Metabolism of Xenobiotics by P450s J. Chem. Inf. Model. 2012, 52, 2471– 2483 DOI: 10.1021/ci3003073
[ACS Full Text ], [CAS], Google Scholar
4
Development of a Computational Tool to Rival Experts in the Prediction of Sites of Metabolism of Xenobiotics by P450s
Campagna-Slater, Valerie; Pottel, Joshua; Therrien, Eric; Cantin, Louis-David; Moitessier, Nicolas
Journal of Chemical Information and Modeling (2012), 52 (9), 2471-2483CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
The metab. of xenobiotics - and more specifically drugs - in the liver is a crit. process controlling their half-life. Although there exist exptl. methods, which measure the metabolic stability of xenobiotics and identify their metabolites, developing higher throughput predictive methods is an avenue of research. It is expected that predicting the chem. nature of the metabolites would be an asset for designing safer drugs and/or drugs with modulated half-lives. We have developed IMPACTS (In-silico Metab. Prediction by Activated Cytochromes and Transition States), a computational tool combining docking to metabolic enzymes, transition state modeling, and rule-based substrate reactivity prediction to predict the site of metab. (SoM) of xenobiotics. Its application to sets of CYP1A2, CYP2C9, CYP2D6, and CYP3A4 substrates and comparison to experts' predictions demonstrates its accuracy and significance. IMPACTS identified an exptl. obsd. SoM in the top 2 predicted sites for 77% of the substrates, while the accuracy of biotransformation experts' prediction was 65%. Application of IMPACTS to external sets and comparison of its accuracy to those of eleven other methods further validated the method implemented in IMPACTS.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC38Xht1CisrvP&md5=c954b56c19c486106cb7fec0f51fe68a
5
Crivori, P.; Poggesi, I. Computational Approaches for Predicting CYP-Related Metabolism Properties in the Screening of New Drugs Eur. J. Med. Chem. 2006, 41, 795– 808 DOI: 10.1016/j.ejmech.2006.03.003
[Crossref], [PubMed], [CAS], Google Scholar
5
Computational approaches for predicting CYP-related metabolism properties in the screening of new drugs
Crivori, P.; Poggesi, I.
European Journal of Medicinal Chemistry (2006), 41 (7), 795-808CODEN: EJMCA5; ISSN:0223-5234. (Elsevier B.V.)
A review. The site of biotransformation, the extent and rate of metab. and the no. of active metabolic pathways are among the most important characteristics of the pharmacokinetics of a drug. The catalytic activity of drug metabolizing enzymes is likely the most influential determinant of the pharmacokinetic variability. Metabolic stability is the prerequisite for sustaining the therapeutically relevant concns. Metabolic inhibition and induction can give rise to clin. important drug-drug interactions. A variety of computational approaches are currently available for predicting different cytochrome P 450 (CYP)-related metab. endpoints. The present review will describe these approaches and their impact on drug development process. Indications on the available software for the implementation will also be given.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD28XmvFeltro%253D&md5=1de13b7845575861e4f5ef85af3d809b
6
Tarcsay, Á.; Keserű, G. M. In Silico Site of Metabolism Prediction of Cytochrome P450-Mediated Biotransformations Expert Opin. Drug Metab. Toxicol. 2011, 7, 299– 312 DOI: 10.1517/17425255.2011.553599
[Crossref], [PubMed], [CAS], Google Scholar
6
In silico site of metabolism prediction of cytochrome P450-mediated biotransformations
Tarcsay, Akos; Keseru, Gyorgy M.
Expert Opinion on Drug Metabolism & Toxicology (2011), 7 (3), 299-312CODEN: EODMAP; ISSN:1742-5255. (Informa Healthcare)
A review. Preclin. research involves the in vitro monitoring of metabolic stability to deliver compds. with improved ADME profiles. Prediction of the metabolically vulnerable points can substantially help in analyzing CYP-mediated metab. data and support optimization efforts in drug discovery programs. Moreover, fast and reliable in silico predictions could accelerate the characterization of in vitro/in vivo metabolites. This paper reviews in silico methods available for CYP-mediated site of metab. (SOM) prediction. Comprehensive and practical knowledge in this field can guide the identification of best practice and may inspire ideas for the development of novel approaches. Comparison of the efficacy of SOM prediction methodologies revealed the general dependency on the studied isoform and substrate set. Increasing knowledge on P 450 X-ray structures, on biotransformations and on the mechanistic details of the catalytic cycle revolutionized the prediction of SOM. Although no ultimate soln. exits, combined methods covering both steric and electronic effects are preferred on most of the pharmaceutically relevant isoforms.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3MXit1Wku7k%253D&md5=f3631600a6ebeee3c7338521ccdb7424
7
Kirchmair, J.; Williamson, M. J.; Tyzack, J. D.; Tan, L.; Bond, P. J.; Bender, A.; Glen, R. C. Computational Prediction of Metabolism: Sites, Products, SAR, P450 Enzyme Dynamics, and Mechanisms J. Chem. Inf. Model. 2012, 52, 617– 648 DOI: 10.1021/ci200542m
[ACS Full Text ], [CAS], Google Scholar
7
Computational Prediction of Metabolism: Sites, Products, SAR, P450 Enzyme Dynamics, and Mechanisms
Kirchmair, Johannes; Williamson, Mark J.; Tyzack, Jonathan D.; Tan, Lu; Bond, Peter J.; Bender, Andreas; Glen, Robert C.
Journal of Chemical Information and Modeling (2012), 52 (3), 617-648CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
A review perspective. Metab. of xenobiotics remains a central challenge for the discovery and development of drugs, cosmetics, nutritional supplements, and agrochems. Metabolic transformations are frequently related to the incidence of toxic effects that may result from the emergence of reactive species, the systemic accumulation of metabolites, or by induction of metabolic pathways. Exptl. investigation of the metab. of small org. mols. is particularly resource demanding; hence, computational methods are of considerable interest to complement exptl. approaches. This review provides a broad overview of structure- and ligand-based computational methods for the prediction of xenobiotic metab. Current computational approaches to address xenobiotic metab. are discussed from three major perspectives: (i) prediction of sites of metab. (SOMs), (ii) elucidation of potential metabolites and their chem. structures, and (iii) prediction of direct and indirect effects of xenobiotics on metabolizing enzymes, where the focus is on the cytochrome P 450 (CYP) superfamily of enzymes, the cardinal xenobiotics metabolizing enzymes. For each of these domains, a variety of approaches and their applications are systematically reviewed, including expert systems, data mining approaches, quant. structure-activity relationships (QSARs), and machine learning-based methods, pharmacophore-based algorithms, shape-focused techniques, mol. interaction fields (MIFs), reactivity-focused techniques, protein-ligand docking, mol. dynamics (MD) simulations, and combinations of methods. Predictive metab. is a developing area, and there is still enormous potential for improvement. However, it is clear that the combination of rapidly increasing amts. of available ligand- and structure-related exptl. data (in particular, quant. data) with novel and diverse simulation and modeling approaches is accelerating the development of effective tools for prediction of in vivo metab., which is reflected by the diverse and comprehensive data sources and methods for metab. prediction reviewed here. This review attempts to survey the range and scope of computational methods applied to metab. prediction and also to compare and contrast their applicability and performance.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC38XisVyitL4%253D&md5=d144f6054d9f27774f476d026a066805
8
Raunio, H.; Kuusisto, M.; Juvonen, R. O.; Pentikäinen, O. T. Modeling of Interactions between Xenobiotics and Cytochrome P450 (CYP) Enzymes Front. Pharmacol. 2015, 6, 123 DOI: 10.3389/fphar.2015.00123
[Crossref], [PubMed], [CAS], Google Scholar
8
Modeling of interactions between xenobiotics and cytochrome P450 (CYP) enzymes
Raunio Hannu; Juvonen Risto O; Kuusisto Mira; Pentikainen Olli T
Frontiers in pharmacology (2015), 6 (), 123 ISSN:1663-9812.
The adverse effects to humans and environment of only few chemicals are well known. Absorption, distribution, metabolism, and excretion (ADME) are the steps of pharmaco/toxicokinetics that determine the internal dose of chemicals to which the organism is exposed. Of all the xenobiotic-metabolizing enzymes, the cytochrome P450 (CYP) enzymes are the most important due to their abundance and versatility. Reactions catalyzed by CYPs usually turn xenobiotics to harmless and excretable metabolites, but sometimes an innocuous xenobiotic is transformed into a toxic metabolite. Data on ADME and toxicity properties of compounds are increasingly generated using in vitro and modeling (in silico) tools. Both physics-based and empirical modeling approaches are used. Numerous ligand-based and target-based as well as combined modeling methods have been employed to evaluate determinants of CYP ligand binding as well as predicting sites of metabolism and inhibition characteristics of test molecules. In silico prediction of CYP-ligand interactions have made crucial contributions in understanding (1) determinants of CYP ligand binding recognition and affinity; (2) prediction of likely metabolites from substrates; (3) prediction of inhibitors and their inhibition potency. Truly predictive models of toxic outcomes cannot be created without incorporating metabolic characteristics; in silico methods help producing such information and filling gaps in experimentally derived data. Currently modeling methods are not mature enough to replace standard in vitro and in vivo approaches, but they are already used as an important component in risk assessment of drugs and other chemicals.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A280%3ADC%252BC2MbosFGiuw%253D%253D&md5=aaf28501a8579f1a24ec8e5c80f24a3a
9
Bezhentsev, V. M.; Tarasova, O. A.; Dmitriev, A. V.; Rudik, A. V.; Lagunin, A. A.; Filimonov, D. A.; Poroikov, V. V. Computer-Aided Prediction of Xenobiotic Metabolism in the Human Body Russ. Chem. Rev. 2016, 85, 854 DOI: 10.1070/RCR4614
[Crossref], [CAS], Google Scholar
9
Computer-aided prediction of xenobiotic metabolism in humans
Bezhentsev, Vladislav M.; Tarasova, Olga A.; Dmitriev, Aleksandr V.; Rudik, Anastasiya V.; Lagunin, Aleksey A.; Filimonov, Dmitriy A.; Poroikov, Vladimir V.
Russian Chemical Reviews (2016), 85 (8), 854-879CODEN: RCRVAB; ISSN:0036-021X. (IOP Publishing Ltd.)
A review. The review describes the main databases contg. information about the metab. of xenobiotics, including data on drug metab., metabolic enzymes, schemes of biotransformation and the structures of some substrates and metabolites. Computational approaches used to predict the interaction of xenobiotics with metabolic enzymes, to identify the sites of metab. in the mol. and to generate structures of potential metabolites for subsequent evaluation of their properties are considered. The advantages and limitations of particular computational methods for metab. prediction are indicated and the prospects for their applications to improve the safety and efficacy of new drugs are discussed.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC28XhvF2jsb7L&md5=4ea739a940a38faeaabadba470f2cd05
10
Rydberg, P. Reactivity-Based Approaches and Machine Learning Methods for Predicting the Sites of Cytochrome P450-Mediated Metabolism. In Drug Metabolism Prediction; Kirchmair, J., Ed.; Wiley-VCH Verlag GmbH & Co. KGaA: Weinheim, 2014; pp 265– 292.
[Crossref], Google Scholar
There is no corresponding record for this reference.
11
Rydberg, P.; Olsen, L. Predicting Drug Metabolism by Cytochrome P450 2C9: Comparison with the 2D6 and 3A4 Isoforms ChemMedChem 2012, 7, 1202– 1209 DOI: 10.1002/cmdc.201200160
[Crossref], [PubMed], [CAS], Google Scholar
11
Predicting Drug Metabolism by Cytochrome P450 2C9: Comparison with the 2D6 and 3A4 Isoforms
Rydberg, Patrik; Olsen, Lars
ChemMedChem (2012), 7 (7), 1202-1209, S1202/1-S1202/33CODEN: CHEMGX; ISSN:1860-7179. (Wiley-VCH Verlag GmbH & Co. KGaA)
By the use of knowledge gained through modeling of drug metab. mediated by the cytochrome P 450 2D6 and 3A4 isoforms, we constructed a 2D-based model for site-of-metab. prediction for the cytochrome P 450 2C9 isoform. The similarities and differences between the models for the 2C9 and 2D6 isoforms are discussed through structural knowledge from the X-ray crystal structures and trends in exptl. data. The final model was validated on an independent test set, resulting in an area under the curve value of 0.92, and a site of metab. was found among the top two ranked atoms for 77 % of the compds.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC38XntFWitr0%253D&md5=0dd476527c405af9de49a076a30c1181
12
Darvas, F. Predicting Metabolic Pathways by Logic Programming J. Mol. Graphics 1988, 6, 80– 86 DOI: 10.1016/0263-7855(88)85004-5
[Crossref], [CAS], Google Scholar
12
Predicting metabolic pathways by logic programming
Darvas, Ferenc
Journal of Molecular Graphics (1988), 6 (2), 80-6CODEN: JMGRDV; ISSN:0263-7855.
A discussion of logic programming is presented and its application is described in an expert system used to simulate the metabolic fate of substances. An expert system called Metabolexpert accepts the formula of the compd. to be metabolized and produces a treelike picture of the metabolites generated together with the formula of the compds. On request, 3-dimensional pictures of the metabolites are also displayed and hydrophobicity values of the compds. calcd. A retrospective investigation of Metabolexpert's achievement showed that the expert system can reproduce almost all primary, secondary, and tertiary metabolites of amphetamine. A compd. series has been suggested for benchmark testing of metabolic transformation knowledge bases.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADyaL1cXkslaltL8%253D&md5=1d73455bb8ab8d32c02dff7b3a4d78a4
13
Klopman, G.; Dimayuga, M.; Talafous, J. META. 1. A Program for the Evaluation of Metabolic Transformation of Chemicals J. Chem. Inf. Model. 1994, 34, 1320– 1325 DOI: 10.1021/ci00022a014
[ACS Full Text ], [CAS], Google Scholar
13
META. 1. A Program for the Evaluation of Metabolic Transformation of Chemicals
Klopman, Gilles; Dimayuga, Mario; Talafous, Joseph
Journal of Chemical Information and Computer Sciences (1994), 34 (6), 1320-5CODEN: JCISD8; ISSN:0095-2338.
A new metab. program, META, is introduced. In this paper, the basic principles on which the program operates are described. META is an expert system, capable of predicting the sites of potential enzymic attack and the nature of the chems. formed by such metabolic transformations. It operates from dictionaries of transformation operators, created by experts to represent known metabolic paths.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADyaK2cXmsVyitrw%253D&md5=9f145c8b540affcaf4a67b6306b991aa
14
Talafous, J.; Sayre, L. M.; Mieyal, J. J.; Klopman, G. META. 2. A Dictionary Model of Mammalian Xenobiotic Metabolism J. Chem. Inf. Model. 1994, 34, 1326– 1333 DOI: 10.1021/ci00022a015
[ACS Full Text ], Google Scholar
There is no corresponding record for this reference.
15
Greene, N.; Judson, P. N.; Langowski, J. J.; Marchant, C. A. Knowledge-Based Expert Systems for Toxicity and Metabolism Prediction: DEREK, StAR and METEOR SAR QSAR Environ. Res. 1999, 10, 299– 314 DOI: 10.1080/10629369908039182
[Crossref], [PubMed], [CAS], Google Scholar
15
Knowledge-based expert systems for toxicity and metabolism prediction: DEREK, StAR and METEOR
Greene, N.; Judson, P. N.; Langowski, J. J.; Marchant, C. A.
SAR and QSAR in Environmental Research (1999), 10 (2-3), 299-314, 2 platesCODEN: SQERED; ISSN:1062-936X. (Gordon & Breach Science Publishers)
It has long been recognized that the ability to predict the metabolic fate of a chem. substance and the potential toxicity of either the parent compd. or its metabolites are important in novel drug design. The popularity of using computer models as an aid in this area has grown considerably in recent years. LHASA Limited has been developing knowledge-based expert systems for toxicity and metab. prediction in collaboration with industry and regulatory authorities. These systems, DEREK, StAR, and METEOR, use rules to describe the relationship between chem. structure and either toxicity in the case of DEREK and StAR or metabolic fate in the case of METEOR. The rule refinement process for DEREK often involves assessing the predictions for a novel set of compds. and comparing them to their biol. assay results as a measure of the system's performance. For example, 266 non-congeneric chems. from the National Toxicol. Program database have been processed through the DEREK mutagenicity knowledge base and the predictions compared to their Salmonella typhimurium mutagenicity data. Initially, 81 of 114 mutagens (71%) and 117 of 152 non-mutagens (77%) were correctly identified. Following further knowledge base development, the no. of correctly identified mutagens has increased to 96 (84%). Further work on improving the predictive capabilities of DEREK, StAR, and METEOR is in progress.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADyaK1MXlvFCgsbc%253D&md5=5988de2a40351b9416bc21a9688a3a6d
16
Hou, B. K.; Wackett, L. P.; Ellis, L. B. M. Microbial Pathway Prediction: A Functional Group Approach J. Chem. Inf. Comput. Sci. 2003, 43, 1051– 1057 DOI: 10.1021/ci034018f
[ACS Full Text ], [CAS], Google Scholar
16
Microbial pathway prediction: a functional group approach
Hou, Bo Kyeng; Wackett, Lawrence P.; Ellis, Lynda B. M.
Journal of Chemical Information and Computer Sciences (2003), 43 (3), 1051-1057CODEN: JCISD8; ISSN:0095-2338. (American Chemical Society)
We have developed a system to predict microbial catabolism, using the University of Minnesota Biocatalysis/Biodegrdn. Database (UM-BBD, http://umbbd.ahc.umn.edu/) as a knowledge base. The present system, available on the Web (http://umbbd.ahc.umn.edu/predict/), can predict biodegrdn. of most of the major aliph. and arom. org. functional groups contg. C, H, N, O, and halogens. It can duplicate at least one known biodegrdn. pathway for 60% of the compds. in a 84-member validation set; most pathways that did not completely duplicate known metab. could plausibly occur in nature. Users are encouraged, and have begun, to submit addnl. biotransformation rules and comment on existing rules; the system will further develop under the direction of the scientific community.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD3sXis1Kgt74%253D&md5=c7a314ccbde95b8ceef6af0276469840
17
Hatzimanikatis, V.; Li, C.; Ionita, J. A.; Henry, C. S.; Jankowski, M. D.; Broadbelt, L. J. Exploring the Diversity of Complex Metabolic Networks Bioinformatics 2005, 21, 1603– 1609 DOI: 10.1093/bioinformatics/bti213
[Crossref], [PubMed], [CAS], Google Scholar
17
Exploring the diversity of complex metabolic networks
Hatzimanikatis, Vassily; Li, Chunhui; Ionita, Justin A.; Henry, Christopher S.; Jankowski, Matthew D.; Broadbelt, Linda J.
Bioinformatics (2005), 21 (8), 1603-1609CODEN: BOINFP; ISSN:1367-4803. (Oxford University Press)
Motivation: Metab., the network of chem. reactions that make life possible, is one of the most complex processes in nature. We describe here the development of a computational approach for the identification of every possible biochem. reaction from a given set of enzyme reaction rules that allows the de novo synthesis of metabolic pathways composed of these reactions, and the evaluation of these novel pathways with respect to their thermodn. properties. Results: We applied this framework to the anal. of the arom. amino acid pathways and discovered almost 75,000 novel biochem. routes from chorismate to phenylalanine, more than 350,000 from chorismate to tyrosine, but only 13 from chorismate to tryptophan. Thermodn. anal. of these pathways suggests that the native pathways are thermodynamically more favorable than the alternative possible pathways. The pathways generated involve compds. that exist in biol. databases, as well as compds. that exist in chem. databases and novel compds., suggesting novel biochem. routes for these compds. and the existence of biochem. compds. that remain to be discovered or synthesized through enzyme and pathway engineering.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD2MXjtlGhuro%253D&md5=343be8db9eada86fcba34ae047529dec
18
Ridder, L.; Wagener, M. SyGMa: Combining Expert Knowledge and Empirical Scoring in the Prediction of Metabolites ChemMedChem 2008, 3, 821– 832 DOI: 10.1002/cmdc.200700312
[Crossref], [PubMed], [CAS], Google Scholar
18
SyGMa: combining expert knowledge and empirical scoring in the prediction of metabolites
Ridder, Lars; Wagener, Markus
ChemMedChem (2008), 3 (5), 821-832CODEN: CHEMGX; ISSN:1860-7179. (Wiley-VCH Verlag GmbH & Co. KGaA)
Predictions of potential metabolites based on chem. structure are becoming increasingly important in drug discovery to guide medicinal chem. efforts that address metabolic issues and to support exptl. metabolite screening and identification. Herein we present a novel rule-based method, SyGMa (Systematic Generation of potential Metabolites), to predict the potential metabolites of a given parent structure. A set of reaction rules covering a broad range of phase 1 and phase 2 metab. has been derived from metabolic reactions reported in the Metabolite Database to occur in humans. An empirical probability score is assigned to each rule representing the fraction of correctly predicted metabolites in the training database. This score is used to refine the rules and to rank predicted metabolites. The current rule set of SyGMa covers approx. 70% of biotransformation reactions obsd. in humans. Evaluation of the rule-based predictions demonstrated a significant enrichment of true metabolites in the top of the ranking list: while in total, 68% of all obsd. metabolites in an independent test set were reproduced by SyGMa, a large part, 30% of the obsd. metabolites, were identified among the top three predictions. From a subset of cytochrome P 450 specific metabolites, 84% were reproduced overall, with 66% in the top three predicted phase 1 metabolites. A similarity anal. of the reactions present in the database was performed to obtain an overview of the metabolic reactions predicted by SyGMa and to support ongoing efforts to extend the rules. Specific examples demonstrate the use of SyGMa in exptl. metabolite identification and the application of SyGMa to suggest chem. modifications that improve the metabolic stability of compds.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD1cXms1emtLw%253D&md5=92ed84b2e97af5ad9a50beef151fa5db
19
Gao, J.; Ellis, L. B. M.; Wackett, L. P. The University of Minnesota Pathway Prediction System: Multi-Level Prediction and Visualization Nucleic Acids Res. 2011, 39, W406– W411 DOI: 10.1093/nar/gkr200
[Crossref], [PubMed], [CAS], Google Scholar
19
The University of Minnesota Pathway Prediction System: multi-level prediction and visualization
Gao, Junfeng; Ellis, Lynda B. M.; Wackett, Lawrence P.
Nucleic Acids Research (2011), 39 (Web Server), W406-W411CODEN: NARHAD; ISSN:0305-1048. (Oxford University Press)
The University of Minnesota Pathway Prediction System (UM-PPS, http://umbbd.msi.umn.edu/predict/) is a rule-based system that predicts microbial catabolism of org. compds. Currently, its knowledge base contains 250 biotransformation rules and five types of metabolic logic entities. The original UM-PPS predicted up to two prediction levels at a time. Users had to choose a predicted product to continue the prediction. This approach provided a limited view of prediction results and heavily relied on manual intervention. The new UM-PPS produces a multi-level prediction within an acceptable time frame, and allows users to view prediction alternatives much more easily as a directed acyclic graph.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3MXosVOmsLs%253D&md5=4b27083f37d70ab33ba63c1e27a2959b
20
Mu, F.; Unkefer, C. J.; Unkefer, P. J.; Hlavacek, W. S. Prediction of Metabolic Reactions Based on Atomic and Molecular Properties of Small-Molecule Compounds Bioinformatics 2011, 27, 1537– 1545 DOI: 10.1093/bioinformatics/btr177
[Crossref], [PubMed], [CAS], Google Scholar
20
Prediction of metabolic reactions based on atomic and molecular properties of small-molecule compounds
Mu, Fangping; Unkefer, Clifford J.; Unkefer, Pat J.; Hlavacek, William S.
Bioinformatics (2011), 27 (11), 1537-1545CODEN: BOINFP; ISSN:1367-4803. (Oxford University Press)
Our knowledge of the metabolites in cells and their reactions is far from complete as revealed by metabolomic measurements that detect many more small mols. than are documented in metabolic databases. Here, we develop an approach for predicting the reactivity of small-mol. metabolites in enzyme-catalyzed reactions that combines expert knowledge, computational chem. and machine learning. We classified 4843 reactions documented in the KEGG database, from all six Enzyme Commission classes (EC 1-6), into 80 reaction classes, each of which is marked by a characteristic functional group transformation. Reaction centers and surrounding local structures in substrates and products of these reactions were represented using SMARTS. We found that each of the SMARTS-defined chem. substructures is widely distributed among metabolites, but only a fraction of the functional groups in these substructures are reactive. Using at. properties of atoms in a putative reaction center and mol. properties as features, we trained support vector machine (SVM) classifiers to discriminate between functional groups that are reactive and non-reactive. Classifier accuracy was assessed by cross-validation anal. A typical sensitivity [TP/(TP+FN)] or specificity [TN/(TN+FP)] is ≈0.8. Our results suggest that metabolic reactivity of small-mol. compds. can be predicted with reasonable accuracy based on the presence of a potentially reactive functional group and the chem. features of its local environment. The classifiers presented here can be used to predict reactions via a web site (http://cellsignaling.lanl.gov/Reactivity/). Contact: [email protected].
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3MXmvVajsLs%253D&md5=97709401a9db367e98b2278a9d601816
21
Yousofshahi, M.; Manteiga, S.; Wu, C.; Lee, K.; Hassoun, S. PROXIMAL: A Method for Prediction of Xenobiotic Metabolism BMC Syst. Biol. 2015, 9, 94 DOI: 10.1186/s12918-015-0241-4
[Crossref], [PubMed], [CAS], Google Scholar
21
PROXIMAL: a method for Prediction of Xenobiotic Metabolism
Yousofshahi, Mona; Manteiga, Sara; Wu, Charmian; Lee, Kyongbum; Hassoun, Soha
BMC Systems Biology (2015), 9 (), 94/1-94/17CODEN: BSBMCC; ISSN:1752-0509. (BioMed Central Ltd.)
Background: Contamination of the environment with bioactive chems. has emerged as a potential public health risk. These substances that may cause distress or disease in humans can be found in air, water and food supplies. An open question is whether these chems. transform into potentially more active or toxic derivs. via xenobiotic metabolizing enzymes expressed in the body. We present a new prediction tool, which we call PROXIMAL (Prediction of Xenobiotic Metab.) for identifying possible transformation products of xenobiotic chems. in the liver. Using reaction data from DrugBank and KEGG, PROXIMAL builds look-up tables that catalog the sites and types of structural modifications performed by Phase I and Phase II enzymes. Given a compd. of interest, PROXIMAL searches for substructures that match the sites cataloged in the look-up tables, applies the corresponding modifications to generate a panel of possible transformation products, and ranks the products based on the activity and abundance of the enzymes involved. Results: PROXIMAL generates transformations that are specific for the chem. of interest by analyzing the chem.'s substructures. We evaluate the accuracy of PROXIMAL's predictions through case studies on two environmental chems. with suspected endocrine disrupting activity, bisphenol A (BPA) and 4-chlorobiphenyl (PCB3). Comparisons with published reports confirm 5 out of 7 and 17 out of 26 of the predicted derivs. for BPA and PCB3, resp. We also compare biotransformation predictions generated by PROXIMAL with those generated by METEOR and Metaprint2D-react, two other prediction tools. Conclusions: PROXIMAL can predict transformations of chems. that contain substructures recognizable by human liver enzymes. It also has the ability to rank the predicted metabolites based on the activity and abundance of enzymes involved in xenobiotic transformation.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC2sXjtVyltbo%253D&md5=131eaa3ba02457ab89a73d1481d38ea1
22
Sun, H.; Scott, D. O. Structure-Based Drug Metabolism Predictions for Drug Design Chem. Biol. Drug Des. 2010, 75, 3– 17 DOI: 10.1111/j.1747-0285.2009.00899.x
[Crossref], [PubMed], [CAS], Google Scholar
22
Structure-based drug metabolism predictions for drug design
Sun, Hao; Scott, Dennis O.
Chemical Biology & Drug Design (2010), 75 (1), 3-17CODEN: CBDDAL; ISSN:1747-0277. (Wiley-Blackwell)
A review. Significant progress has been made in structure-based drug design by pharmaceutical companies at different stages of drug discovery such as identifying new hits, enhancing mol. binding affinity in hit-to-lead, and reducing toxicities in lead optimization. Drug metab. is a major consideration for modifying drug clearance and also a primary source for drug metabolite-induced toxicity. With major cytochrome P 450 structures identified and characterized recently, structure-based drug metab. prediction becomes increasingly attractive. In silico methods based on mol. and quantum mechanics such as docking, mol. dynamics and ab initio chem. reactivity calcns. bring us closer to understand drug metab. and predict drug-drug interactions. In this study, we review important progress in drug metab. and common in silico techniques adopted to predict drug regioselectivity, stereoselectivity, reactive metabolites, induction, inhibition and mechanism-based inactivation, as well as their implementation in hit-to-lead drug discovery.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD1MXhs1WhsbfO&md5=52666dc1d960035e9def127cb1963860
23
Kingsley, L. J.; Wilson, G. L.; Essex, M. E.; Lill, M. A. Combining Structure- and Ligand-Based Approaches to Improve Site of Metabolism Prediction in CYP2C9 Substrates Pharm. Res. 2015, 32, 986– 1001 DOI: 10.1007/s11095-014-1511-3
[Crossref], Google Scholar
There is no corresponding record for this reference.
24
Sheridan, R. P.; Korzekwa, K. R.; Torres, R. A.; Walker, M. J. Empirical Regioselectivity Models for Human Cytochromes P450 3A4, 2D6, and 2C9 J. Med. Chem. 2007, 50, 3173– 3184 DOI: 10.1021/jm0613471
[ACS Full Text ], [CAS], Google Scholar
24
Empirical Regioselectivity Models for Human Cytochromes P450 3A4, 2D6, and 2C9
Sheridan, Robert P.; Korzekwa, Kenneth R.; Torres, Rhonda A.; Walker, Matthew J.
Journal of Medicinal Chemistry (2007), 50 (14), 3173-3184CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
Cytochromes P 450 3A4, 2D6, and 2C9 metabolize a large fraction of drugs. Knowing where these enzymes will preferentially oxidize a mol., the regioselectivity, allows medicinal chemists to plan how best to block its metab. The authors present QSAR-based regioselectivity models for these enzymes calibrated against compiled literature data of drugs and drug-like compds. These models are purely empirical and use only the structures of the substrates, in contrast to those models that simulate a specific mechanism like hydrogen radical abstraction, and/or use explicit models of active sites. The authors most predictive models use three substructure descriptors and two phys. property descriptors. Descriptor importance from the random forest QSAR method show that other factors than the immediate chem. environment and the accessibility of the hydrogen affect regioselectivity in all three isoforms. The cross-validated predictions of the models are compared to predictions from the authors earlier mechanistic model (Singh et al. J. Med. Chem. 2003, 46, 1330-1336) and predictions from MetaSite (Cruciani et al. J. Med. Chem. 2005, 48, 6970-6979).
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD2sXmsFGnsr0%253D&md5=645bd5ddeb0459636bf6067e0194dd34
25
Rydberg, P.; Gloriam, D. E.; Zaretzki, J.; Breneman, C.; Olsen, L. SMARTCyp: A 2D Method for Prediction of Cytochrome P450-Mediated Drug Metabolism ACS Med. Chem. Lett. 2010, 1, 96– 100 DOI: 10.1021/ml100016x
[ACS Full Text ], [CAS], Google Scholar
25
SMARTCyp: A 2D Method for Prediction of Cytochrome P450-Mediated Drug Metabolism
Rydberg, Patrik; Gloriam, David E.; Zaretzki, Jed; Breneman, Curt; Olsen, Lars
ACS Medicinal Chemistry Letters (2010), 1 (3), 96-100CODEN: AMCLCT; ISSN:1948-5875. (American Chemical Society)
SMARTCyp is an in silico method that predicts the sites of cytochrome P 450-mediated metab. of druglike mols. The method is foremost a reactivity model, and as such, it shows a preference for predicting sites that are metabolized by the cytochrome P 450 3A4 isoform. SMARTCyp predicts the site of metab. directly from the 2D structure of a mol., without requiring calcn. of electronic properties or generation of 3D structures. This is a major advantage, because it makes SMARTCyp very fast. Other advantages are that exptl. data are not a prerequisite to create the model, and it can easily be integrated with other methods to create models for other cytochrome P 450 isoforms. Benchmarking tests on a database of 394 3A4 substrates show that SMARTCyp successfully identifies at least one metabolic site in the top two ranked positions 76% of the time. SMARTCyp is available for download at http://www.farma.ku.dk/P 450.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3cXjtFOqsL0%253D&md5=4761de933c2c3cef5da2ffb29ac076d0
26
Rydberg, P.; Gloriam, D. E.; Olsen, L. The SMARTCyp Cytochrome P450 Metabolism Prediction Server Bioinformatics 2010, 26, 2988– 2989 DOI: 10.1093/bioinformatics/btq584
[Crossref], [PubMed], [CAS], Google Scholar
26
The SMARTCyp cytochrome P450 metabolism prediction server
Rydberg, Patrik; Gloriam, David E.; Olsen, Lars
Bioinformatics (2010), 26 (23), 2988-2989CODEN: BOINFP; ISSN:1367-4803. (Oxford University Press)
The SMARTCyp server is the first web application for site of metab. prediction of cytochrome P 450-mediated drug metab.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3cXhsVKjtL3I&md5=a8689591d976ca80d0ae6a3e15b32636
27
Rydberg, P.; Rostkowski, M.; Gloriam, D. E.; Olsen, L. The Contribution of Atom Accessibility to Site of Metabolism Models for Cytochromes P450 Mol. Pharmaceutics 2013, 10, 1216– 1223 DOI: 10.1021/mp3005116
[ACS Full Text ], [CAS], Google Scholar
27
The Contribution of Atom Accessibility to Site of Metabolism Models for Cytochromes P450
Rydberg, Patrik; Rostkowski, Michal; Gloriam, David E.; Olsen, Lars
Molecular Pharmaceutics (2013), 10 (4), 1216-1223CODEN: MPOHBP; ISSN:1543-8384. (American Chemical Society)
Three different types of atom accessibility descriptors are investigated in relation to site of metab. predictions. To enable the integration of local accessibility we have constructed 2DSASA, a method for the calcn. of the at. solvent accessible surface area that is independent of 3D coordinates. The method was implemented in the SMARTCyp site of metab. prediction models and improved the results by up to 4 percentage points for nine cytochrome P 450 isoforms. The final models are made available at http://www.farma.ku.dk/smartcyp.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3sXhtFyjs7c%253D&md5=206ed0792c118423008bca0e6a793096
28
Zaretzki, J.; Bergeron, C.; Rydberg, P.; Huang, T.-W.; Bennett, K. P.; Breneman, C. M. RS-Predictor: A New Tool for Predicting Sites of Cytochrome P450-Mediated Metabolism Applied to CYP 3A4 J. Chem. Inf. Model. 2011, 51, 1667– 1689 DOI: 10.1021/ci2000488
[ACS Full Text ], [CAS], Google Scholar
28
RS-Predictor: A New Tool for Predicting Sites of Cytochrome P450-Mediated Metabolism Applied to CYP 3A4
Zaretzki, Jed; Bergeron, Charles; Rydberg, Patrik; Huang, Tao-wei; Bennett, Kristin P.; Breneman, Curt M.
Journal of Chemical Information and Modeling (2011), 51 (7), 1667-1689CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
This article describes RegioSelectivity-Predictor (RS-Predictor), a new in silico method for generating predictive models of P 450-mediated metab. for drug-like compds. Within this method, potential sites of metab. (SOMs) are represented as "metabolophores": A concept that describes the hierarchical combination of topol. and quantum chem. descriptors needed to represent the reactivity of potential metabolic reaction sites. RS-Predictor modeling involves the use of metabolophore descriptors together with multiple-instance ranking (MIRank) to generate an optimized descriptor wt. vector that encodes regioselectivity trends across all cases in a training set. The resulting pathway-independent (O-dealkylation vs. N-oxidn. vs. Csp3 hydroxylation, etc.), isoenzyme-specific regioselectivity model may be used to predict potential metabolic liabilities. In the present work, cross-validated RS-Predictor models were generated for a set of 394 substrates of CYP 3A4 as a proof-of-principle for the method. Rank aggregation was then employed to merge independently generated predictions for each substrate into a single consensus prediction. The resulting consensus RS-Predictor models were shown to reliably identify at least one obsd. site of metab. in the top two rank-positions on 78% of the substrates. Comparisons between RS-Predictor and previously described regioselectivity prediction methods reveal new insights into how in silico metabolite prediction methods should be compared.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3MXnsVertLs%253D&md5=3d863c866d45fc1c8e90d033a16ed6a1
29
Zaretzki, J.; Rydberg, P.; Bergeron, C.; Bennett, K. P.; Olsen, L.; Breneman, C. M. RS-Predictor Models Augmented with SMARTCyp Reactivities: Robust Metabolic Regioselectivity Predictions for Nine CYP Isozymes J. Chem. Inf. Model. 2012, 52, 1637– 1659 DOI: 10.1021/ci300009z
[ACS Full Text ], [CAS], Google Scholar
29
RS-Predictor Models Augmented with SMARTCyp Reactivities: Robust Metabolic Regioselectivity Predictions for Nine CYP Isozymes
Zaretzki, Jed; Rydberg, Patrik; Bergeron, Charles; Bennett, Kristin P.; Olsen, Lars; Breneman, Curt M.
Journal of Chemical Information and Modeling (2012), 52 (6), 1637-1659CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
RS-Predictor is a tool for creating pathway-independent, isoenzyme-specific, site of metab. (SOM) prediction models using any set of known cytochrome P 450 (CYP) substrates and metabolites. Until now, the RS-Predictor method was only trained and validated on CYP 3A4 data, but in the present study, we report on the versatility the RS-Predictor modeling paradigm by creating and testing regioselectivity models for substrates of the nine most important CYP isoenzymes. Through curation of source literature, we have assembled 680 substrates distributed among CYPs 1A2, 2A6, 2B6, 2C19, 2C8, 2C9, 2D6, 2E1, and 3A4, the largest publicly accessible collection of P 450 ligands and metabolites released to date. A comprehensive investigation into the importance of different descriptor classes for identifying the regioselectivity mediated by each isoenzyme is made through the generation of multiple independent RS-Predictor models for each set of isoenzyme substrates. Two of these models include a d. functional theory (DFT) reactivity descriptor derived from SMARTCyp. Optimal combinations of RS-Predictor and SMARTCyp are shown to have stronger performance than either method alone, while also exceeding the accuracy of the com. regioselectivity prediction methods distributed by Optibrium and Schroedinger, correctly identifying a large proportion of the metabolites in each substrate set within the top two rank-positions: 1A2 (83.0%), 2A6 (85.7%), 2B6 (82.1%), 2C19 (86.2%), 2C8 (83.8%), 2C9 (84.5%), 2D6 (85.9%), 2E1 (82.8%), 3A4 (82.3%), and merged (86.0%). Comprehensive datamining of each substrate set and careful statistical analyses of the predictions made by the different models revealed new insights into mol. features that control metabolic regioselectivity and enable accurate prospective prediction of likely SOMs.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC38XlvFCrsLk%253D&md5=99d070a003355b0caecdad0945077de7
30
Zaretzki, J.; Bergeron, C.; Huang, T.-W.; Rydberg, P.; Swamidass, S. J.; Breneman, C. M. RS-WebPredictor: A Server for Predicting CYP-Mediated Sites of Metabolism on Drug-like Molecules Bioinformatics 2013, 29, 497– 498 DOI: 10.1093/bioinformatics/bts705
[Crossref], [PubMed], [CAS], Google Scholar
30
RS-WebPredictor: a server for predicting CYP-mediated sites of metabolism on drug-like molecules
Zaretzki, Jed; Bergeron, Charles; Huang, Tao-wei; Rydberg, Patrik; Swamidass, S. Joshua; Breneman, Curt M.
Bioinformatics (2013), 29 (4), 497-498CODEN: BOINFP; ISSN:1367-4803. (Oxford University Press)
Summary: Regioselectivity-WebPredictor (RS-WebPredictor) is a server that predicts isoenzyme-specific cytochrome P 450 (CYP)-mediated sites of metab. (SOMs) on drug-like mols. Predictions may be made for the promiscuous 2C9, 2D6 and 3A4 CYP isoenzymes, as well as CYPs 1A2, 2A6, 2B6, 2C8, 2C19 and 2E1. RS-WebPredictor is the first freely accessible server that predicts the regioselectivity of the last six isoenzymes. Server execution time is fast, taking on av. 2s to encode a submitted mol. and 1s to apply a given model, allowing for high-throughput use in lead optimization projects.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3sXis1Ojtrw%253D&md5=bb01bdc036c5d338984efd663bd5ed78
31
Zaretzki, J.; Matlock, M.; Swamidass, S. J. XenoSite: Accurately Predicting CYP-Mediated Sites of Metabolism with Neural Networks J. Chem. Inf. Model. 2013, 53, 3373– 3383 DOI: 10.1021/ci400518g
[ACS Full Text ], [CAS], Google Scholar
31
XenoSite: Accurately Predicting CYP-Mediated Sites of Metabolism with Neural Networks
Zaretzki, Jed; Matlock, Matthew; Swamidass, S. Joshua
Journal of Chemical Information and Modeling (2013), 53 (12), 3373-3383CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
Understanding how xenobiotic mols. are metabolized is important because it influences the safety, efficacy, and dose of medicines and how they can be modified to improve these properties. The cytochrome P450s (CYPs) are proteins responsible for metabolizing 90% of drugs on the market, and many computational methods can predict which at. sites of a mol.-sites of metab. (SOMs)-are modified during CYP-mediated metab. This study improves on prior methods of predicting CYP-mediated SOMs by using new descriptors and machine learning based on neural networks. The new method, XenoSite, is faster to train and more accurate by as much as 4% or 5% for some isoenzymes. Furthermore, some "incorrect" predictions made by XenoSite were subsequently validated as correct predictions by revaluation of the source literature. Moreover, XenoSite output is interpretable as a probability, which reflects both the confidence of the model that a particular atom is metabolized and the statistical likelihood that its prediction for that atom is correct.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3sXhslKks77E&md5=1de6321b1500c9571db548006767e708
32
Matlock, M. K.; Hughes, T. B.; Swamidass, S. J. XenoSite Server: A Web-Available Site of Metabolism Prediction Tool Bioinformatics 2015, 31, 1136– 1137 DOI: 10.1093/bioinformatics/btu761
[Crossref], [PubMed], [CAS], Google Scholar
32
XenoSite server: a web-available site of metabolism prediction tool
Matlock, Matthew K.; Hughes, Tyler B.; Joshua, Swamidass S.
Bioinformatics (2015), 31 (7), 1136-1137CODEN: BOINFP; ISSN:1367-4803. (Oxford University Press)
Summary: Cytochrome P 450 enzymes (P450s) are metabolic enzymes that process the majority of FDA-approved, small-mol. drugs. Understanding how these enzymes modify mol. structure is key to the development of safe, effective drugs. XenoSite server is an online implementation of the XenoSite, a recently published computational model for P 450 metab. XenoSite predicts which at. sites of a mol.-sites of metab. (SOMs)-are modified by P450s. XenoSite server accepts input in common chem. file formats including SDF and SMILES and provides tools for visualizing the likelihood that each at. site is a site of metab. for a variety of important P450s, as well as a flat file download of SOM predictions.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC28Xht1Gntb7E&md5=8d7928ef60f9ef495f7cf776aadc4d0e
33
Kirchmair, J.; Williamson, M. J.; Afzal, A. M.; Tyzack, J. D.; Choy, A. P. K.; Howlett, A.; Rydberg, P.; Glen, R. C. FAst MEtabolizer (FAME): A Rapid and Accurate Predictor of Sites of Metabolism in Multiple Species by Endogenous Enzymes J. Chem. Inf. Model. 2013, 53, 2896– 2907 DOI: 10.1021/ci400503s
[ACS Full Text ], [CAS], Google Scholar
33
FAst MEtabolizer (FAME): A Rapid and Accurate Predictor of Sites of Metabolism in Multiple Species by Endogenous Enzymes
Kirchmair, Johannes; Williamson, Mark J.; Afzal, Avid M.; Tyzack, Jonathan D.; Choy, Alison P. K.; Howlett, Andrew; Rydberg, Patrik; Glen, Robert C.
Journal of Chemical Information and Modeling (2013), 53 (11), 2896-2907CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
FAst MEtabolizer (FAME) is a fast and accurate predictor of sites of metab. (SoMs). It is based on a collection of random forest models trained on diverse chem. data sets of >20,000 mols. annotated with their exptl. detd. SoMs. Using a comprehensive set of available data, FAME aims to assess metabolic processes from a holistic point of view. It is not limited to a specific enzyme family or species. Besides a global model, dedicated models are available for human, rat, and dog metab.; specific prediction of phase I and II metab. is also supported. FAME is able to identify at least one known SoM among the top-1, top-2, and top-3 highest ranked atom positions in up to 71%, 81%, and 87% of all cases tested, resp. These prediction rates are comparable to or better than SoM predictors focused on specific enzyme families (such as cytochrome P450s), despite the fact that FAME uses only seven chem. descriptors. FAME covers a very broad chem. space, which together with its inter- and extrapolation power makes it applicable to a wide range of chems. Predictions take <2.5 s per mol. in batch mode on an Ultrabook. Results are visualized using Jmol, with the most likely SoMs highlighted.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3sXhslGmt7rO&md5=131ead027225bea9851fba02e7b63a09
34
Tyzack, J. D.; Hunt, P. A.; Segall, M. D. Predicting Regioselectivity and Lability of Cytochrome P450 Metabolism Using Quantum Mechanical Simulations J. Chem. Inf. Model. 2016, 56, 2180– 2193 DOI: 10.1021/acs.jcim.6b00233
[ACS Full Text ], [CAS], Google Scholar
34
Predicting Regioselectivity and Lability of Cytochrome P450 Metabolism Using Quantum Mechanical Simulations
Tyzack, Jonathan D.; Hunt, Peter A.; Segall, Matthew D.
Journal of Chemical Information and Modeling (2016), 56 (11), 2180-2193CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
Methods are described for predicting cytochrome P 450 (CYP) metab. incorporating both pathway-specific reactivity and isoform-specific accessibility considerations. Semi-empirical quantum mech. (QM) simulations, parameterized using exptl. data and ab initio calcns., estd. the reactivity of each potential site of metab. in the context of the whole mol. Ligand-based models, trained using high quality regioselectivity data, cor. for orientation and steric effects of the different CYP isoform binding pockets. The resulting models identified a site of metab. in the top 2 predictions for between 82 and 91% of compds. in independent test sets across 7 CYP isoforms. In addn. to predicting the relative proportion of metabolite formation at each site, these methods estd. the activation energy at each site, from which addnl. information could be derived regarding their lability in abs. terms. The authors illustrated how this could guide the design of compds. to overcome issues with rapid CYP metab.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC28Xhs1yksrfO&md5=3f4daa77aacbfaefaec652c8ff456740
35
He, S.-B.; Li, M.-M.; Zhang, B.-X.; Ye, X.-T.; Du, R.-F.; Wang, Y.; Qiao, Y.-J. Construction of Metabolism Prediction Models for CYP450 3A4, 2D6, and 2C9 Based on Microsomal Metabolic Reaction System Int. J. Mol. Sci. 2016, 17, E1686 DOI: 10.3390/ijms17101686
[Crossref], [PubMed], [CAS], Google Scholar
35
Construction of metabolism prediction models for CYP450 3A4, 2D6, and 2C9 based on microsomal metabolic reaction system
He, Shuai-Bing; Li, Man-Man; Zhang, Bai-Xia; Ye, Xiao-Tong; Du, Ran-Feng; Wang, Yun; Qiao, Yan-Jiang
International Journal of Molecular Sciences (2016), 17 (10), 1686/1-1686/18CODEN: IJMCFK; ISSN:1422-0067. (MDPI AG)
During the past decades, there have been continuous attempts in the prediction of metab. mediated by cytochrome P450s (CYP450s) 3A4, 2D6, and 2C9. However, it has indeed remained a huge challenge to accurately predict the metab. of xenobiotics mediated by these enzymes. To address this issue, microsomal metabolic reaction system (MMRS)-a novel concept, which integrates information about site of metab. (SOM) and enzyme-was introduced. By incorporating the use of multiple feature selection (FS) techniques (ChiSquared (CHI), InfoGain (IG), GainRatio (GR), Relief) and hybrid classification procedures (Kstar, Bayes (BN), K-nearest neighbors (IBK), C4.5 decision tree (J48), RandomForest (RF), Support vector machines (SVM), AdaBoostM1, Bagging), metab. prediction models were established based on metab. data released by Sheridan et al. Four major biotransformations, including aliph. C-hydroxylation, arom. C-hydroxylation, N-dealkylation and O-dealkylation, were involved. For validation, the overall accuracies of all four biotransformations exceeded 0.95. For receiver operating characteristic (ROC) anal., each of these models gave a significant area under curve (AUC) value >0.98. In addn., an external test was performed based on dataset published previously. As a result, 87.7% of the potential SOMs were correctly identified by our four models. In summary, four MMRS-based models were established, which can be used to predict the metab. mediated by CYP3A4, 2D6, and 2C9 with high accuracy.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC28XitVahsLjJ&md5=9b44af1c8b3412453be673ab9775bc9f
36
Finkelmann, A. R.; Göller, A. H.; Schneider, G. Site of Metabolism Prediction Based on Ab Initio Derived Atom Representations ChemMedChem 2017, 12, 606– 612 DOI: 10.1002/cmdc.201700097
[Crossref], [PubMed], [CAS], Google Scholar
36
Site of Metabolism Prediction Based on ab initio Derived Atom Representations
Finkelmann, Arndt R.; Goeller, Andreas H.; Schneider, Gisbert
ChemMedChem (2017), 12 (8), 606-612CODEN: CHEMGX; ISSN:1860-7179. (Wiley-VCH Verlag GmbH & Co. KGaA)
Machine learning models for site of metab. (SoM) prediction offer the ability to identify metabolic soft spots in low-mol.-wt. drug mols. at low computational cost and enable data-based reactivity prediction. SoM prediction is an atom classification problem. Successful construction of machine learning models requires atom representations that capture the reactivity-detg. features of a potential reaction site. We have developed a descriptor scheme that characterizes an atom's steric and electronic environment and its relative location in the mol. structure. The partial charge distributions were obtained from fast quantum mech. calcns. We successfully trained machine learning classifiers on curated cytochrome P 450 metab. data. The models based on the new atom descriptors showed sustained accuracy for retrospective analyses of metab. optimization campaigns and lead optimization projects from Bayer Pharmaceuticals. The results obtained demonstrate the practicality of quantum-chem.-supported machine learning models for hit-to-lead optimization.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC2sXlsV2ltL8%253D&md5=5dbee6380ec1bed9bae7887550e9ec6c
37
Geurts, P.; Ernst, D.; Wehenkel, L. Extremely Randomized Trees Mach. Learn. 2006, 63, 3– 42 DOI: 10.1007/s10994-006-6226-1
[Crossref], Google Scholar
There is no corresponding record for this reference.
38
Tyzack, J. D.; Williamson, M. J.; Torella, R.; Glen, R. C. Prediction of Cytochrome P450 Xenobiotic Metabolism: Tethered Docking and Reactivity Derived from Ligand Molecular Orbital Analysis J. Chem. Inf. Model. 2013, 53, 1294– 1305 DOI: 10.1021/ci400058s
[ACS Full Text ], [CAS], Google Scholar
38
Prediction of Cytochrome P450 Xenobiotic Metabolism: Tethered Docking and Reactivity Derived from Ligand Molecular Orbital Analysis
Tyzack, Jonathan D.; Williamson, Mark J.; Torella, Rubben; Glen, Robert C.
Journal of Chemical Information and Modeling (2013), 53 (6), 1294-1305CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
Metab. of xenobiotic and endogenous compds. is frequently complex, not completely elucidated, and therefore often ambiguous. The prediction of sites of metab. (SoM) can be particularly helpful as a first step toward the identification of metabolites, a process esp. relevant to drug discovery. This paper describes a reactivity approach for predicting SoM whereby reactivity is derived directly from the ground state ligand MO anal., calcd. using D. Functional Theory, using a novel implementation of the av. local ionization energy. Thus each potential SoM is sampled in the context of the whole ligand, in contrast to other popular approaches where activation energies are calcd. for a predefined database of mol. fragments and assigned to matching moieties in a query ligand. In addn., one of the first descriptions of mol. dynamics of cytochrome P 450 (CYP) isoforms 3A4, 2D6, and 2C9 in their Compd. I state is reported, and, from the representative protein structures obtained, an anal. and evaluation of various docking approaches using GOLD is performed. In particular, a covalent docking approach is described coupled with the modeling of important electrostatic interactions between CYP and ligand using spherical constraints. Combining the docking and reactivity results, obtained using std. functionality from common docking and quantum chem. applications, enables a SoM to be identified in the top 2 predictions for 75%, 80%, and 78% of the data sets for 3A4, 2D6, and 2C9, resp., results that are accessible and competitive with other recently published prediction tools.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3sXot1Srtb0%253D&md5=8215558c1c9459a4d64f5ee88e16a3e3
39
Huang, T.-W.; Zaretzki, J.; Bergeron, C.; Bennett, K. P.; Breneman, C. M. DR-Predictor: Incorporating Flexible Docking with Specialized Electronic Reactivity and Machine Learning Techniques to Predict CYP-Mediated Sites of Metabolism J. Chem. Inf. Model. 2013, 53, 3352– 3366 DOI: 10.1021/ci4004688
[ACS Full Text ], [CAS], Google Scholar
39
DR-Predictor: Incorporating Flexible Docking with Specialized Electronic Reactivity and Machine Learning Techniques to Predict CYP-Mediated Sites of Metabolism
Huang, Tao-wei; Zaretzki, Jed; Bergeron, Charles; Bennett, Kristin P.; Breneman, Curt M.
Journal of Chemical Information and Modeling (2013), 53 (12), 3352-3366CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
Computational methods that can identify CYP-mediated sites of metab. (SOMs) of drug-like compds. have become required tools for early stage lead optimization. In recent years, methods that combine CYP binding site features with CYP/ligand binding information have been sought to increase the prediction accuracy of such hybrid models over those that use only one representation. Two challenges that any hybrid ligand/structure-based method must overcome are (1) identification of the best binding pose for a specific ligand with a given CYP and (2) appropriately incorporating the results of docking with ligand reactivity. To address these challenges the authors have created Docking-Regioselectivity-Predictor (DR-Predictor) - a method that incorporates flexible docking-derived information with specialized electronic reactivity and multiple-instance-learning methods to predict CYP-mediated SOMs. The hybrid ligand-structure-based DR-Predictor method was tested on substrate sets for CYP 1A2 and CYP 2A6. For these data, the DR-Predictor model was found to identify the exptl. obsd. SOM within the top two predicted rank-positions for 86% of the 261 1A2 substrates and 83% of the 100 2A6 substrates. Given the accuracy and extendibility of the DR-Predictor method, the authors anticipate that it will further facilitate the prediction of CYP metab. liabilities and aid in in-silico ADMET assessment of novel structures.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3sXhslWlsrzF&md5=1d0592336da0de3bd1374b5d5b8fd991
40
Zaretzki, J. M.; Browning, M. R.; Hughes, T. B.; Swamidass, S. J. Extending P450 Site-of-Metabolism Models with Region-Resolution Data Bioinformatics 2015, 31, 1966– 1973 DOI: 10.1093/bioinformatics/btv100
[Crossref], [PubMed], [CAS], Google Scholar
40
Extending P450 site-of-metabolism models with region-resolution data
Zaretzki, Jed M.; Browning, Michael R.; Hughes, Tyler B.; Swamidass, S. Joshua
Bioinformatics (2015), 31 (12), 1966-1973CODEN: BOINFP; ISSN:1367-4803. (Oxford University Press)
Motivation: Cytochrome P450s are a family of enzymes responsible for the metab. of approx. 90% of FDA-approved drugs. Medicinal chemists often want to know which atoms of a mol.-its metabolized sites-are oxidized by Cytochrome P450s in order to modify their metab. Consequently, there are several methods that use literature-derived, atom-resoln. data to train models that can predict a mol.'s sites of metab. There is, however, much more data available at a lower resoln., where the exact site of metab. is not known, but the region of the mol. that is oxidized is known. Until now, no site-of-metab. models made use of region- resoln. data. Results: Here, we describe XenoSite-Region, the first reported method for training site-of-metab. models with region-resoln. data. Our approach uses the Expectation Maximization algorithm to train a site-of-metab. model. Region-resoln. metab. data was simulated from a large site-of-metab. dataset, contg. 2000 mols. with 3400 metabolized and 30 000 un-metabolized sites and covering nine Cytochrome P 450 isoenzymes. When training on the same mols. (but with only region-level information), we find that this approach yields models almost as accurate as models trained with atom-resoln. data. Moreover, we find that atom-resoln. trained models are more accurate when also trained with region-resoln. data from addnl. mols. Our approach, therefore, opens up a way to extend the applicable domain of site-of-metab. models into larger regions of chem. space. This meets a crit. need in drug development by tapping into underutilized data commonly available in most large drug companies.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC28Xht1Cit77E&md5=5463287e64149487a190db5ad62f6bee
41
Powers, D. Evaluation: From Precision, Recall and F-Measure to ROC, Informedness, Markedness & Correlation J. Mach. Learn. Technol. 2011, 2, 37– 63
Google Scholar
There is no corresponding record for this reference.
42
Adams, S. E. Molecular Similarity and Xenobiotic Metabolism; University of Cambridge, 2010.
Google Scholar
There is no corresponding record for this reference.
43
Boyer, S.; Arnby, C. H.; Carlsson, L.; Smith, J.; Stein, V.; Glen, R. C. Reaction Site Mapping of Xenobiotic Biotransformations J. Chem. Inf. Model. 2007, 47, 583– 590 DOI: 10.1021/ci600376q
[ACS Full Text ], [CAS], Google Scholar
43
Reaction Site Mapping of Xenobiotic Biotransformations
Boyer, Scott; Arnby, Catrin Hasselgren; Carlsson, Lars; Smith, James; Stein, Viktor; Glen, Robert C.
Journal of Chemical Information and Modeling (2007), 47 (2), 583-590CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
Predictive metab. methods can be used in drug discovery projects to enhance the understanding of structure-metab. relationships. The present study uses data mining methods to exploit biotransformation data that have been recorded in the MDL Metabolite database. Reacting center fingerprints were derived from a comparison of substrates and their corresponding products listed in the database. This process yields two fingerprint databases: all atoms in all substrates and all reacting centers. The metabolic reaction data are then mined by submitting a new mol. and searching for fingerprint matches to every atom in the new mol. in both databases. An "occurrence ratio" is derived from the fingerprint matches between the submitted compd. and the reacting center and substrate fingerprint databases. Normalization of the occurrence ratio within each submitted mol. enables the results of the search to be rank-ordered as a measure of the relative frequency of a reaction occurring at a specific site within the submitted mol. Predictive performance that would allow this method to be used by drug discovery teams to generate useful hypotheses regarding structure metab. relationships was obsd.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD2sXhslaitrc%253D&md5=5feeecf97e20393b3fc3a36be29eae6d
44
Tyzack, J. D.; Mussa, H. Y.; Williamson, M. J.; Kirchmair, J.; Glen, R. C. Cytochrome P450 Site of Metabolism Prediction from 2D Topological Fingerprints Using GPU Accelerated Probabilistic Classifiers J. Cheminf. 2014, 6, 29 DOI: 10.1186/1758-2946-6-29
[Crossref], [CAS], Google Scholar
44
Cytochrome P450 site of metabolism prediction from 2D topological fingerprints using GPU accelerated probabilistic classifiers
Tyzack, Jonathan D.; Mussa, Hamse Y.; Williamson, Mark J.; Kirchmair, Johannes; Glen, Robert C.
Journal of Cheminformatics (2014), 6 (), 29/1-29/14, 14 pp.CODEN: JCOHB3; ISSN:1758-2946. (Chemistry Central Ltd.)
Background: The prediction of sites and products of metab. in xenobiotic compds. is key to the development of new chem. entities, where screening potential metabolites for toxicity or unwanted side-effects is of crucial importance. In this work 2D topol. fingerprints are used to encode at. sites and three probabilistic machine learning methods are applied: Parzen-Rosenblatt Window (PRW), Naive Bayesian (NB) and a novel approach called RASCAL (Random Attribute Subsampling Classification Algorithm). These are implemented by randomly subsampling descriptor space to alleviate the problem often suffered by data mining methods of having to exactly match fingerprints, and in the case of PRW by measuring a distance between feature vectors rather than exact matching. The classifiers have been implemented in CUDA/C++ to exploit the parallel architecture of graphical processing units (GPUs) and is freely available in a public repository. Results: It is shown that for PRW a SoM (Site of Metab.) is identified in the top two predictions for 85%, 91% and 88% of the CYP 3A4, 2D6 and 2C9 data sets resp., with RASCAL giving similar performance of 83%, 91% and 88%, resp. These results put PRW and RASCAL performance ahead of NB which gave a much lower classification performance of 51%, 73% and 74%, resp. Conclusions: 2D topol. fingerprints calcd. to a bond depth of 4-6 contain sufficient information to allow the identification of SoMs using classifiers based on relatively small data sets. Thus, the machine learning methods outlined in this paper are conceptually simpler and more efficient than other methods tested and the use of simple topol. descriptors derived from 2D structure give results competitive with other approaches using more expensive quantum chem. descriptors. The descriptor space subsampling approach and ensemble methodol. allow the methods to be applied to mols. more distant from the training data where data mining would be more likely to fail due to the lack of common fingerprints. The RASCAL algorithm is shown to give equiv. classification performance to PRW but at lower computational expense allowing it to be applied more efficiently in the ensemble scheme.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC2MXivFSks74%253D&md5=cb64fc4f7a201503285ead1d4d732c6b
45
Stewart, J. J. MOPAC: A Semiempirical Molecular Orbital Program J. Comput.-Aided Mol. Des. 1990, 4, 1– 105 DOI: 10.1007/BF00128336
[Crossref], [PubMed], [CAS], Google Scholar
45
MOPAC: a semiempirical molecular orbital program
Stewart, James J. P.
Journal of Computer-Aided Molecular Design (1990), 4 (1), 1-105CODEN: JCADEQ; ISSN:0920-654X.
An overview is presented of MOPAC, a semiempirical MO program for the study of chem. reactions involving mols., ions, and linear polymers. The program implements the semiempirical Hamiltonians MNDO, AM1, MINDO/3, and MNDO-PM3 and combines the calcns. of vibrational spectra, thermodn. quantities, isotopic substitution effects, and force consts. in a fully integrated program.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADyaK3cXksFyjurc%253D&md5=983366b4ba2797f0fd8a91aec6d366ed
46
MOPAC2016. http://openmopac.net/home.html (accessed Apr 7, 2017) .
Google Scholar
There is no corresponding record for this reference.
47
Schüürmann, G. Quantitative Structure-Property Relationships for the Polarizability, Solvatochromic Parameters and Lipophilicity Quant. Struct.-Act. Relat. 1990, 9, 326– 333 DOI: 10.1002/qsar.19900090406
[Crossref], [CAS], Google Scholar
47
Quantitative structure-property relationships for the polarizability, solvatochromic parameters and lipophilicity
Schueuermann, Gerrit
Quantitative Structure-Activity Relationships (1990), 9 (4), 326-33CODEN: QSARDI; ISSN:0931-8771.
The polarizability αvol, the solvatochromic parameters π* and β and the lipophilicity as expressed by log Kow are subjected to regression analyses using calcd. mol. parameters within the MNDO scheme. The resulting linear one- and more-variable regression equations enable rapid approx. calcns. for αvol, π* and β and log Kow of untested compds.; in particular, αvol is split into a mol. size and an electronic part which offers new possibilities for applications in quant. structure-activity relationships.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADyaK3MXhvFaqtrY%253D&md5=e7e356b88e9179b4698e7515c47e1f0e
48
Coulson, C. A.; Longuet-Higgins, H. C. The Electronic Structure of Conjugated Systems. II. Unsaturated Hydrocarbons and Their Hetero-Derivatives Proc. R. Soc. London, Ser. A 1947, 192, 16– 32 DOI: 10.1098/rspa.1947.0136
[Crossref], Google Scholar
There is no corresponding record for this reference.
49
Fukui, K.; Kato, H.; Yonezawa, T. A New Quantum-Mechanical Reactivity Index for Saturated Compounds Bull. Chem. Soc. Jpn. 1961, 34, 1111– 1115 DOI: 10.1246/bcsj.34.1111
[Crossref], Google Scholar
There is no corresponding record for this reference.
50
Gopinathan, M. S.; Siddarth, P.; Ravimohan, C. Valency and Molecular Structure Theor. Chim. Acta 1986, 70, 303– 322 DOI: 10.1007/BF00534237
[Crossref], Google Scholar
There is no corresponding record for this reference.
51
Mulliken, R. S. Electronic Population Analysis on LCAO-MO Molecular Wave Functions. I J. Chem. Phys. 1955, 23, 1833– 1840 DOI: 10.1063/1.1740588
[Crossref], [CAS], Google Scholar
51
Electronic population analysis on LCAO-MO [linear combination of atomic orbital-molecular orbital] molecular wave functions. I
Mulliken, R. S.
Journal of Chemical Physics (1955), 23 (), 1833-40CODEN: JCPSA6; ISSN:0021-9606.
An analysis in quant. form was given in terms of breakdowns of the electronic population into partial and total "gross at. populations," or into partial and total "net at. populations" together with "overlap populations." Gross at. populations distribute the electrons almost perfectly among the various at. orbitals of the various atoms in the mol. From these nos., a definite figure is obtained for the amt. of promotion (e.g. from 2s to 2p) in each atom; and also for the gross charge Q on each atom if the bonds are polar. The total overlap population for any pair of atoms in a mol. is in general made up of pos. and neg. contributions. If the total overlap population between 2 atoms is pos., they are bonded; if neg., they are antibonded. Tables of gross at. populations and overlap populations were calcd. for CO and H2O. The amt. of s-p promotion was nearly the same for the O atom in CO and in H2O (0.14 electron in CO and 0.15e in H2O). For the C atom in CO it is 0.50e. For the N atom in N2 it is 0.26e. In spite of very strong polarity in the π bonds in CO, the σ and π overlap populations are very similar to those in N. In CO the total overlap population for the π electrons is about twice that for the σ electrons. The most easily ionized electrons of CO are in a mol. orbital such that its gross at. population is 94% localized on the C atom; these electrons account for the weak electron donor properties of CO.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADyaG28Xnt1Kq&md5=5aab51c06ce6a3250219cf12d1b5f395
52
Mulliken, R. S. Criteria for the Construction of Good Self-Consistent-Field Molecular Orbital Wave Functions, and the Significance of LCAO-MO Population Analysis J. Chem. Phys. 1962, 36, 3428– 3439 DOI: 10.1063/1.1732476
[Crossref], [CAS], Google Scholar
52
Criteria for the construction of good self-consistent-field molecular orbital wave functions, and the significance of L.C.A.O.M.O. population analysis
Mulliken, R. S.
Journal of Chemical Physics (1962), 36 (), 3428-39CODEN: JCPSA6; ISSN:0021-9606.
Criteria for the optimal choice of finite linear combinations of Slater-type orbitals (S.T.O.) adequate to approx. closely S.C.F. M.O. were examd. in the light of computations on HF and other mols. Some aspects of the A.O. (generalized Heitler-London) method were discussed. The inherent limitations on the meaning of charges on atoms in a mol., or of degree of ionic character, were discussed. Unacceptable at. charges are found from an L.C.A.O. mol. orbital population analysis made on S.C.F. M.O. wave functions, approximated by using unbalanced S.T.O. sets, while acceptable results are obtained with judiciously balanced and sufficiently complete S.T.O. sets. The effects of insufficiently complete and unbalanced S.T.O. sets on the computed dipole moments of S.C.F. M.O. wave functions were discussed, with examples.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADyaF38Xks1KrtL8%253D&md5=9fb6c9a93a70358d02154b028efe63fc
53
Holm, S. A Simple Sequentially Rejective Multiple Test Procedure Scand. Stat. Theory Appl. 1979, 6, 65– 70
Google Scholar
There is no corresponding record for this reference.
54
Friedman, M. A Comparison of Alternative Tests of Significance for the Problem of m Rankings Ann. Math. Stat. 1940, 11, 86– 92 DOI: 10.1214/aoms/1177731944
[Crossref], Google Scholar
There is no corresponding record for this reference.
55
Friedman, M. A Correction: The Use of Ranks to Avoid the Assumption of Normality Implicit in the Analysis of Variance J. Am. Stat. Assoc. 1939, 34, 109– 109 DOI: 10.2307/2279169
[Crossref], Google Scholar
There is no corresponding record for this reference.
56
Friedman, M. The Use of Ranks to Avoid the Assumption of Normality Implicit in the Analysis of Variance J. Am. Stat. Assoc. 1937, 32, 675– 701 DOI: 10.1080/01621459.1937.10503522
[Crossref], Google Scholar
There is no corresponding record for this reference.
57
Shapiro, S. S.; Wilk, M. B. An Analysis of Variance Test for Normality (Complete Samples) Biometrika 1965, 52, 591– 611 DOI: 10.1093/biomet/52.3-4.591
[Crossref], Google Scholar
There is no corresponding record for this reference.
58
Mauchly, J. W. Significance Test for Sphericity of a Normal N-Variate Distribution Ann. Math. Stat. 1940, 11, 204– 209 DOI: 10.1214/aoms/1177731915
[Crossref], Google Scholar
There is no corresponding record for this reference.
59
Greenhouse, S. W.; Geisser, S. On Methods in the Analysis of Profile Data Psychometrika 1959, 24, 95– 112 DOI: 10.1007/BF02289823
[Crossref], Google Scholar
There is no corresponding record for this reference.
60
Huynh, H.; Feldt, L. S. Estimation of the Box Correction for Degrees of Freedom from Sample Data in Randomized Block and Split-Plot Designs J. Educ. Behav. Stat. 1976, 1, 69– 82 DOI: 10.3102/10769986001001069
[Crossref], Google Scholar
There is no corresponding record for this reference.
61
de Bruyn Kops, C.; Friedrich, N.-O.; Kirchmair, J. Alignment-Based Prediction of Sites of Metabolism J. Chem. Inf. Model. 2017, 57 (6) 1258– 1264
[ACS Full Text ], [CAS], Google Scholar
61
Alignment-Based Prediction of Sites of Metabolism
de Bruyn Kops, Christina; Friedrich, Nils-Ole; Kirchmair, Johannes
Journal of Chemical Information and Modeling (2017), 57 (6), 1258-1264CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
Prediction of metabolically labile atom positions in a mol. (sites of metab.) is a key component of the simulation of xenobiotic metab. as a whole, providing crucial information for the development of safe and effective drugs. In 2008, an exploratory study was published in which sites of metab. were derived based on mol. shape- and chem. feature-based alignment to a mol. whose site of metab. (SoM) had been detd. by expts. The authors present a detailed anal. of the breadth of applicability of alignment-based SoM prediction, including transfer of the approach from a structure- to ligand-based method and extension of the applicability of the models from cytochrome P 450 2C9 to all cytochrome P 450 isoenzymes involved in drug metab. The authors evaluate the effect of mol. similarity of the query and ref. mols. on the ability of this approach to accurately predict SoMs. In addn., the authors combine the alignment-based method with a leading chem. reactivity model to take reactivity into account. The combined model yielded superior performance in comparison to the alignment-based approach and the reactivity models with an av. area under the receiver operating characteristic curve of 0.85 in cross-validation expts. In particular, early enrichment was improved, as evidenced by higher BEDROC scores (mean BEDROC = 0.59 for α = 20.0, mean BEDROC = 0.73 for α = 80.5).
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC2sXnvV2hs7w%253D&md5=3a48e4bbfc64e569c9a9ff3d4040bc38
62
OMEGA, version 2.5.1.4; OpenEye Scientific Software: Santa Fe, NM, 2011; https://www.eyesopen.com (accessed Apr 7, 2017).
Google Scholar
There is no corresponding record for this reference.
63
Hawkins, P. C. D.; Skillman, A. G.; Warren, G. L.; Ellingson, B. A.; Stahl, M. T. Conformer Generation with OMEGA: Algorithm and Validation Using High Quality Structures from the Protein Databank and Cambridge Structural Database J. Chem. Inf. Model. 2010, 50, 572– 584 DOI: 10.1021/ci100031x
[ACS Full Text ], [CAS], Google Scholar
63
Conformer Generation with OMEGA: Algorithm and Validation Using High Quality Structures from the Protein Databank and Cambridge Structural Database
Hawkins, Paul C. D.; Skillman, A. Geoffrey; Warren, Gregory L.; Ellingson, Benjamin A.; Stahl, Matthew T.
Journal of Chemical Information and Modeling (2010), 50 (4), 572-584CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
Here, we present the algorithm and validation for OMEGA, a systematic, knowledge-based conformer generator. The algorithm consists of three phases: assembly of an initial 3D structure from a library of fragments; exhaustive enumeration of all rotatable torsions using values drawn from a knowledge-based list of angles, thereby generating a large set of conformations; and sampling of this set by geometric and energy criteria. Validation of conformer generators like OMEGA has often been undertaken by comparing computed conformer sets to exptl. mol. conformations from crystallog., usually from the Protein Databank (PDB). Such an approach is fraught with difficulty due to the systematic problems with small mol. structures in the PDB. Methods are presented to identify a diverse set of small mol. structures from cocomplexes in the PDB that has maximal reliability. A challenging set of 197 high quality, carefully selected ligand structures from well-solved models was obtained using these methods. This set will provide a sound basis for comparison and validation of conformer generators in the future. Validation results from this set are compared to the results using structures of a set of druglike mols. extd. from the Cambridge Structural Database (CSD). OMEGA is found to perform very well in reproducing the crystallog. conformations from both these data sets using two complementary metrics of success.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3cXjtlaisrY%253D&md5=fb87ecc9c51eddef63b41fffcd9babee
64
RDKit 2016.03.4. https://github.com/rdkit/rdkit/releases/tag/Release_2016_03_4 (Accessed April 7, 2017).
Google Scholar
There is no corresponding record for this reference.
65
Steinbeck, C.; Han, Y.; Kuhn, S.; Horlacher, O.; Luttmann, E.; Willighagen, E. The Chemistry Development Kit (CDK): An Open-Source Java Library for Chemo- and Bioinformatics J. Chem. Inf. Comput. Sci. 2003, 43, 493– 500 DOI: 10.1021/ci025584y
[ACS Full Text ], [CAS], Google Scholar
65
The Chemistry Development Kit (CDK): An Open-Source Java Library for Chemo- and Bioinformatics
Steinbeck, Christoph; Han, Yongquan; Kuhn, Stefan; Horlacher, Oliver; Luttmann, Edgar; Willighagen, Egon
Journal of Chemical Information and Computer Sciences (2003), 43 (2), 493-500CODEN: JCISD8; ISSN:0095-2338. (American Chemical Society)
The Chem. Development Kit (CDK) is a freely available open-source Java library for Structural Chemo- and Bioinformatics. Its architecture and capabilities as well as the development as an open-source project by a team of international collaborators from academic and industrial institutions is described. The CDK provides methods for many common tasks in mol. informatics, including 2D and 3D rendering of chem. structures, I/O routines, SMILES parsing and generation, ring searches, isomorphism checking, structure diagram generation, etc. Application scenarios as well as access information for interested users and potential contributors are given.
>> More from SciFinder ^®
https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD3sXhtVaktbg%253D&md5=afc8fd10783af301c73a8183727230bf
66
Chemistry Development Kit 1.4.19. https://github.com/cdk/cdk/releases/tag/cdk-1.4.19 (accessed Apr 7, 2017) .
Google Scholar
There is no corresponding record for this reference.
67
scikit-learn 0.18. http://scikit-learn.org/0.18/documentation.html (accessed Apr 7, 2017) .
Google Scholar
There is no corresponding record for this reference.

Cited By

This article is cited by 40 publications.

Yanjun Feng, Changda Gong, Jieyu Zhu, Guixia Liu, Yun Tang, Weihua Li. Prediction of Sites of Metabolism of CYP3A4 Substrates Utilizing Docking-Derived Geometric Features. Journal of Chemical Information and Modeling 2023, 63 (13) , 4158-4169. https://doi.org/10.1021/acs.jcim.3c00549
Mario Öeren, Sylvia C. Kaempf, David J. Ponting, Peter A. Hunt, Matthew D. Segall. Predicting Regioselectivity of Cytosolic Sulfotransferase Metabolism for Drugs. Journal of Chemical Information and Modeling 2023, 63 (11) , 3340-3349. https://doi.org/10.1021/acs.jcim.3c00275
Mario Öeren, Peter J. Walton, James Suri, David J. Ponting, Peter A. Hunt, Matthew D. Segall. Predicting Regioselectivity of AO, CYP, FMO, and UGT Metabolism Using Quantum Mechanical Simulations and Machine Learning. Journal of Medicinal Chemistry 2022, 65 (20) , 14066-14081. https://doi.org/10.1021/acs.jmedchem.2c01303
Siyang Tian, Xuan Cao, Russell Greiner, Carin Li, AnChi Guo, David S. Wishart. CyProduct: A Software Tool for Accurately Predicting the Byproducts of Human Cytochrome P450 Metabolism. Journal of Chemical Information and Modeling 2021, 61 (6) , 3128-3140. https://doi.org/10.1021/acs.jcim.1c00144
Christina de Bruyn Kops, Martin Šícho, Angelica Mazzolari, Johannes Kirchmair. GLORYx: Prediction of the Metabolites Resulting from Phase 1 and Phase 2 Biotransformations of Xenobiotics. Chemical Research in Toxicology 2021, 34 (2) , 286-299. https://doi.org/10.1021/acs.chemrestox.0c00224
Charleen G. Don, Martin Smieško. Deciphering Reaction Determinants of Altered-Activity CYP2D6 Variants by Well-Tempered Metadynamics Simulation and QM/MM Calculations. Journal of Chemical Information and Modeling 2020, 60 (12) , 6642-6653. https://doi.org/10.1021/acs.jcim.0c01091
Na Le Dang, Matthew K. Matlock, Tyler B. Hughes, S. Joshua Swamidass. The Metabolic Rainbow: Deep Learning Phase I Metabolism in Five Colors. Journal of Chemical Information and Modeling 2020, 60 (3) , 1146-1164. https://doi.org/10.1021/acs.jcim.9b00836
Xin Yang, Yifei Wang, Ryan Byrne, Gisbert Schneider, Shengyong Yang. Concepts of Artificial Intelligence for Computer-Assisted Drug Discovery. Chemical Reviews 2019, 119 (18) , 10520-10594. https://doi.org/10.1021/acs.chemrev.8b00728
Martin Šícho, Conrad Stork, Angelica Mazzolari, Christina de Bruyn Kops, Alessandro Pedretti, Bernard Testa, Giulio Vistoli, Daniel Svozil, Johannes Kirchmair. FAME 3: Predicting the Sites of Metabolism in Synthetic Compounds and Natural Products for Phase 1 and Phase 2 Metabolic Enzymes. Journal of Chemical Information and Modeling 2019, 59 (8) , 3400-3412. https://doi.org/10.1021/acs.jcim.9b00376
Angelica Mazzolari, Avid M. Afzal, Alessandro Pedretti, Bernard Testa, Giulio Vistoli, Andreas Bender. Prediction of UGT-mediated Metabolism Using the Manually Curated MetaQSAR Database. ACS Medicinal Chemistry Letters 2019, 10 (4) , 633-638. https://doi.org/10.1021/acsmedchemlett.8b00603
Andrea Volkamer, Sereina Riniker, Eva Nittinger, Jessica Lanini, Francesca Grisoni, Emma Evertsson, Raquel Rodríguez-Pérez, Nadine Schneider. Machine learning for small molecule drug discovery in academia and industry. Artificial Intelligence in the Life Sciences 2023, 3 , 100056. https://doi.org/10.1016/j.ailsci.2022.100056
Aditya Kate, Ekkita Seth, Ananya Singh, Chandrashekhar Mahadeo Chakole, Meenakshi Kanwar Chauhan, Ravi Kant Singh, Shrirang Maddalwar, Mohit Mishra. Artificial Intelligence for Computer-Aided Drug Discovery. Drug Research 2023, 63 https://doi.org/10.1055/a-2076-3359
Vincent-Alexander Scholz, Conrad Stork, Markus Frericks, Johannes Kirchmair. Computational prediction of the metabolites of agrochemicals formed in rats. Science of The Total Environment 2023, , 165039. https://doi.org/10.1016/j.scitotenv.2023.165039
Thi Tuyet Van Tran, Hilal Tayara, Kil To Chong. Artificial Intelligence in Drug Metabolism and Excretion Prediction: Recent Advances, Challenges, and Future Perspectives. Pharmaceutics 2023, 15 (4) , 1260. https://doi.org/10.3390/pharmaceutics15041260
Érika Yoko Suzuki, Alice Simon, Thaisa Francielle Souza Domingos, Bárbara de Azevedo Abrahim Vieira, Alessandra Mendonça Teles de Souza, Carlos Rangel Rodrigues, Valeria Pereira de Sousa, Flávia Almada do Carmo, Lucio Mendes Cabral. Alternative Methods for Pulmonary-Administered Drugs Metabolism: A Breath of Change. Mini-Reviews in Medicinal Chemistry 2023, 23 (2) , 170-186. https://doi.org/10.2174/1389557522666220620125623
Peter Ertl, Grégori Gerebtzoff, Richard Lewis, Hagen Muenkler, Nadine Schneider, Finton Sirockin, Nikolaus Stiefl, Paolo Tosco. Chemical Reactivity Prediction: Current Methods and Different Application Areas. Molecular Informatics 2022, 41 (6) https://doi.org/10.1002/minf.202100277
Christophe Muller, Obdulia Rabal, Constantino Diaz Gonzalez. Artificial Intelligence, Machine Learning, and Deep Learning in Real-Life Drug Design Cases. 2022, 383-407. https://doi.org/10.1007/978-1-0716-1787-8_16
Alan Talevi. Site of Metabolism Predictions. 2022, 1073-1081. https://doi.org/10.1007/978-3-030-84860-6_151
Jonathan D Tyzack. Prediction of Drug Metabolism: Use of Structural Biology and In Silico Tools. 2022, 769-791. https://doi.org/10.1016/B978-0-12-820472-6.00067-0
Ruben Goncalves, Romain Pelletier, Aurélien Couette, Thomas Gicquel, Brendan Le Daré. Suitability of high-resolution mass spectrometry in analytical toxicology: Focus on drugs of abuse. Toxicologie Analytique et Clinique 2022, 34 (1) , 29-41. https://doi.org/10.1016/j.toxac.2021.11.006
Mario Thevis, Thomas Piper, Andreas Thomas. Recent advances in identifying and utilizing metabolites of selected doping agents in human sports drug testing. Journal of Pharmaceutical and Biomedical Analysis 2021, 205 , 114312. https://doi.org/10.1016/j.jpba.2021.114312
Mary Alexandra Schleiff, Deepika Dhaware, Jasleen K. Sodhi. Recent advances in computational metabolite structure predictions and altered metabolic pathways assessment to inform drug development processes. Drug Metabolism Reviews 2021, 53 (2) , 173-187. https://doi.org/10.1080/03602532.2021.1910292
Mario Öeren, Peter J. Walton, Peter A. Hunt, David J. Ponting, Matthew D. Segall. Predicting reactivity to drug metabolism: beyond P450s—modelling FMOs and UGTs. Journal of Computer-Aided Molecular Design 2021, 35 (4) , 541-555. https://doi.org/10.1007/s10822-020-00321-1
Bekir Engin Eser, Yan Zhang, Li Zong, Zheng Guo. Self-sufficient Cytochrome P450s and their potential applications in biotechnology. Chinese Journal of Chemical Engineering 2021, 30 , 121-135. https://doi.org/10.1016/j.cjche.2020.12.002
Baddipadige Raju, Shalki Choudhary, Gera Narendra, Himanshu Verma, Om Silakari. Molecular modeling approaches to address drug-metabolizing enzymes (DMEs) mediated chemoresistance: a review. Drug Metabolism Reviews 2021, 53 (1) , 45-75. https://doi.org/10.1080/03602532.2021.1874406
Hyunho Kim, Eunyoung Kim, Ingoo Lee, Bongsung Bae, Minsu Park, Hojung Nam. Artificial Intelligence in Drug Discovery: A Comprehensive Review of Data-driven and Machine Learning Approaches. Biotechnology and Bioprocess Engineering 2020, 25 (6) , 895-930. https://doi.org/10.1007/s12257-020-0049-y
Santosh Malik, Ananya Ghosh, Rout George Kerry, Jyoti Ranjan Rout. Nanotechnology in Preclinical Pharmacokinetics. 2020, 461-478. https://doi.org/10.1007/978-981-15-2195-9_30
Hussam AL-barakati, Niraj Thapa, Saigo Hiroto, Kaushik Roy, Robert H. Newman, Dukka KC. RF-MaloSite and DL-Malosite: Methods based on random forest and deep learning to identify malonylation sites. Computational and Structural Biotechnology Journal 2020, 18 , 852-860. https://doi.org/10.1016/j.csbj.2020.02.012
Manavalan, Basith, Shin, Lee, Wei, Lee. 4mCpred-EL: An Ensemble Learning Framework for Identification of DNA N4-methylcytosine Sites in the Mouse Genome. Cells 2019, 8 (11) , 1332. https://doi.org/10.3390/cells8111332
Kazuma Kaitoh, Masaaki Kotera, Kimito Funatsu. Novel Electrotopological Atomic Descriptors for the Prediction of Xenobiotic Cytochrome P450 Reactions. Molecular Informatics 2019, 38 (10) https://doi.org/10.1002/minf.201900010
Christina de Bruyn Kops, Conrad Stork, Martin Šícho, Nikolay Kochev, Daniel Svozil, Nina Jeliazkova, Johannes Kirchmair. GLORY: Generator of the Structures of Likely Cytochrome P450 Metabolites Based on Predicted Sites of Metabolism. Frontiers in Chemistry 2019, 7 https://doi.org/10.3389/fchem.2019.00402
Hussam J. AL-barakati, Hiroto Saigo, Robert H. Newman, Dukka B. KC. RF-GlutarySite: a random forest based predictor for glutarylation sites. Molecular Omics 2019, 15 (3) , 189-204. https://doi.org/10.1039/C9MO00028C
Balachandran Manavalan, Shaherin Basith, Tae Hwan Shin, Leyi Wei, Gwang Lee. Meta-4mCpred: A Sequence-Based Meta-Predictor for Accurate DNA 4mC Site Prediction Using Effective Feature Representation. Molecular Therapy - Nucleic Acids 2019, 16 , 733-744. https://doi.org/10.1016/j.omtn.2019.04.019
Leonardo L.G. Ferreira, Adriano D. Andricopulo. ADMET modeling approaches in drug discovery. Drug Discovery Today 2019, 24 (5) , 1157-1165. https://doi.org/10.1016/j.drudis.2019.03.015
Yanmin Zhang, Yuchen Wang, Weineng Zhou, Yuanrong Fan, Junnan Zhao, Lu Zhu, Shuai Lu, Tao Lu, Yadong Chen, Haichun Liu. A combined drug discovery strategy based on machine learning and molecular docking. Chemical Biology & Drug Design 2019, 93 (5) , 685-699. https://doi.org/10.1111/cbdd.13494
Jonathan D. Tyzack, Johannes Kirchmair. Computational methods and tools to predict cytochrome P450 metabolism for drug discovery. Chemical Biology & Drug Design 2019, 93 (4) , 377-386. https://doi.org/10.1111/cbdd.13445
Marco Montefiori, Casper Lyngholm-Kjærby, Anthony Long, Lars Olsen, Flemming Steen Jørgensen. Fast Methods for Prediction of Aldehyde Oxidase-Mediated Site-of-Metabolism. Computational and Structural Biotechnology Journal 2019, 17 , 345-351. https://doi.org/10.1016/j.csbj.2019.03.003
Magdalena Galster, Marius Löppenberg, Fabian Galla, Frederik Börgel, Oriana Agoglitta, Johannes Kirchmair, Ralph Holl. Phenylethylene glycol-derived LpxC inhibitors with diverse Zn2+-binding groups. Tetrahedron 2019, 75 (4) , 486-509. https://doi.org/10.1016/j.tet.2018.12.011
Arndt R. Finkelmann, Daria Goldmann, Gisbert Schneider, Andreas H. Göller. MetScore: Site of Metabolism Prediction Beyond Cytochrome P450 Enzymes. ChemMedChem 2018, 13 (21) , 2281-2289. https://doi.org/10.1002/cmdc.201800309
Balachandran Manavalan, Rajiv Gandhi Govindaraj, Tae Hwan Shin, Myeong Ok Kim, Gwang Lee. iBCE-EL: A New Ensemble Learning Framework for Improved Linear B-Cell Epitope Prediction. Frontiers in Immunology 2018, 9 https://doi.org/10.3389/fimmu.2018.01695

Download PDF

Abstract
High Resolution Image
Download MS PowerPoint Slide
Figure 1
Figure 1. Overview of the data preparation, model building, and model evaluation workflow.
High Resolution Image
Download MS PowerPoint Slide
Figure 2
Figure 2. Example showing how circular atom type fingerprints and descriptors of up to bond depth 3 were calculated for a single atom. In this example, neighboring atoms up to three bonds away from the sp³ hybridized oxygen were encoded. Two examples of data matrix parts corresponding to neighbors encoded in bond depths 0 and 3 are shown. In both the fingerprints and descriptors, atoms were grouped by their atom type (highlighted in blue in the column names) in each bond depth (highlighted in orange). In the case of the fingerprints, the occurrences of each atom type were noted, while for real-valued descriptors, the average value of the basic descriptor among the grouped atoms was calculated and assigned to the corresponding atom type. Therefore, for the sigmaElectronegativity descriptor in this example, the third layer would contain 8.35 (8.35/1 = 8.35) for the sp³ hybridized nitrogen type and 8.31 ((8.31 + 8.31)/2 = 8.31) for the aromatic carbon type. Consequently, what can be derived from this matrix fragment is that there are two aromatic carbons (which are topologically identical) within three bonds of the oxygen atom that have a sigmaElectronegativity equal to 8.31 each.
High Resolution Image
Download MS PowerPoint Slide
Figure 3
Figure 3. Visualization of the relationship between maximum bond depth of circular descriptors and model performance on the (a) training set (measured as the mean MCC across the 10 folds in cross-validation of the optimized model), (b) independent test set, (c) uncorrelated test set, and (d) full Zaretzki data set (measured as the mean MCC across the 10 folds in cross-validation of the optimized model). Models that do not use circular descriptors are shown to have a bond depth equal to 0.
High Resolution Image
Download MS PowerPoint Slide
Figure 4
Figure 4. Consensus matrix combining results from parametric and nonparametric posthoc tests. The pairs of models for which the null hypothesis (the equality of means for t test or similar performance ranking for Nemenyi test) was rejected at an α level of 0.05 by both the parametric and nonparametric method are highlighted in teal.
High Resolution Image
Download MS PowerPoint Slide
Figure 5
Figure 5. Point plot which represents the expected performance of each model by indicating 95% confidence intervals for mean MCC. The confidence intervals were obtained from a statistical analysis of predictions of 100 random subsamples (50 molecules each, sampled with replacement) of the original test set.
High Resolution Image
Download MS PowerPoint Slide
Figure 6
Figure 6. Histogram of maximum Tanimoto similarities (a) between compounds in the training set and the independent test set (136 molecules) and (b) between compounds within the uncorrelated test set (71 molecules).
High Resolution Image
Download MS PowerPoint Slide
Figure 7
Figure 7. Example of prediction visualization by FAME 2.
High Resolution Image
Download MS PowerPoint Slide
References
ARTICLE SECTIONS
Jump To
This article references 67 other publications.
1. 1
  Kirchmair, J.; Howlett, A.; Peironcely, J. E.; Murrell, D. S.; Williamson, M. J.; Adams, S. E.; Hankemeier, T.; van Buren, L.; Duchateau, G.; Klaffke, W.; Glen, R. C. How Do Metabolites Differ from Their Parent Molecules and How Are They Excreted? J. Chem. Inf. Model. 2013, 53, 354– 367 DOI: 10.1021/ci300487z
  [ACS Full Text ], [CAS], Google Scholar
  1
  How Do Metabolites Differ from Their Parent Molecules and How Are They Excreted?
  Kirchmair, Johannes; Howlett, Andrew; Peironcely, Julio E.; Murrell, Daniel S.; Williamson, Mark J.; Adams, Samuel E.; Hankemeier, Thomas; van Buren, Leo; Duchateau, Guus; Klaffke, Werner; Glen, Robert C.
  Journal of Chemical Information and Modeling (2013), 53 (2), 354-367CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
  Understanding which physicochem. properties, or property distributions, are favorable for successful design and development of drugs, nutritional supplements, cosmetics, and agrochems. is of great importance. In this study the authors have analyzed mols. from three distinct chem. spaces (i) approved drugs, (ii) human metabolites, and (iii) traditional Chinese medicine (TCM) to investigate four aspects detg. the disposition of small org. mols. First, the authors examd. the physicochem. properties of these three classes of mols. and identified characteristic features resulting from their distinctive biol. functions. For example, human metabolites and TCM mols. can be larger and more hydrophobic than drugs, which makes them less likely to cross membranes. The authors then quantified the shifts in physicochem. property space induced by metab. from a holistic perspective by analyzing a data set of several thousand exptl. obsd. metabolic trees. Results show how the metabolic system aims to retain nutrients/micronutrients while facilitating a rapid elimination of xenobiotics. In the third part the authors compared these global shifts with the contributions made by individual metabolic reactions. For better resoln., all reactions were classified into phase I and phase II biotransformations. Interestingly, not all metabolic reactions lead to more hydrophilic mols. The authors were able to identify biotransformations leading to an increase of logP by more than one log unit, which could be used for the design of drugs with enhanced efficacy. The study closes with the anal. of the physicochem. properties of metabolites found in the bile, feces, and urine. Metabolites in the bile can be large and are often neg. charged. Mols. with mol. wt. >500 Da are rarely found in the urine, and most of these large mols. are charged phase II conjugates.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3sXhsVemtrY%253D&md5=ea563b68312736c7736b40939d7620a6
2. 2
  Testa, B.; Pedretti, A.; Vistoli, G. Reactions and Enzymes in the Metabolism of Drugs and Other Xenobiotics Drug Discovery Today 2012, 17, 549– 560 DOI: 10.1016/j.drudis.2012.01.017
  [Crossref], [PubMed], [CAS], Google Scholar
  2
  Reactions and enzymes in the metabolism of drugs and other xenobiotics
  Testa, Bernard; Pedretti, Alessandro; Vistoli, Giulio
  Drug Discovery Today (2012), 17 (11-12), 549-560CODEN: DDTOFS; ISSN:1359-6446. (Elsevier Ltd.)
  In this article, we offer an overview of the compared quant. importance of biotransformation reactions in the metab. of drugs and other xenobiotics, based on a meta-anal. of current research interests. Also, we assess the relative significance the enzyme (super)families or categories catalyzing these reactions. We put the facts unveiled by the anal. into a drug discovery context and draw some implications. The results confirm the primary role of cytochrome P 450-catalyzed oxidns. and UDP-glucuronosyl-catalyzed glucuronidations, but they also document the marked significance of several other reactions. Thus, there is a need for several drug discovery scientists to better grasp the variety of drug metab. reactions and enzymes and their consequences.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC38XnslOiur8%253D&md5=e11b5b1b7ea18e8bc817bff83efad456
3. 3
  Kirchmair, J.; Göller, A. H.; Lang, D.; Kunze, J.; Testa, B.; Wilson, I. D.; Glen, R. C.; Schneider, G. Predicting Drug Metabolism: Experiment and/or Computation? Nat. Rev. Drug Discovery 2015, 14, 387– 404 DOI: 10.1038/nrd4581
  [Crossref], [PubMed], [CAS], Google Scholar
  3
  Predicting drug metabolism: experiment and/or computation?
  Kirchmair, Johannes; Goeller, Andreas H.; Lang, Dieter; Kunze, Jens; Testa, Bernard; Wilson, Ian D.; Glen, Robert C.; Schneider, Gisbert
  Nature Reviews Drug Discovery (2015), 14 (6), 387-404CODEN: NRDDAG; ISSN:1474-1776. (Nature Publishing Group)
  A review. Drug metab. can produce metabolites with physicochem. and pharmacol. properties that differ substantially from those of the parent drug, and consequently has important implications for both drug safety and efficacy. To reduce the risk of costly clin.-stage attrition due to the metabolic characteristics of drug candidates, there is a need for efficient and reliable ways to predict drug metab. in vitro, in silico and in vivo. In this Perspective, we provide an overview of the state of the art of exptl. and computational approaches for investigating drug metab. We highlight the scope and limitations of these methods, and indicate strategies to harvest the synergies that result from combining measurement and prediction of drug metab.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC2MXntFCju7k%253D&md5=231060e7860047948cffd071b8a57197
4. 4
  Campagna-Slater, V.; Pottel, J.; Therrien, E.; Cantin, L.-D.; Moitessier, N. Development of a Computational Tool to Rival Experts in the Prediction of Sites of Metabolism of Xenobiotics by P450s J. Chem. Inf. Model. 2012, 52, 2471– 2483 DOI: 10.1021/ci3003073
  [ACS Full Text ], [CAS], Google Scholar
  4
  Development of a Computational Tool to Rival Experts in the Prediction of Sites of Metabolism of Xenobiotics by P450s
  Campagna-Slater, Valerie; Pottel, Joshua; Therrien, Eric; Cantin, Louis-David; Moitessier, Nicolas
  Journal of Chemical Information and Modeling (2012), 52 (9), 2471-2483CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
  The metab. of xenobiotics - and more specifically drugs - in the liver is a crit. process controlling their half-life. Although there exist exptl. methods, which measure the metabolic stability of xenobiotics and identify their metabolites, developing higher throughput predictive methods is an avenue of research. It is expected that predicting the chem. nature of the metabolites would be an asset for designing safer drugs and/or drugs with modulated half-lives. We have developed IMPACTS (In-silico Metab. Prediction by Activated Cytochromes and Transition States), a computational tool combining docking to metabolic enzymes, transition state modeling, and rule-based substrate reactivity prediction to predict the site of metab. (SoM) of xenobiotics. Its application to sets of CYP1A2, CYP2C9, CYP2D6, and CYP3A4 substrates and comparison to experts' predictions demonstrates its accuracy and significance. IMPACTS identified an exptl. obsd. SoM in the top 2 predicted sites for 77% of the substrates, while the accuracy of biotransformation experts' prediction was 65%. Application of IMPACTS to external sets and comparison of its accuracy to those of eleven other methods further validated the method implemented in IMPACTS.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC38Xht1CisrvP&md5=c954b56c19c486106cb7fec0f51fe68a
5. 5
  Crivori, P.; Poggesi, I. Computational Approaches for Predicting CYP-Related Metabolism Properties in the Screening of New Drugs Eur. J. Med. Chem. 2006, 41, 795– 808 DOI: 10.1016/j.ejmech.2006.03.003
  [Crossref], [PubMed], [CAS], Google Scholar
  5
  Computational approaches for predicting CYP-related metabolism properties in the screening of new drugs
  Crivori, P.; Poggesi, I.
  European Journal of Medicinal Chemistry (2006), 41 (7), 795-808CODEN: EJMCA5; ISSN:0223-5234. (Elsevier B.V.)
  A review. The site of biotransformation, the extent and rate of metab. and the no. of active metabolic pathways are among the most important characteristics of the pharmacokinetics of a drug. The catalytic activity of drug metabolizing enzymes is likely the most influential determinant of the pharmacokinetic variability. Metabolic stability is the prerequisite for sustaining the therapeutically relevant concns. Metabolic inhibition and induction can give rise to clin. important drug-drug interactions. A variety of computational approaches are currently available for predicting different cytochrome P 450 (CYP)-related metab. endpoints. The present review will describe these approaches and their impact on drug development process. Indications on the available software for the implementation will also be given.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD28XmvFeltro%253D&md5=1de13b7845575861e4f5ef85af3d809b
6. 6
  Tarcsay, Á.; Keserű, G. M. In Silico Site of Metabolism Prediction of Cytochrome P450-Mediated Biotransformations Expert Opin. Drug Metab. Toxicol. 2011, 7, 299– 312 DOI: 10.1517/17425255.2011.553599
  [Crossref], [PubMed], [CAS], Google Scholar
  6
  In silico site of metabolism prediction of cytochrome P450-mediated biotransformations
  Tarcsay, Akos; Keseru, Gyorgy M.
  Expert Opinion on Drug Metabolism & Toxicology (2011), 7 (3), 299-312CODEN: EODMAP; ISSN:1742-5255. (Informa Healthcare)
  A review. Preclin. research involves the in vitro monitoring of metabolic stability to deliver compds. with improved ADME profiles. Prediction of the metabolically vulnerable points can substantially help in analyzing CYP-mediated metab. data and support optimization efforts in drug discovery programs. Moreover, fast and reliable in silico predictions could accelerate the characterization of in vitro/in vivo metabolites. This paper reviews in silico methods available for CYP-mediated site of metab. (SOM) prediction. Comprehensive and practical knowledge in this field can guide the identification of best practice and may inspire ideas for the development of novel approaches. Comparison of the efficacy of SOM prediction methodologies revealed the general dependency on the studied isoform and substrate set. Increasing knowledge on P 450 X-ray structures, on biotransformations and on the mechanistic details of the catalytic cycle revolutionized the prediction of SOM. Although no ultimate soln. exits, combined methods covering both steric and electronic effects are preferred on most of the pharmaceutically relevant isoforms.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3MXit1Wku7k%253D&md5=f3631600a6ebeee3c7338521ccdb7424
7. 7
  Kirchmair, J.; Williamson, M. J.; Tyzack, J. D.; Tan, L.; Bond, P. J.; Bender, A.; Glen, R. C. Computational Prediction of Metabolism: Sites, Products, SAR, P450 Enzyme Dynamics, and Mechanisms J. Chem. Inf. Model. 2012, 52, 617– 648 DOI: 10.1021/ci200542m
  [ACS Full Text ], [CAS], Google Scholar
  7
  Computational Prediction of Metabolism: Sites, Products, SAR, P450 Enzyme Dynamics, and Mechanisms
  Kirchmair, Johannes; Williamson, Mark J.; Tyzack, Jonathan D.; Tan, Lu; Bond, Peter J.; Bender, Andreas; Glen, Robert C.
  Journal of Chemical Information and Modeling (2012), 52 (3), 617-648CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
  A review perspective. Metab. of xenobiotics remains a central challenge for the discovery and development of drugs, cosmetics, nutritional supplements, and agrochems. Metabolic transformations are frequently related to the incidence of toxic effects that may result from the emergence of reactive species, the systemic accumulation of metabolites, or by induction of metabolic pathways. Exptl. investigation of the metab. of small org. mols. is particularly resource demanding; hence, computational methods are of considerable interest to complement exptl. approaches. This review provides a broad overview of structure- and ligand-based computational methods for the prediction of xenobiotic metab. Current computational approaches to address xenobiotic metab. are discussed from three major perspectives: (i) prediction of sites of metab. (SOMs), (ii) elucidation of potential metabolites and their chem. structures, and (iii) prediction of direct and indirect effects of xenobiotics on metabolizing enzymes, where the focus is on the cytochrome P 450 (CYP) superfamily of enzymes, the cardinal xenobiotics metabolizing enzymes. For each of these domains, a variety of approaches and their applications are systematically reviewed, including expert systems, data mining approaches, quant. structure-activity relationships (QSARs), and machine learning-based methods, pharmacophore-based algorithms, shape-focused techniques, mol. interaction fields (MIFs), reactivity-focused techniques, protein-ligand docking, mol. dynamics (MD) simulations, and combinations of methods. Predictive metab. is a developing area, and there is still enormous potential for improvement. However, it is clear that the combination of rapidly increasing amts. of available ligand- and structure-related exptl. data (in particular, quant. data) with novel and diverse simulation and modeling approaches is accelerating the development of effective tools for prediction of in vivo metab., which is reflected by the diverse and comprehensive data sources and methods for metab. prediction reviewed here. This review attempts to survey the range and scope of computational methods applied to metab. prediction and also to compare and contrast their applicability and performance.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC38XisVyitL4%253D&md5=d144f6054d9f27774f476d026a066805
8. 8
  Raunio, H.; Kuusisto, M.; Juvonen, R. O.; Pentikäinen, O. T. Modeling of Interactions between Xenobiotics and Cytochrome P450 (CYP) Enzymes Front. Pharmacol. 2015, 6, 123 DOI: 10.3389/fphar.2015.00123
  [Crossref], [PubMed], [CAS], Google Scholar
  8
  Modeling of interactions between xenobiotics and cytochrome P450 (CYP) enzymes
  Raunio Hannu; Juvonen Risto O; Kuusisto Mira; Pentikainen Olli T
  Frontiers in pharmacology (2015), 6 (), 123 ISSN:1663-9812.
  The adverse effects to humans and environment of only few chemicals are well known. Absorption, distribution, metabolism, and excretion (ADME) are the steps of pharmaco/toxicokinetics that determine the internal dose of chemicals to which the organism is exposed. Of all the xenobiotic-metabolizing enzymes, the cytochrome P450 (CYP) enzymes are the most important due to their abundance and versatility. Reactions catalyzed by CYPs usually turn xenobiotics to harmless and excretable metabolites, but sometimes an innocuous xenobiotic is transformed into a toxic metabolite. Data on ADME and toxicity properties of compounds are increasingly generated using in vitro and modeling (in silico) tools. Both physics-based and empirical modeling approaches are used. Numerous ligand-based and target-based as well as combined modeling methods have been employed to evaluate determinants of CYP ligand binding as well as predicting sites of metabolism and inhibition characteristics of test molecules. In silico prediction of CYP-ligand interactions have made crucial contributions in understanding (1) determinants of CYP ligand binding recognition and affinity; (2) prediction of likely metabolites from substrates; (3) prediction of inhibitors and their inhibition potency. Truly predictive models of toxic outcomes cannot be created without incorporating metabolic characteristics; in silico methods help producing such information and filling gaps in experimentally derived data. Currently modeling methods are not mature enough to replace standard in vitro and in vivo approaches, but they are already used as an important component in risk assessment of drugs and other chemicals.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A280%3ADC%252BC2MbosFGiuw%253D%253D&md5=aaf28501a8579f1a24ec8e5c80f24a3a
9. 9
  Bezhentsev, V. M.; Tarasova, O. A.; Dmitriev, A. V.; Rudik, A. V.; Lagunin, A. A.; Filimonov, D. A.; Poroikov, V. V. Computer-Aided Prediction of Xenobiotic Metabolism in the Human Body Russ. Chem. Rev. 2016, 85, 854 DOI: 10.1070/RCR4614
  [Crossref], [CAS], Google Scholar
  9
  Computer-aided prediction of xenobiotic metabolism in humans
  Bezhentsev, Vladislav M.; Tarasova, Olga A.; Dmitriev, Aleksandr V.; Rudik, Anastasiya V.; Lagunin, Aleksey A.; Filimonov, Dmitriy A.; Poroikov, Vladimir V.
  Russian Chemical Reviews (2016), 85 (8), 854-879CODEN: RCRVAB; ISSN:0036-021X. (IOP Publishing Ltd.)
  A review. The review describes the main databases contg. information about the metab. of xenobiotics, including data on drug metab., metabolic enzymes, schemes of biotransformation and the structures of some substrates and metabolites. Computational approaches used to predict the interaction of xenobiotics with metabolic enzymes, to identify the sites of metab. in the mol. and to generate structures of potential metabolites for subsequent evaluation of their properties are considered. The advantages and limitations of particular computational methods for metab. prediction are indicated and the prospects for their applications to improve the safety and efficacy of new drugs are discussed.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC28XhvF2jsb7L&md5=4ea739a940a38faeaabadba470f2cd05
10. 10
  Rydberg, P. Reactivity-Based Approaches and Machine Learning Methods for Predicting the Sites of Cytochrome P450-Mediated Metabolism. In Drug Metabolism Prediction; Kirchmair, J., Ed.; Wiley-VCH Verlag GmbH & Co. KGaA: Weinheim, 2014; pp 265– 292.
  [Crossref], Google Scholar
  There is no corresponding record for this reference.
11. 11
  Rydberg, P.; Olsen, L. Predicting Drug Metabolism by Cytochrome P450 2C9: Comparison with the 2D6 and 3A4 Isoforms ChemMedChem 2012, 7, 1202– 1209 DOI: 10.1002/cmdc.201200160
  [Crossref], [PubMed], [CAS], Google Scholar
  11
  Predicting Drug Metabolism by Cytochrome P450 2C9: Comparison with the 2D6 and 3A4 Isoforms
  Rydberg, Patrik; Olsen, Lars
  ChemMedChem (2012), 7 (7), 1202-1209, S1202/1-S1202/33CODEN: CHEMGX; ISSN:1860-7179. (Wiley-VCH Verlag GmbH & Co. KGaA)
  By the use of knowledge gained through modeling of drug metab. mediated by the cytochrome P 450 2D6 and 3A4 isoforms, we constructed a 2D-based model for site-of-metab. prediction for the cytochrome P 450 2C9 isoform. The similarities and differences between the models for the 2C9 and 2D6 isoforms are discussed through structural knowledge from the X-ray crystal structures and trends in exptl. data. The final model was validated on an independent test set, resulting in an area under the curve value of 0.92, and a site of metab. was found among the top two ranked atoms for 77 % of the compds.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC38XntFWitr0%253D&md5=0dd476527c405af9de49a076a30c1181
12. 12
  Darvas, F. Predicting Metabolic Pathways by Logic Programming J. Mol. Graphics 1988, 6, 80– 86 DOI: 10.1016/0263-7855(88)85004-5
  [Crossref], [CAS], Google Scholar
  12
  Predicting metabolic pathways by logic programming
  Darvas, Ferenc
  Journal of Molecular Graphics (1988), 6 (2), 80-6CODEN: JMGRDV; ISSN:0263-7855.
  A discussion of logic programming is presented and its application is described in an expert system used to simulate the metabolic fate of substances. An expert system called Metabolexpert accepts the formula of the compd. to be metabolized and produces a treelike picture of the metabolites generated together with the formula of the compds. On request, 3-dimensional pictures of the metabolites are also displayed and hydrophobicity values of the compds. calcd. A retrospective investigation of Metabolexpert's achievement showed that the expert system can reproduce almost all primary, secondary, and tertiary metabolites of amphetamine. A compd. series has been suggested for benchmark testing of metabolic transformation knowledge bases.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADyaL1cXkslaltL8%253D&md5=1d73455bb8ab8d32c02dff7b3a4d78a4
13. 13
  Klopman, G.; Dimayuga, M.; Talafous, J. META. 1. A Program for the Evaluation of Metabolic Transformation of Chemicals J. Chem. Inf. Model. 1994, 34, 1320– 1325 DOI: 10.1021/ci00022a014
  [ACS Full Text ], [CAS], Google Scholar
  13
  META. 1. A Program for the Evaluation of Metabolic Transformation of Chemicals
  Klopman, Gilles; Dimayuga, Mario; Talafous, Joseph
  Journal of Chemical Information and Computer Sciences (1994), 34 (6), 1320-5CODEN: JCISD8; ISSN:0095-2338.
  A new metab. program, META, is introduced. In this paper, the basic principles on which the program operates are described. META is an expert system, capable of predicting the sites of potential enzymic attack and the nature of the chems. formed by such metabolic transformations. It operates from dictionaries of transformation operators, created by experts to represent known metabolic paths.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADyaK2cXmsVyitrw%253D&md5=9f145c8b540affcaf4a67b6306b991aa
14. 14
  Talafous, J.; Sayre, L. M.; Mieyal, J. J.; Klopman, G. META. 2. A Dictionary Model of Mammalian Xenobiotic Metabolism J. Chem. Inf. Model. 1994, 34, 1326– 1333 DOI: 10.1021/ci00022a015
  [ACS Full Text ], Google Scholar
  There is no corresponding record for this reference.
15. 15
  Greene, N.; Judson, P. N.; Langowski, J. J.; Marchant, C. A. Knowledge-Based Expert Systems for Toxicity and Metabolism Prediction: DEREK, StAR and METEOR SAR QSAR Environ. Res. 1999, 10, 299– 314 DOI: 10.1080/10629369908039182
  [Crossref], [PubMed], [CAS], Google Scholar
  15
  Knowledge-based expert systems for toxicity and metabolism prediction: DEREK, StAR and METEOR
  Greene, N.; Judson, P. N.; Langowski, J. J.; Marchant, C. A.
  SAR and QSAR in Environmental Research (1999), 10 (2-3), 299-314, 2 platesCODEN: SQERED; ISSN:1062-936X. (Gordon & Breach Science Publishers)
  It has long been recognized that the ability to predict the metabolic fate of a chem. substance and the potential toxicity of either the parent compd. or its metabolites are important in novel drug design. The popularity of using computer models as an aid in this area has grown considerably in recent years. LHASA Limited has been developing knowledge-based expert systems for toxicity and metab. prediction in collaboration with industry and regulatory authorities. These systems, DEREK, StAR, and METEOR, use rules to describe the relationship between chem. structure and either toxicity in the case of DEREK and StAR or metabolic fate in the case of METEOR. The rule refinement process for DEREK often involves assessing the predictions for a novel set of compds. and comparing them to their biol. assay results as a measure of the system's performance. For example, 266 non-congeneric chems. from the National Toxicol. Program database have been processed through the DEREK mutagenicity knowledge base and the predictions compared to their Salmonella typhimurium mutagenicity data. Initially, 81 of 114 mutagens (71%) and 117 of 152 non-mutagens (77%) were correctly identified. Following further knowledge base development, the no. of correctly identified mutagens has increased to 96 (84%). Further work on improving the predictive capabilities of DEREK, StAR, and METEOR is in progress.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADyaK1MXlvFCgsbc%253D&md5=5988de2a40351b9416bc21a9688a3a6d
16. 16
  Hou, B. K.; Wackett, L. P.; Ellis, L. B. M. Microbial Pathway Prediction: A Functional Group Approach J. Chem. Inf. Comput. Sci. 2003, 43, 1051– 1057 DOI: 10.1021/ci034018f
  [ACS Full Text ], [CAS], Google Scholar
  16
  Microbial pathway prediction: a functional group approach
  Hou, Bo Kyeng; Wackett, Lawrence P.; Ellis, Lynda B. M.
  Journal of Chemical Information and Computer Sciences (2003), 43 (3), 1051-1057CODEN: JCISD8; ISSN:0095-2338. (American Chemical Society)
  We have developed a system to predict microbial catabolism, using the University of Minnesota Biocatalysis/Biodegrdn. Database (UM-BBD, http://umbbd.ahc.umn.edu/) as a knowledge base. The present system, available on the Web (http://umbbd.ahc.umn.edu/predict/), can predict biodegrdn. of most of the major aliph. and arom. org. functional groups contg. C, H, N, O, and halogens. It can duplicate at least one known biodegrdn. pathway for 60% of the compds. in a 84-member validation set; most pathways that did not completely duplicate known metab. could plausibly occur in nature. Users are encouraged, and have begun, to submit addnl. biotransformation rules and comment on existing rules; the system will further develop under the direction of the scientific community.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD3sXis1Kgt74%253D&md5=c7a314ccbde95b8ceef6af0276469840
17. 17
  Hatzimanikatis, V.; Li, C.; Ionita, J. A.; Henry, C. S.; Jankowski, M. D.; Broadbelt, L. J. Exploring the Diversity of Complex Metabolic Networks Bioinformatics 2005, 21, 1603– 1609 DOI: 10.1093/bioinformatics/bti213
  [Crossref], [PubMed], [CAS], Google Scholar
  17
  Exploring the diversity of complex metabolic networks
  Hatzimanikatis, Vassily; Li, Chunhui; Ionita, Justin A.; Henry, Christopher S.; Jankowski, Matthew D.; Broadbelt, Linda J.
  Bioinformatics (2005), 21 (8), 1603-1609CODEN: BOINFP; ISSN:1367-4803. (Oxford University Press)
  Motivation: Metab., the network of chem. reactions that make life possible, is one of the most complex processes in nature. We describe here the development of a computational approach for the identification of every possible biochem. reaction from a given set of enzyme reaction rules that allows the de novo synthesis of metabolic pathways composed of these reactions, and the evaluation of these novel pathways with respect to their thermodn. properties. Results: We applied this framework to the anal. of the arom. amino acid pathways and discovered almost 75,000 novel biochem. routes from chorismate to phenylalanine, more than 350,000 from chorismate to tyrosine, but only 13 from chorismate to tryptophan. Thermodn. anal. of these pathways suggests that the native pathways are thermodynamically more favorable than the alternative possible pathways. The pathways generated involve compds. that exist in biol. databases, as well as compds. that exist in chem. databases and novel compds., suggesting novel biochem. routes for these compds. and the existence of biochem. compds. that remain to be discovered or synthesized through enzyme and pathway engineering.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD2MXjtlGhuro%253D&md5=343be8db9eada86fcba34ae047529dec
18. 18
  Ridder, L.; Wagener, M. SyGMa: Combining Expert Knowledge and Empirical Scoring in the Prediction of Metabolites ChemMedChem 2008, 3, 821– 832 DOI: 10.1002/cmdc.200700312
  [Crossref], [PubMed], [CAS], Google Scholar
  18
  SyGMa: combining expert knowledge and empirical scoring in the prediction of metabolites
  Ridder, Lars; Wagener, Markus
  ChemMedChem (2008), 3 (5), 821-832CODEN: CHEMGX; ISSN:1860-7179. (Wiley-VCH Verlag GmbH & Co. KGaA)
  Predictions of potential metabolites based on chem. structure are becoming increasingly important in drug discovery to guide medicinal chem. efforts that address metabolic issues and to support exptl. metabolite screening and identification. Herein we present a novel rule-based method, SyGMa (Systematic Generation of potential Metabolites), to predict the potential metabolites of a given parent structure. A set of reaction rules covering a broad range of phase 1 and phase 2 metab. has been derived from metabolic reactions reported in the Metabolite Database to occur in humans. An empirical probability score is assigned to each rule representing the fraction of correctly predicted metabolites in the training database. This score is used to refine the rules and to rank predicted metabolites. The current rule set of SyGMa covers approx. 70% of biotransformation reactions obsd. in humans. Evaluation of the rule-based predictions demonstrated a significant enrichment of true metabolites in the top of the ranking list: while in total, 68% of all obsd. metabolites in an independent test set were reproduced by SyGMa, a large part, 30% of the obsd. metabolites, were identified among the top three predictions. From a subset of cytochrome P 450 specific metabolites, 84% were reproduced overall, with 66% in the top three predicted phase 1 metabolites. A similarity anal. of the reactions present in the database was performed to obtain an overview of the metabolic reactions predicted by SyGMa and to support ongoing efforts to extend the rules. Specific examples demonstrate the use of SyGMa in exptl. metabolite identification and the application of SyGMa to suggest chem. modifications that improve the metabolic stability of compds.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD1cXms1emtLw%253D&md5=92ed84b2e97af5ad9a50beef151fa5db
19. 19
  Gao, J.; Ellis, L. B. M.; Wackett, L. P. The University of Minnesota Pathway Prediction System: Multi-Level Prediction and Visualization Nucleic Acids Res. 2011, 39, W406– W411 DOI: 10.1093/nar/gkr200
  [Crossref], [PubMed], [CAS], Google Scholar
  19
  The University of Minnesota Pathway Prediction System: multi-level prediction and visualization
  Gao, Junfeng; Ellis, Lynda B. M.; Wackett, Lawrence P.
  Nucleic Acids Research (2011), 39 (Web Server), W406-W411CODEN: NARHAD; ISSN:0305-1048. (Oxford University Press)
  The University of Minnesota Pathway Prediction System (UM-PPS, http://umbbd.msi.umn.edu/predict/) is a rule-based system that predicts microbial catabolism of org. compds. Currently, its knowledge base contains 250 biotransformation rules and five types of metabolic logic entities. The original UM-PPS predicted up to two prediction levels at a time. Users had to choose a predicted product to continue the prediction. This approach provided a limited view of prediction results and heavily relied on manual intervention. The new UM-PPS produces a multi-level prediction within an acceptable time frame, and allows users to view prediction alternatives much more easily as a directed acyclic graph.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3MXosVOmsLs%253D&md5=4b27083f37d70ab33ba63c1e27a2959b
20. 20
  Mu, F.; Unkefer, C. J.; Unkefer, P. J.; Hlavacek, W. S. Prediction of Metabolic Reactions Based on Atomic and Molecular Properties of Small-Molecule Compounds Bioinformatics 2011, 27, 1537– 1545 DOI: 10.1093/bioinformatics/btr177
  [Crossref], [PubMed], [CAS], Google Scholar
  20
  Prediction of metabolic reactions based on atomic and molecular properties of small-molecule compounds
  Mu, Fangping; Unkefer, Clifford J.; Unkefer, Pat J.; Hlavacek, William S.
  Bioinformatics (2011), 27 (11), 1537-1545CODEN: BOINFP; ISSN:1367-4803. (Oxford University Press)
  Our knowledge of the metabolites in cells and their reactions is far from complete as revealed by metabolomic measurements that detect many more small mols. than are documented in metabolic databases. Here, we develop an approach for predicting the reactivity of small-mol. metabolites in enzyme-catalyzed reactions that combines expert knowledge, computational chem. and machine learning. We classified 4843 reactions documented in the KEGG database, from all six Enzyme Commission classes (EC 1-6), into 80 reaction classes, each of which is marked by a characteristic functional group transformation. Reaction centers and surrounding local structures in substrates and products of these reactions were represented using SMARTS. We found that each of the SMARTS-defined chem. substructures is widely distributed among metabolites, but only a fraction of the functional groups in these substructures are reactive. Using at. properties of atoms in a putative reaction center and mol. properties as features, we trained support vector machine (SVM) classifiers to discriminate between functional groups that are reactive and non-reactive. Classifier accuracy was assessed by cross-validation anal. A typical sensitivity [TP/(TP+FN)] or specificity [TN/(TN+FP)] is ≈0.8. Our results suggest that metabolic reactivity of small-mol. compds. can be predicted with reasonable accuracy based on the presence of a potentially reactive functional group and the chem. features of its local environment. The classifiers presented here can be used to predict reactions via a web site (http://cellsignaling.lanl.gov/Reactivity/). Contact: [email protected].
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3MXmvVajsLs%253D&md5=97709401a9db367e98b2278a9d601816
21. 21
  Yousofshahi, M.; Manteiga, S.; Wu, C.; Lee, K.; Hassoun, S. PROXIMAL: A Method for Prediction of Xenobiotic Metabolism BMC Syst. Biol. 2015, 9, 94 DOI: 10.1186/s12918-015-0241-4
  [Crossref], [PubMed], [CAS], Google Scholar
  21
  PROXIMAL: a method for Prediction of Xenobiotic Metabolism
  Yousofshahi, Mona; Manteiga, Sara; Wu, Charmian; Lee, Kyongbum; Hassoun, Soha
  BMC Systems Biology (2015), 9 (), 94/1-94/17CODEN: BSBMCC; ISSN:1752-0509. (BioMed Central Ltd.)
  Background: Contamination of the environment with bioactive chems. has emerged as a potential public health risk. These substances that may cause distress or disease in humans can be found in air, water and food supplies. An open question is whether these chems. transform into potentially more active or toxic derivs. via xenobiotic metabolizing enzymes expressed in the body. We present a new prediction tool, which we call PROXIMAL (Prediction of Xenobiotic Metab.) for identifying possible transformation products of xenobiotic chems. in the liver. Using reaction data from DrugBank and KEGG, PROXIMAL builds look-up tables that catalog the sites and types of structural modifications performed by Phase I and Phase II enzymes. Given a compd. of interest, PROXIMAL searches for substructures that match the sites cataloged in the look-up tables, applies the corresponding modifications to generate a panel of possible transformation products, and ranks the products based on the activity and abundance of the enzymes involved. Results: PROXIMAL generates transformations that are specific for the chem. of interest by analyzing the chem.'s substructures. We evaluate the accuracy of PROXIMAL's predictions through case studies on two environmental chems. with suspected endocrine disrupting activity, bisphenol A (BPA) and 4-chlorobiphenyl (PCB3). Comparisons with published reports confirm 5 out of 7 and 17 out of 26 of the predicted derivs. for BPA and PCB3, resp. We also compare biotransformation predictions generated by PROXIMAL with those generated by METEOR and Metaprint2D-react, two other prediction tools. Conclusions: PROXIMAL can predict transformations of chems. that contain substructures recognizable by human liver enzymes. It also has the ability to rank the predicted metabolites based on the activity and abundance of enzymes involved in xenobiotic transformation.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC2sXjtVyltbo%253D&md5=131eaa3ba02457ab89a73d1481d38ea1
22. 22
  Sun, H.; Scott, D. O. Structure-Based Drug Metabolism Predictions for Drug Design Chem. Biol. Drug Des. 2010, 75, 3– 17 DOI: 10.1111/j.1747-0285.2009.00899.x
  [Crossref], [PubMed], [CAS], Google Scholar
  22
  Structure-based drug metabolism predictions for drug design
  Sun, Hao; Scott, Dennis O.
  Chemical Biology & Drug Design (2010), 75 (1), 3-17CODEN: CBDDAL; ISSN:1747-0277. (Wiley-Blackwell)
  A review. Significant progress has been made in structure-based drug design by pharmaceutical companies at different stages of drug discovery such as identifying new hits, enhancing mol. binding affinity in hit-to-lead, and reducing toxicities in lead optimization. Drug metab. is a major consideration for modifying drug clearance and also a primary source for drug metabolite-induced toxicity. With major cytochrome P 450 structures identified and characterized recently, structure-based drug metab. prediction becomes increasingly attractive. In silico methods based on mol. and quantum mechanics such as docking, mol. dynamics and ab initio chem. reactivity calcns. bring us closer to understand drug metab. and predict drug-drug interactions. In this study, we review important progress in drug metab. and common in silico techniques adopted to predict drug regioselectivity, stereoselectivity, reactive metabolites, induction, inhibition and mechanism-based inactivation, as well as their implementation in hit-to-lead drug discovery.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD1MXhs1WhsbfO&md5=52666dc1d960035e9def127cb1963860
23. 23
  Kingsley, L. J.; Wilson, G. L.; Essex, M. E.; Lill, M. A. Combining Structure- and Ligand-Based Approaches to Improve Site of Metabolism Prediction in CYP2C9 Substrates Pharm. Res. 2015, 32, 986– 1001 DOI: 10.1007/s11095-014-1511-3
  [Crossref], Google Scholar
  There is no corresponding record for this reference.
24. 24
  Sheridan, R. P.; Korzekwa, K. R.; Torres, R. A.; Walker, M. J. Empirical Regioselectivity Models for Human Cytochromes P450 3A4, 2D6, and 2C9 J. Med. Chem. 2007, 50, 3173– 3184 DOI: 10.1021/jm0613471
  [ACS Full Text ], [CAS], Google Scholar
  24
  Empirical Regioselectivity Models for Human Cytochromes P450 3A4, 2D6, and 2C9
  Sheridan, Robert P.; Korzekwa, Kenneth R.; Torres, Rhonda A.; Walker, Matthew J.
  Journal of Medicinal Chemistry (2007), 50 (14), 3173-3184CODEN: JMCMAR; ISSN:0022-2623. (American Chemical Society)
  Cytochromes P 450 3A4, 2D6, and 2C9 metabolize a large fraction of drugs. Knowing where these enzymes will preferentially oxidize a mol., the regioselectivity, allows medicinal chemists to plan how best to block its metab. The authors present QSAR-based regioselectivity models for these enzymes calibrated against compiled literature data of drugs and drug-like compds. These models are purely empirical and use only the structures of the substrates, in contrast to those models that simulate a specific mechanism like hydrogen radical abstraction, and/or use explicit models of active sites. The authors most predictive models use three substructure descriptors and two phys. property descriptors. Descriptor importance from the random forest QSAR method show that other factors than the immediate chem. environment and the accessibility of the hydrogen affect regioselectivity in all three isoforms. The cross-validated predictions of the models are compared to predictions from the authors earlier mechanistic model (Singh et al. J. Med. Chem. 2003, 46, 1330-1336) and predictions from MetaSite (Cruciani et al. J. Med. Chem. 2005, 48, 6970-6979).
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD2sXmsFGnsr0%253D&md5=645bd5ddeb0459636bf6067e0194dd34
25. 25
  Rydberg, P.; Gloriam, D. E.; Zaretzki, J.; Breneman, C.; Olsen, L. SMARTCyp: A 2D Method for Prediction of Cytochrome P450-Mediated Drug Metabolism ACS Med. Chem. Lett. 2010, 1, 96– 100 DOI: 10.1021/ml100016x
  [ACS Full Text ], [CAS], Google Scholar
  25
  SMARTCyp: A 2D Method for Prediction of Cytochrome P450-Mediated Drug Metabolism
  Rydberg, Patrik; Gloriam, David E.; Zaretzki, Jed; Breneman, Curt; Olsen, Lars
  ACS Medicinal Chemistry Letters (2010), 1 (3), 96-100CODEN: AMCLCT; ISSN:1948-5875. (American Chemical Society)
  SMARTCyp is an in silico method that predicts the sites of cytochrome P 450-mediated metab. of druglike mols. The method is foremost a reactivity model, and as such, it shows a preference for predicting sites that are metabolized by the cytochrome P 450 3A4 isoform. SMARTCyp predicts the site of metab. directly from the 2D structure of a mol., without requiring calcn. of electronic properties or generation of 3D structures. This is a major advantage, because it makes SMARTCyp very fast. Other advantages are that exptl. data are not a prerequisite to create the model, and it can easily be integrated with other methods to create models for other cytochrome P 450 isoforms. Benchmarking tests on a database of 394 3A4 substrates show that SMARTCyp successfully identifies at least one metabolic site in the top two ranked positions 76% of the time. SMARTCyp is available for download at http://www.farma.ku.dk/P 450.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3cXjtFOqsL0%253D&md5=4761de933c2c3cef5da2ffb29ac076d0
26. 26
  Rydberg, P.; Gloriam, D. E.; Olsen, L. The SMARTCyp Cytochrome P450 Metabolism Prediction Server Bioinformatics 2010, 26, 2988– 2989 DOI: 10.1093/bioinformatics/btq584
  [Crossref], [PubMed], [CAS], Google Scholar
  26
  The SMARTCyp cytochrome P450 metabolism prediction server
  Rydberg, Patrik; Gloriam, David E.; Olsen, Lars
  Bioinformatics (2010), 26 (23), 2988-2989CODEN: BOINFP; ISSN:1367-4803. (Oxford University Press)
  The SMARTCyp server is the first web application for site of metab. prediction of cytochrome P 450-mediated drug metab.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3cXhsVKjtL3I&md5=a8689591d976ca80d0ae6a3e15b32636
27. 27
  Rydberg, P.; Rostkowski, M.; Gloriam, D. E.; Olsen, L. The Contribution of Atom Accessibility to Site of Metabolism Models for Cytochromes P450 Mol. Pharmaceutics 2013, 10, 1216– 1223 DOI: 10.1021/mp3005116
  [ACS Full Text ], [CAS], Google Scholar
  27
  The Contribution of Atom Accessibility to Site of Metabolism Models for Cytochromes P450
  Rydberg, Patrik; Rostkowski, Michal; Gloriam, David E.; Olsen, Lars
  Molecular Pharmaceutics (2013), 10 (4), 1216-1223CODEN: MPOHBP; ISSN:1543-8384. (American Chemical Society)
  Three different types of atom accessibility descriptors are investigated in relation to site of metab. predictions. To enable the integration of local accessibility we have constructed 2DSASA, a method for the calcn. of the at. solvent accessible surface area that is independent of 3D coordinates. The method was implemented in the SMARTCyp site of metab. prediction models and improved the results by up to 4 percentage points for nine cytochrome P 450 isoforms. The final models are made available at http://www.farma.ku.dk/smartcyp.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3sXhtFyjs7c%253D&md5=206ed0792c118423008bca0e6a793096
28. 28
  Zaretzki, J.; Bergeron, C.; Rydberg, P.; Huang, T.-W.; Bennett, K. P.; Breneman, C. M. RS-Predictor: A New Tool for Predicting Sites of Cytochrome P450-Mediated Metabolism Applied to CYP 3A4 J. Chem. Inf. Model. 2011, 51, 1667– 1689 DOI: 10.1021/ci2000488
  [ACS Full Text ], [CAS], Google Scholar
  28
  RS-Predictor: A New Tool for Predicting Sites of Cytochrome P450-Mediated Metabolism Applied to CYP 3A4
  Zaretzki, Jed; Bergeron, Charles; Rydberg, Patrik; Huang, Tao-wei; Bennett, Kristin P.; Breneman, Curt M.
  Journal of Chemical Information and Modeling (2011), 51 (7), 1667-1689CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
  This article describes RegioSelectivity-Predictor (RS-Predictor), a new in silico method for generating predictive models of P 450-mediated metab. for drug-like compds. Within this method, potential sites of metab. (SOMs) are represented as "metabolophores": A concept that describes the hierarchical combination of topol. and quantum chem. descriptors needed to represent the reactivity of potential metabolic reaction sites. RS-Predictor modeling involves the use of metabolophore descriptors together with multiple-instance ranking (MIRank) to generate an optimized descriptor wt. vector that encodes regioselectivity trends across all cases in a training set. The resulting pathway-independent (O-dealkylation vs. N-oxidn. vs. Csp3 hydroxylation, etc.), isoenzyme-specific regioselectivity model may be used to predict potential metabolic liabilities. In the present work, cross-validated RS-Predictor models were generated for a set of 394 substrates of CYP 3A4 as a proof-of-principle for the method. Rank aggregation was then employed to merge independently generated predictions for each substrate into a single consensus prediction. The resulting consensus RS-Predictor models were shown to reliably identify at least one obsd. site of metab. in the top two rank-positions on 78% of the substrates. Comparisons between RS-Predictor and previously described regioselectivity prediction methods reveal new insights into how in silico metabolite prediction methods should be compared.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3MXnsVertLs%253D&md5=3d863c866d45fc1c8e90d033a16ed6a1
29. 29
  Zaretzki, J.; Rydberg, P.; Bergeron, C.; Bennett, K. P.; Olsen, L.; Breneman, C. M. RS-Predictor Models Augmented with SMARTCyp Reactivities: Robust Metabolic Regioselectivity Predictions for Nine CYP Isozymes J. Chem. Inf. Model. 2012, 52, 1637– 1659 DOI: 10.1021/ci300009z
  [ACS Full Text ], [CAS], Google Scholar
  29
  RS-Predictor Models Augmented with SMARTCyp Reactivities: Robust Metabolic Regioselectivity Predictions for Nine CYP Isozymes
  Zaretzki, Jed; Rydberg, Patrik; Bergeron, Charles; Bennett, Kristin P.; Olsen, Lars; Breneman, Curt M.
  Journal of Chemical Information and Modeling (2012), 52 (6), 1637-1659CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
  RS-Predictor is a tool for creating pathway-independent, isoenzyme-specific, site of metab. (SOM) prediction models using any set of known cytochrome P 450 (CYP) substrates and metabolites. Until now, the RS-Predictor method was only trained and validated on CYP 3A4 data, but in the present study, we report on the versatility the RS-Predictor modeling paradigm by creating and testing regioselectivity models for substrates of the nine most important CYP isoenzymes. Through curation of source literature, we have assembled 680 substrates distributed among CYPs 1A2, 2A6, 2B6, 2C19, 2C8, 2C9, 2D6, 2E1, and 3A4, the largest publicly accessible collection of P 450 ligands and metabolites released to date. A comprehensive investigation into the importance of different descriptor classes for identifying the regioselectivity mediated by each isoenzyme is made through the generation of multiple independent RS-Predictor models for each set of isoenzyme substrates. Two of these models include a d. functional theory (DFT) reactivity descriptor derived from SMARTCyp. Optimal combinations of RS-Predictor and SMARTCyp are shown to have stronger performance than either method alone, while also exceeding the accuracy of the com. regioselectivity prediction methods distributed by Optibrium and Schroedinger, correctly identifying a large proportion of the metabolites in each substrate set within the top two rank-positions: 1A2 (83.0%), 2A6 (85.7%), 2B6 (82.1%), 2C19 (86.2%), 2C8 (83.8%), 2C9 (84.5%), 2D6 (85.9%), 2E1 (82.8%), 3A4 (82.3%), and merged (86.0%). Comprehensive datamining of each substrate set and careful statistical analyses of the predictions made by the different models revealed new insights into mol. features that control metabolic regioselectivity and enable accurate prospective prediction of likely SOMs.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC38XlvFCrsLk%253D&md5=99d070a003355b0caecdad0945077de7
30. 30
  Zaretzki, J.; Bergeron, C.; Huang, T.-W.; Rydberg, P.; Swamidass, S. J.; Breneman, C. M. RS-WebPredictor: A Server for Predicting CYP-Mediated Sites of Metabolism on Drug-like Molecules Bioinformatics 2013, 29, 497– 498 DOI: 10.1093/bioinformatics/bts705
  [Crossref], [PubMed], [CAS], Google Scholar
  30
  RS-WebPredictor: a server for predicting CYP-mediated sites of metabolism on drug-like molecules
  Zaretzki, Jed; Bergeron, Charles; Huang, Tao-wei; Rydberg, Patrik; Swamidass, S. Joshua; Breneman, Curt M.
  Bioinformatics (2013), 29 (4), 497-498CODEN: BOINFP; ISSN:1367-4803. (Oxford University Press)
  Summary: Regioselectivity-WebPredictor (RS-WebPredictor) is a server that predicts isoenzyme-specific cytochrome P 450 (CYP)-mediated sites of metab. (SOMs) on drug-like mols. Predictions may be made for the promiscuous 2C9, 2D6 and 3A4 CYP isoenzymes, as well as CYPs 1A2, 2A6, 2B6, 2C8, 2C19 and 2E1. RS-WebPredictor is the first freely accessible server that predicts the regioselectivity of the last six isoenzymes. Server execution time is fast, taking on av. 2s to encode a submitted mol. and 1s to apply a given model, allowing for high-throughput use in lead optimization projects.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3sXis1Ojtrw%253D&md5=bb01bdc036c5d338984efd663bd5ed78
31. 31
  Zaretzki, J.; Matlock, M.; Swamidass, S. J. XenoSite: Accurately Predicting CYP-Mediated Sites of Metabolism with Neural Networks J. Chem. Inf. Model. 2013, 53, 3373– 3383 DOI: 10.1021/ci400518g
  [ACS Full Text ], [CAS], Google Scholar
  31
  XenoSite: Accurately Predicting CYP-Mediated Sites of Metabolism with Neural Networks
  Zaretzki, Jed; Matlock, Matthew; Swamidass, S. Joshua
  Journal of Chemical Information and Modeling (2013), 53 (12), 3373-3383CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
  Understanding how xenobiotic mols. are metabolized is important because it influences the safety, efficacy, and dose of medicines and how they can be modified to improve these properties. The cytochrome P450s (CYPs) are proteins responsible for metabolizing 90% of drugs on the market, and many computational methods can predict which at. sites of a mol.-sites of metab. (SOMs)-are modified during CYP-mediated metab. This study improves on prior methods of predicting CYP-mediated SOMs by using new descriptors and machine learning based on neural networks. The new method, XenoSite, is faster to train and more accurate by as much as 4% or 5% for some isoenzymes. Furthermore, some "incorrect" predictions made by XenoSite were subsequently validated as correct predictions by revaluation of the source literature. Moreover, XenoSite output is interpretable as a probability, which reflects both the confidence of the model that a particular atom is metabolized and the statistical likelihood that its prediction for that atom is correct.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3sXhslKks77E&md5=1de6321b1500c9571db548006767e708
32. 32
  Matlock, M. K.; Hughes, T. B.; Swamidass, S. J. XenoSite Server: A Web-Available Site of Metabolism Prediction Tool Bioinformatics 2015, 31, 1136– 1137 DOI: 10.1093/bioinformatics/btu761
  [Crossref], [PubMed], [CAS], Google Scholar
  32
  XenoSite server: a web-available site of metabolism prediction tool
  Matlock, Matthew K.; Hughes, Tyler B.; Joshua, Swamidass S.
  Bioinformatics (2015), 31 (7), 1136-1137CODEN: BOINFP; ISSN:1367-4803. (Oxford University Press)
  Summary: Cytochrome P 450 enzymes (P450s) are metabolic enzymes that process the majority of FDA-approved, small-mol. drugs. Understanding how these enzymes modify mol. structure is key to the development of safe, effective drugs. XenoSite server is an online implementation of the XenoSite, a recently published computational model for P 450 metab. XenoSite predicts which at. sites of a mol.-sites of metab. (SOMs)-are modified by P450s. XenoSite server accepts input in common chem. file formats including SDF and SMILES and provides tools for visualizing the likelihood that each at. site is a site of metab. for a variety of important P450s, as well as a flat file download of SOM predictions.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC28Xht1Gntb7E&md5=8d7928ef60f9ef495f7cf776aadc4d0e
33. 33
  Kirchmair, J.; Williamson, M. J.; Afzal, A. M.; Tyzack, J. D.; Choy, A. P. K.; Howlett, A.; Rydberg, P.; Glen, R. C. FAst MEtabolizer (FAME): A Rapid and Accurate Predictor of Sites of Metabolism in Multiple Species by Endogenous Enzymes J. Chem. Inf. Model. 2013, 53, 2896– 2907 DOI: 10.1021/ci400503s
  [ACS Full Text ], [CAS], Google Scholar
  33
  FAst MEtabolizer (FAME): A Rapid and Accurate Predictor of Sites of Metabolism in Multiple Species by Endogenous Enzymes
  Kirchmair, Johannes; Williamson, Mark J.; Afzal, Avid M.; Tyzack, Jonathan D.; Choy, Alison P. K.; Howlett, Andrew; Rydberg, Patrik; Glen, Robert C.
  Journal of Chemical Information and Modeling (2013), 53 (11), 2896-2907CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
  FAst MEtabolizer (FAME) is a fast and accurate predictor of sites of metab. (SoMs). It is based on a collection of random forest models trained on diverse chem. data sets of >20,000 mols. annotated with their exptl. detd. SoMs. Using a comprehensive set of available data, FAME aims to assess metabolic processes from a holistic point of view. It is not limited to a specific enzyme family or species. Besides a global model, dedicated models are available for human, rat, and dog metab.; specific prediction of phase I and II metab. is also supported. FAME is able to identify at least one known SoM among the top-1, top-2, and top-3 highest ranked atom positions in up to 71%, 81%, and 87% of all cases tested, resp. These prediction rates are comparable to or better than SoM predictors focused on specific enzyme families (such as cytochrome P450s), despite the fact that FAME uses only seven chem. descriptors. FAME covers a very broad chem. space, which together with its inter- and extrapolation power makes it applicable to a wide range of chems. Predictions take <2.5 s per mol. in batch mode on an Ultrabook. Results are visualized using Jmol, with the most likely SoMs highlighted.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3sXhslGmt7rO&md5=131ead027225bea9851fba02e7b63a09
34. 34
  Tyzack, J. D.; Hunt, P. A.; Segall, M. D. Predicting Regioselectivity and Lability of Cytochrome P450 Metabolism Using Quantum Mechanical Simulations J. Chem. Inf. Model. 2016, 56, 2180– 2193 DOI: 10.1021/acs.jcim.6b00233
  [ACS Full Text ], [CAS], Google Scholar
  34
  Predicting Regioselectivity and Lability of Cytochrome P450 Metabolism Using Quantum Mechanical Simulations
  Tyzack, Jonathan D.; Hunt, Peter A.; Segall, Matthew D.
  Journal of Chemical Information and Modeling (2016), 56 (11), 2180-2193CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
  Methods are described for predicting cytochrome P 450 (CYP) metab. incorporating both pathway-specific reactivity and isoform-specific accessibility considerations. Semi-empirical quantum mech. (QM) simulations, parameterized using exptl. data and ab initio calcns., estd. the reactivity of each potential site of metab. in the context of the whole mol. Ligand-based models, trained using high quality regioselectivity data, cor. for orientation and steric effects of the different CYP isoform binding pockets. The resulting models identified a site of metab. in the top 2 predictions for between 82 and 91% of compds. in independent test sets across 7 CYP isoforms. In addn. to predicting the relative proportion of metabolite formation at each site, these methods estd. the activation energy at each site, from which addnl. information could be derived regarding their lability in abs. terms. The authors illustrated how this could guide the design of compds. to overcome issues with rapid CYP metab.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC28Xhs1yksrfO&md5=3f4daa77aacbfaefaec652c8ff456740
35. 35
  He, S.-B.; Li, M.-M.; Zhang, B.-X.; Ye, X.-T.; Du, R.-F.; Wang, Y.; Qiao, Y.-J. Construction of Metabolism Prediction Models for CYP450 3A4, 2D6, and 2C9 Based on Microsomal Metabolic Reaction System Int. J. Mol. Sci. 2016, 17, E1686 DOI: 10.3390/ijms17101686
  [Crossref], [PubMed], [CAS], Google Scholar
  35
  Construction of metabolism prediction models for CYP450 3A4, 2D6, and 2C9 based on microsomal metabolic reaction system
  He, Shuai-Bing; Li, Man-Man; Zhang, Bai-Xia; Ye, Xiao-Tong; Du, Ran-Feng; Wang, Yun; Qiao, Yan-Jiang
  International Journal of Molecular Sciences (2016), 17 (10), 1686/1-1686/18CODEN: IJMCFK; ISSN:1422-0067. (MDPI AG)
  During the past decades, there have been continuous attempts in the prediction of metab. mediated by cytochrome P450s (CYP450s) 3A4, 2D6, and 2C9. However, it has indeed remained a huge challenge to accurately predict the metab. of xenobiotics mediated by these enzymes. To address this issue, microsomal metabolic reaction system (MMRS)-a novel concept, which integrates information about site of metab. (SOM) and enzyme-was introduced. By incorporating the use of multiple feature selection (FS) techniques (ChiSquared (CHI), InfoGain (IG), GainRatio (GR), Relief) and hybrid classification procedures (Kstar, Bayes (BN), K-nearest neighbors (IBK), C4.5 decision tree (J48), RandomForest (RF), Support vector machines (SVM), AdaBoostM1, Bagging), metab. prediction models were established based on metab. data released by Sheridan et al. Four major biotransformations, including aliph. C-hydroxylation, arom. C-hydroxylation, N-dealkylation and O-dealkylation, were involved. For validation, the overall accuracies of all four biotransformations exceeded 0.95. For receiver operating characteristic (ROC) anal., each of these models gave a significant area under curve (AUC) value >0.98. In addn., an external test was performed based on dataset published previously. As a result, 87.7% of the potential SOMs were correctly identified by our four models. In summary, four MMRS-based models were established, which can be used to predict the metab. mediated by CYP3A4, 2D6, and 2C9 with high accuracy.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC28XitVahsLjJ&md5=9b44af1c8b3412453be673ab9775bc9f
36. 36
  Finkelmann, A. R.; Göller, A. H.; Schneider, G. Site of Metabolism Prediction Based on Ab Initio Derived Atom Representations ChemMedChem 2017, 12, 606– 612 DOI: 10.1002/cmdc.201700097
  [Crossref], [PubMed], [CAS], Google Scholar
  36
  Site of Metabolism Prediction Based on ab initio Derived Atom Representations
  Finkelmann, Arndt R.; Goeller, Andreas H.; Schneider, Gisbert
  ChemMedChem (2017), 12 (8), 606-612CODEN: CHEMGX; ISSN:1860-7179. (Wiley-VCH Verlag GmbH & Co. KGaA)
  Machine learning models for site of metab. (SoM) prediction offer the ability to identify metabolic soft spots in low-mol.-wt. drug mols. at low computational cost and enable data-based reactivity prediction. SoM prediction is an atom classification problem. Successful construction of machine learning models requires atom representations that capture the reactivity-detg. features of a potential reaction site. We have developed a descriptor scheme that characterizes an atom's steric and electronic environment and its relative location in the mol. structure. The partial charge distributions were obtained from fast quantum mech. calcns. We successfully trained machine learning classifiers on curated cytochrome P 450 metab. data. The models based on the new atom descriptors showed sustained accuracy for retrospective analyses of metab. optimization campaigns and lead optimization projects from Bayer Pharmaceuticals. The results obtained demonstrate the practicality of quantum-chem.-supported machine learning models for hit-to-lead optimization.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC2sXlsV2ltL8%253D&md5=5dbee6380ec1bed9bae7887550e9ec6c
37. 37
  Geurts, P.; Ernst, D.; Wehenkel, L. Extremely Randomized Trees Mach. Learn. 2006, 63, 3– 42 DOI: 10.1007/s10994-006-6226-1
  [Crossref], Google Scholar
  There is no corresponding record for this reference.
38. 38
  Tyzack, J. D.; Williamson, M. J.; Torella, R.; Glen, R. C. Prediction of Cytochrome P450 Xenobiotic Metabolism: Tethered Docking and Reactivity Derived from Ligand Molecular Orbital Analysis J. Chem. Inf. Model. 2013, 53, 1294– 1305 DOI: 10.1021/ci400058s
  [ACS Full Text ], [CAS], Google Scholar
  38
  Prediction of Cytochrome P450 Xenobiotic Metabolism: Tethered Docking and Reactivity Derived from Ligand Molecular Orbital Analysis
  Tyzack, Jonathan D.; Williamson, Mark J.; Torella, Rubben; Glen, Robert C.
  Journal of Chemical Information and Modeling (2013), 53 (6), 1294-1305CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
  Metab. of xenobiotic and endogenous compds. is frequently complex, not completely elucidated, and therefore often ambiguous. The prediction of sites of metab. (SoM) can be particularly helpful as a first step toward the identification of metabolites, a process esp. relevant to drug discovery. This paper describes a reactivity approach for predicting SoM whereby reactivity is derived directly from the ground state ligand MO anal., calcd. using D. Functional Theory, using a novel implementation of the av. local ionization energy. Thus each potential SoM is sampled in the context of the whole ligand, in contrast to other popular approaches where activation energies are calcd. for a predefined database of mol. fragments and assigned to matching moieties in a query ligand. In addn., one of the first descriptions of mol. dynamics of cytochrome P 450 (CYP) isoforms 3A4, 2D6, and 2C9 in their Compd. I state is reported, and, from the representative protein structures obtained, an anal. and evaluation of various docking approaches using GOLD is performed. In particular, a covalent docking approach is described coupled with the modeling of important electrostatic interactions between CYP and ligand using spherical constraints. Combining the docking and reactivity results, obtained using std. functionality from common docking and quantum chem. applications, enables a SoM to be identified in the top 2 predictions for 75%, 80%, and 78% of the data sets for 3A4, 2D6, and 2C9, resp., results that are accessible and competitive with other recently published prediction tools.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3sXot1Srtb0%253D&md5=8215558c1c9459a4d64f5ee88e16a3e3
39. 39
  Huang, T.-W.; Zaretzki, J.; Bergeron, C.; Bennett, K. P.; Breneman, C. M. DR-Predictor: Incorporating Flexible Docking with Specialized Electronic Reactivity and Machine Learning Techniques to Predict CYP-Mediated Sites of Metabolism J. Chem. Inf. Model. 2013, 53, 3352– 3366 DOI: 10.1021/ci4004688
  [ACS Full Text ], [CAS], Google Scholar
  39
  DR-Predictor: Incorporating Flexible Docking with Specialized Electronic Reactivity and Machine Learning Techniques to Predict CYP-Mediated Sites of Metabolism
  Huang, Tao-wei; Zaretzki, Jed; Bergeron, Charles; Bennett, Kristin P.; Breneman, Curt M.
  Journal of Chemical Information and Modeling (2013), 53 (12), 3352-3366CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
  Computational methods that can identify CYP-mediated sites of metab. (SOMs) of drug-like compds. have become required tools for early stage lead optimization. In recent years, methods that combine CYP binding site features with CYP/ligand binding information have been sought to increase the prediction accuracy of such hybrid models over those that use only one representation. Two challenges that any hybrid ligand/structure-based method must overcome are (1) identification of the best binding pose for a specific ligand with a given CYP and (2) appropriately incorporating the results of docking with ligand reactivity. To address these challenges the authors have created Docking-Regioselectivity-Predictor (DR-Predictor) - a method that incorporates flexible docking-derived information with specialized electronic reactivity and multiple-instance-learning methods to predict CYP-mediated SOMs. The hybrid ligand-structure-based DR-Predictor method was tested on substrate sets for CYP 1A2 and CYP 2A6. For these data, the DR-Predictor model was found to identify the exptl. obsd. SOM within the top two predicted rank-positions for 86% of the 261 1A2 substrates and 83% of the 100 2A6 substrates. Given the accuracy and extendibility of the DR-Predictor method, the authors anticipate that it will further facilitate the prediction of CYP metab. liabilities and aid in in-silico ADMET assessment of novel structures.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3sXhslWlsrzF&md5=1d0592336da0de3bd1374b5d5b8fd991
40. 40
  Zaretzki, J. M.; Browning, M. R.; Hughes, T. B.; Swamidass, S. J. Extending P450 Site-of-Metabolism Models with Region-Resolution Data Bioinformatics 2015, 31, 1966– 1973 DOI: 10.1093/bioinformatics/btv100
  [Crossref], [PubMed], [CAS], Google Scholar
  40
  Extending P450 site-of-metabolism models with region-resolution data
  Zaretzki, Jed M.; Browning, Michael R.; Hughes, Tyler B.; Swamidass, S. Joshua
  Bioinformatics (2015), 31 (12), 1966-1973CODEN: BOINFP; ISSN:1367-4803. (Oxford University Press)
  Motivation: Cytochrome P450s are a family of enzymes responsible for the metab. of approx. 90% of FDA-approved drugs. Medicinal chemists often want to know which atoms of a mol.-its metabolized sites-are oxidized by Cytochrome P450s in order to modify their metab. Consequently, there are several methods that use literature-derived, atom-resoln. data to train models that can predict a mol.'s sites of metab. There is, however, much more data available at a lower resoln., where the exact site of metab. is not known, but the region of the mol. that is oxidized is known. Until now, no site-of-metab. models made use of region- resoln. data. Results: Here, we describe XenoSite-Region, the first reported method for training site-of-metab. models with region-resoln. data. Our approach uses the Expectation Maximization algorithm to train a site-of-metab. model. Region-resoln. metab. data was simulated from a large site-of-metab. dataset, contg. 2000 mols. with 3400 metabolized and 30 000 un-metabolized sites and covering nine Cytochrome P 450 isoenzymes. When training on the same mols. (but with only region-level information), we find that this approach yields models almost as accurate as models trained with atom-resoln. data. Moreover, we find that atom-resoln. trained models are more accurate when also trained with region-resoln. data from addnl. mols. Our approach, therefore, opens up a way to extend the applicable domain of site-of-metab. models into larger regions of chem. space. This meets a crit. need in drug development by tapping into underutilized data commonly available in most large drug companies.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC28Xht1Cit77E&md5=5463287e64149487a190db5ad62f6bee
41. 41
  Powers, D. Evaluation: From Precision, Recall and F-Measure to ROC, Informedness, Markedness & Correlation J. Mach. Learn. Technol. 2011, 2, 37– 63
  Google Scholar
  There is no corresponding record for this reference.
42. 42
  Adams, S. E. Molecular Similarity and Xenobiotic Metabolism; University of Cambridge, 2010.
  Google Scholar
  There is no corresponding record for this reference.
43. 43
  Boyer, S.; Arnby, C. H.; Carlsson, L.; Smith, J.; Stein, V.; Glen, R. C. Reaction Site Mapping of Xenobiotic Biotransformations J. Chem. Inf. Model. 2007, 47, 583– 590 DOI: 10.1021/ci600376q
  [ACS Full Text ], [CAS], Google Scholar
  43
  Reaction Site Mapping of Xenobiotic Biotransformations
  Boyer, Scott; Arnby, Catrin Hasselgren; Carlsson, Lars; Smith, James; Stein, Viktor; Glen, Robert C.
  Journal of Chemical Information and Modeling (2007), 47 (2), 583-590CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
  Predictive metab. methods can be used in drug discovery projects to enhance the understanding of structure-metab. relationships. The present study uses data mining methods to exploit biotransformation data that have been recorded in the MDL Metabolite database. Reacting center fingerprints were derived from a comparison of substrates and their corresponding products listed in the database. This process yields two fingerprint databases: all atoms in all substrates and all reacting centers. The metabolic reaction data are then mined by submitting a new mol. and searching for fingerprint matches to every atom in the new mol. in both databases. An "occurrence ratio" is derived from the fingerprint matches between the submitted compd. and the reacting center and substrate fingerprint databases. Normalization of the occurrence ratio within each submitted mol. enables the results of the search to be rank-ordered as a measure of the relative frequency of a reaction occurring at a specific site within the submitted mol. Predictive performance that would allow this method to be used by drug discovery teams to generate useful hypotheses regarding structure metab. relationships was obsd.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD2sXhslaitrc%253D&md5=5feeecf97e20393b3fc3a36be29eae6d
44. 44
  Tyzack, J. D.; Mussa, H. Y.; Williamson, M. J.; Kirchmair, J.; Glen, R. C. Cytochrome P450 Site of Metabolism Prediction from 2D Topological Fingerprints Using GPU Accelerated Probabilistic Classifiers J. Cheminf. 2014, 6, 29 DOI: 10.1186/1758-2946-6-29
  [Crossref], [CAS], Google Scholar
  44
  Cytochrome P450 site of metabolism prediction from 2D topological fingerprints using GPU accelerated probabilistic classifiers
  Tyzack, Jonathan D.; Mussa, Hamse Y.; Williamson, Mark J.; Kirchmair, Johannes; Glen, Robert C.
  Journal of Cheminformatics (2014), 6 (), 29/1-29/14, 14 pp.CODEN: JCOHB3; ISSN:1758-2946. (Chemistry Central Ltd.)
  Background: The prediction of sites and products of metab. in xenobiotic compds. is key to the development of new chem. entities, where screening potential metabolites for toxicity or unwanted side-effects is of crucial importance. In this work 2D topol. fingerprints are used to encode at. sites and three probabilistic machine learning methods are applied: Parzen-Rosenblatt Window (PRW), Naive Bayesian (NB) and a novel approach called RASCAL (Random Attribute Subsampling Classification Algorithm). These are implemented by randomly subsampling descriptor space to alleviate the problem often suffered by data mining methods of having to exactly match fingerprints, and in the case of PRW by measuring a distance between feature vectors rather than exact matching. The classifiers have been implemented in CUDA/C++ to exploit the parallel architecture of graphical processing units (GPUs) and is freely available in a public repository. Results: It is shown that for PRW a SoM (Site of Metab.) is identified in the top two predictions for 85%, 91% and 88% of the CYP 3A4, 2D6 and 2C9 data sets resp., with RASCAL giving similar performance of 83%, 91% and 88%, resp. These results put PRW and RASCAL performance ahead of NB which gave a much lower classification performance of 51%, 73% and 74%, resp. Conclusions: 2D topol. fingerprints calcd. to a bond depth of 4-6 contain sufficient information to allow the identification of SoMs using classifiers based on relatively small data sets. Thus, the machine learning methods outlined in this paper are conceptually simpler and more efficient than other methods tested and the use of simple topol. descriptors derived from 2D structure give results competitive with other approaches using more expensive quantum chem. descriptors. The descriptor space subsampling approach and ensemble methodol. allow the methods to be applied to mols. more distant from the training data where data mining would be more likely to fail due to the lack of common fingerprints. The RASCAL algorithm is shown to give equiv. classification performance to PRW but at lower computational expense allowing it to be applied more efficiently in the ensemble scheme.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC2MXivFSks74%253D&md5=cb64fc4f7a201503285ead1d4d732c6b
45. 45
  Stewart, J. J. MOPAC: A Semiempirical Molecular Orbital Program J. Comput.-Aided Mol. Des. 1990, 4, 1– 105 DOI: 10.1007/BF00128336
  [Crossref], [PubMed], [CAS], Google Scholar
  45
  MOPAC: a semiempirical molecular orbital program
  Stewart, James J. P.
  Journal of Computer-Aided Molecular Design (1990), 4 (1), 1-105CODEN: JCADEQ; ISSN:0920-654X.
  An overview is presented of MOPAC, a semiempirical MO program for the study of chem. reactions involving mols., ions, and linear polymers. The program implements the semiempirical Hamiltonians MNDO, AM1, MINDO/3, and MNDO-PM3 and combines the calcns. of vibrational spectra, thermodn. quantities, isotopic substitution effects, and force consts. in a fully integrated program.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADyaK3cXksFyjurc%253D&md5=983366b4ba2797f0fd8a91aec6d366ed
46. 46
  MOPAC2016. http://openmopac.net/home.html (accessed Apr 7, 2017) .
  Google Scholar
  There is no corresponding record for this reference.
47. 47
  Schüürmann, G. Quantitative Structure-Property Relationships for the Polarizability, Solvatochromic Parameters and Lipophilicity Quant. Struct.-Act. Relat. 1990, 9, 326– 333 DOI: 10.1002/qsar.19900090406
  [Crossref], [CAS], Google Scholar
  47
  Quantitative structure-property relationships for the polarizability, solvatochromic parameters and lipophilicity
  Schueuermann, Gerrit
  Quantitative Structure-Activity Relationships (1990), 9 (4), 326-33CODEN: QSARDI; ISSN:0931-8771.
  The polarizability αvol, the solvatochromic parameters π* and β and the lipophilicity as expressed by log Kow are subjected to regression analyses using calcd. mol. parameters within the MNDO scheme. The resulting linear one- and more-variable regression equations enable rapid approx. calcns. for αvol, π* and β and log Kow of untested compds.; in particular, αvol is split into a mol. size and an electronic part which offers new possibilities for applications in quant. structure-activity relationships.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADyaK3MXhvFaqtrY%253D&md5=e7e356b88e9179b4698e7515c47e1f0e
48. 48
  Coulson, C. A.; Longuet-Higgins, H. C. The Electronic Structure of Conjugated Systems. II. Unsaturated Hydrocarbons and Their Hetero-Derivatives Proc. R. Soc. London, Ser. A 1947, 192, 16– 32 DOI: 10.1098/rspa.1947.0136
  [Crossref], Google Scholar
  There is no corresponding record for this reference.
49. 49
  Fukui, K.; Kato, H.; Yonezawa, T. A New Quantum-Mechanical Reactivity Index for Saturated Compounds Bull. Chem. Soc. Jpn. 1961, 34, 1111– 1115 DOI: 10.1246/bcsj.34.1111
  [Crossref], Google Scholar
  There is no corresponding record for this reference.
50. 50
  Gopinathan, M. S.; Siddarth, P.; Ravimohan, C. Valency and Molecular Structure Theor. Chim. Acta 1986, 70, 303– 322 DOI: 10.1007/BF00534237
  [Crossref], Google Scholar
  There is no corresponding record for this reference.
51. 51
  Mulliken, R. S. Electronic Population Analysis on LCAO-MO Molecular Wave Functions. I J. Chem. Phys. 1955, 23, 1833– 1840 DOI: 10.1063/1.1740588
  [Crossref], [CAS], Google Scholar
  51
  Electronic population analysis on LCAO-MO [linear combination of atomic orbital-molecular orbital] molecular wave functions. I
  Mulliken, R. S.
  Journal of Chemical Physics (1955), 23 (), 1833-40CODEN: JCPSA6; ISSN:0021-9606.
  An analysis in quant. form was given in terms of breakdowns of the electronic population into partial and total "gross at. populations," or into partial and total "net at. populations" together with "overlap populations." Gross at. populations distribute the electrons almost perfectly among the various at. orbitals of the various atoms in the mol. From these nos., a definite figure is obtained for the amt. of promotion (e.g. from 2s to 2p) in each atom; and also for the gross charge Q on each atom if the bonds are polar. The total overlap population for any pair of atoms in a mol. is in general made up of pos. and neg. contributions. If the total overlap population between 2 atoms is pos., they are bonded; if neg., they are antibonded. Tables of gross at. populations and overlap populations were calcd. for CO and H2O. The amt. of s-p promotion was nearly the same for the O atom in CO and in H2O (0.14 electron in CO and 0.15e in H2O). For the C atom in CO it is 0.50e. For the N atom in N2 it is 0.26e. In spite of very strong polarity in the π bonds in CO, the σ and π overlap populations are very similar to those in N. In CO the total overlap population for the π electrons is about twice that for the σ electrons. The most easily ionized electrons of CO are in a mol. orbital such that its gross at. population is 94% localized on the C atom; these electrons account for the weak electron donor properties of CO.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADyaG28Xnt1Kq&md5=5aab51c06ce6a3250219cf12d1b5f395
52. 52
  Mulliken, R. S. Criteria for the Construction of Good Self-Consistent-Field Molecular Orbital Wave Functions, and the Significance of LCAO-MO Population Analysis J. Chem. Phys. 1962, 36, 3428– 3439 DOI: 10.1063/1.1732476
  [Crossref], [CAS], Google Scholar
  52
  Criteria for the construction of good self-consistent-field molecular orbital wave functions, and the significance of L.C.A.O.M.O. population analysis
  Mulliken, R. S.
  Journal of Chemical Physics (1962), 36 (), 3428-39CODEN: JCPSA6; ISSN:0021-9606.
  Criteria for the optimal choice of finite linear combinations of Slater-type orbitals (S.T.O.) adequate to approx. closely S.C.F. M.O. were examd. in the light of computations on HF and other mols. Some aspects of the A.O. (generalized Heitler-London) method were discussed. The inherent limitations on the meaning of charges on atoms in a mol., or of degree of ionic character, were discussed. Unacceptable at. charges are found from an L.C.A.O. mol. orbital population analysis made on S.C.F. M.O. wave functions, approximated by using unbalanced S.T.O. sets, while acceptable results are obtained with judiciously balanced and sufficiently complete S.T.O. sets. The effects of insufficiently complete and unbalanced S.T.O. sets on the computed dipole moments of S.C.F. M.O. wave functions were discussed, with examples.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADyaF38Xks1KrtL8%253D&md5=9fb6c9a93a70358d02154b028efe63fc
53. 53
  Holm, S. A Simple Sequentially Rejective Multiple Test Procedure Scand. Stat. Theory Appl. 1979, 6, 65– 70
  Google Scholar
  There is no corresponding record for this reference.
54. 54
  Friedman, M. A Comparison of Alternative Tests of Significance for the Problem of m Rankings Ann. Math. Stat. 1940, 11, 86– 92 DOI: 10.1214/aoms/1177731944
  [Crossref], Google Scholar
  There is no corresponding record for this reference.
55. 55
  Friedman, M. A Correction: The Use of Ranks to Avoid the Assumption of Normality Implicit in the Analysis of Variance J. Am. Stat. Assoc. 1939, 34, 109– 109 DOI: 10.2307/2279169
  [Crossref], Google Scholar
  There is no corresponding record for this reference.
56. 56
  Friedman, M. The Use of Ranks to Avoid the Assumption of Normality Implicit in the Analysis of Variance J. Am. Stat. Assoc. 1937, 32, 675– 701 DOI: 10.1080/01621459.1937.10503522
  [Crossref], Google Scholar
  There is no corresponding record for this reference.
57. 57
  Shapiro, S. S.; Wilk, M. B. An Analysis of Variance Test for Normality (Complete Samples) Biometrika 1965, 52, 591– 611 DOI: 10.1093/biomet/52.3-4.591
  [Crossref], Google Scholar
  There is no corresponding record for this reference.
58. 58
  Mauchly, J. W. Significance Test for Sphericity of a Normal N-Variate Distribution Ann. Math. Stat. 1940, 11, 204– 209 DOI: 10.1214/aoms/1177731915
  [Crossref], Google Scholar
  There is no corresponding record for this reference.
59. 59
  Greenhouse, S. W.; Geisser, S. On Methods in the Analysis of Profile Data Psychometrika 1959, 24, 95– 112 DOI: 10.1007/BF02289823
  [Crossref], Google Scholar
  There is no corresponding record for this reference.
60. 60
  Huynh, H.; Feldt, L. S. Estimation of the Box Correction for Degrees of Freedom from Sample Data in Randomized Block and Split-Plot Designs J. Educ. Behav. Stat. 1976, 1, 69– 82 DOI: 10.3102/10769986001001069
  [Crossref], Google Scholar
  There is no corresponding record for this reference.
61. 61
  de Bruyn Kops, C.; Friedrich, N.-O.; Kirchmair, J. Alignment-Based Prediction of Sites of Metabolism J. Chem. Inf. Model. 2017, 57 (6) 1258– 1264
  [ACS Full Text ], [CAS], Google Scholar
  61
  Alignment-Based Prediction of Sites of Metabolism
  de Bruyn Kops, Christina; Friedrich, Nils-Ole; Kirchmair, Johannes
  Journal of Chemical Information and Modeling (2017), 57 (6), 1258-1264CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
  Prediction of metabolically labile atom positions in a mol. (sites of metab.) is a key component of the simulation of xenobiotic metab. as a whole, providing crucial information for the development of safe and effective drugs. In 2008, an exploratory study was published in which sites of metab. were derived based on mol. shape- and chem. feature-based alignment to a mol. whose site of metab. (SoM) had been detd. by expts. The authors present a detailed anal. of the breadth of applicability of alignment-based SoM prediction, including transfer of the approach from a structure- to ligand-based method and extension of the applicability of the models from cytochrome P 450 2C9 to all cytochrome P 450 isoenzymes involved in drug metab. The authors evaluate the effect of mol. similarity of the query and ref. mols. on the ability of this approach to accurately predict SoMs. In addn., the authors combine the alignment-based method with a leading chem. reactivity model to take reactivity into account. The combined model yielded superior performance in comparison to the alignment-based approach and the reactivity models with an av. area under the receiver operating characteristic curve of 0.85 in cross-validation expts. In particular, early enrichment was improved, as evidenced by higher BEDROC scores (mean BEDROC = 0.59 for α = 20.0, mean BEDROC = 0.73 for α = 80.5).
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC2sXnvV2hs7w%253D&md5=3a48e4bbfc64e569c9a9ff3d4040bc38
62. 62
  OMEGA, version 2.5.1.4; OpenEye Scientific Software: Santa Fe, NM, 2011; https://www.eyesopen.com (accessed Apr 7, 2017).
  Google Scholar
  There is no corresponding record for this reference.
63. 63
  Hawkins, P. C. D.; Skillman, A. G.; Warren, G. L.; Ellingson, B. A.; Stahl, M. T. Conformer Generation with OMEGA: Algorithm and Validation Using High Quality Structures from the Protein Databank and Cambridge Structural Database J. Chem. Inf. Model. 2010, 50, 572– 584 DOI: 10.1021/ci100031x
  [ACS Full Text ], [CAS], Google Scholar
  63
  Conformer Generation with OMEGA: Algorithm and Validation Using High Quality Structures from the Protein Databank and Cambridge Structural Database
  Hawkins, Paul C. D.; Skillman, A. Geoffrey; Warren, Gregory L.; Ellingson, Benjamin A.; Stahl, Matthew T.
  Journal of Chemical Information and Modeling (2010), 50 (4), 572-584CODEN: JCISD8; ISSN:1549-9596. (American Chemical Society)
  Here, we present the algorithm and validation for OMEGA, a systematic, knowledge-based conformer generator. The algorithm consists of three phases: assembly of an initial 3D structure from a library of fragments; exhaustive enumeration of all rotatable torsions using values drawn from a knowledge-based list of angles, thereby generating a large set of conformations; and sampling of this set by geometric and energy criteria. Validation of conformer generators like OMEGA has often been undertaken by comparing computed conformer sets to exptl. mol. conformations from crystallog., usually from the Protein Databank (PDB). Such an approach is fraught with difficulty due to the systematic problems with small mol. structures in the PDB. Methods are presented to identify a diverse set of small mol. structures from cocomplexes in the PDB that has maximal reliability. A challenging set of 197 high quality, carefully selected ligand structures from well-solved models was obtained using these methods. This set will provide a sound basis for comparison and validation of conformer generators in the future. Validation results from this set are compared to the results using structures of a set of druglike mols. extd. from the Cambridge Structural Database (CSD). OMEGA is found to perform very well in reproducing the crystallog. conformations from both these data sets using two complementary metrics of success.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BC3cXjtlaisrY%253D&md5=fb87ecc9c51eddef63b41fffcd9babee
64. 64
  RDKit 2016.03.4. https://github.com/rdkit/rdkit/releases/tag/Release_2016_03_4 (Accessed April 7, 2017).
  Google Scholar
  There is no corresponding record for this reference.
65. 65
  Steinbeck, C.; Han, Y.; Kuhn, S.; Horlacher, O.; Luttmann, E.; Willighagen, E. The Chemistry Development Kit (CDK): An Open-Source Java Library for Chemo- and Bioinformatics J. Chem. Inf. Comput. Sci. 2003, 43, 493– 500 DOI: 10.1021/ci025584y
  [ACS Full Text ], [CAS], Google Scholar
  65
  The Chemistry Development Kit (CDK): An Open-Source Java Library for Chemo- and Bioinformatics
  Steinbeck, Christoph; Han, Yongquan; Kuhn, Stefan; Horlacher, Oliver; Luttmann, Edgar; Willighagen, Egon
  Journal of Chemical Information and Computer Sciences (2003), 43 (2), 493-500CODEN: JCISD8; ISSN:0095-2338. (American Chemical Society)
  The Chem. Development Kit (CDK) is a freely available open-source Java library for Structural Chemo- and Bioinformatics. Its architecture and capabilities as well as the development as an open-source project by a team of international collaborators from academic and industrial institutions is described. The CDK provides methods for many common tasks in mol. informatics, including 2D and 3D rendering of chem. structures, I/O routines, SMILES parsing and generation, ring searches, isomorphism checking, structure diagram generation, etc. Application scenarios as well as access information for interested users and potential contributors are given.
  >> More from SciFinder ^®
  https://chemport.cas.org/services/resolver?origin=ACS&resolution=options&coi=1%3ACAS%3A528%3ADC%252BD3sXhtVaktbg%253D&md5=afc8fd10783af301c73a8183727230bf
66. 66
  Chemistry Development Kit 1.4.19. https://github.com/cdk/cdk/releases/tag/cdk-1.4.19 (accessed Apr 7, 2017) .
  Google Scholar
  There is no corresponding record for this reference.
67. 67
  scikit-learn 0.18. http://scikit-learn.org/0.18/documentation.html (accessed Apr 7, 2017) .
  Google Scholar
  There is no corresponding record for this reference.
Supporting Information
Supporting Information
ARTICLE SECTIONS
Jump To
The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jcim.7b00250.
- Additional figures and tables with calculated CDK descriptors, hyperparameter optimization results, model validation results, and performance of the random forest algorithm (PDF)
- ci7b00250_si_001.pdf (684.23 kb)
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.

PDF [ 4MB]

All Types

FAME 2: Simple and Effective Machine Learning Model of Cytochrome P450 Regioselectivity

Article Views

Altmetric

Citations

Abstract

Introduction

Results

Figure 1

Model Evaluation Measures

Model Performance as a Function of Descriptor Types and Model Parameters

Figure 2

Optimized Model Parameters

Decision Threshold and Class Weights

Maximum Number of Considered Features per Split

ANOVA F-Test

Influence of Descriptors on Model Performance During Training

Figure 3

Model Performance on an Independent Test Set

Figure 4

Figure 5

Model Performance in a Structurally Uncorrelated Test Set

Figure 6

Comparison with Other Methods

Performance Comparison with the Finkelmann Models

Performance Comparison with Xenosite

Isoform-Specific Models

Computation Time

Software Package Description

Figure 7

Conclusions

Methods

Data Set

Descriptors

Descriptor Selection

Value Imputation

Model Building

Supporting Information

Terms & Conditions

Author Information

Acknowledgment

List of Abbreviations

References

Cited By

Abstract

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

References

Supporting Information

Supporting Information

Terms & Conditions

STEP 1:

STEP 2: