Extension of an Atom–Atom Dispersion Function to Halogen Bonds and Its Use for Rational Design of Drugs and Biocatalysts

A dispersion function Das in the form of a damped atom–atom asymptotic expansion fitted to ab initio dispersion energies from symmetry-adapted perturbation theory was improved and extended to systems containing heavier halogen atoms. To illustrate its performance, the revised Das function was implemented in the multipole first-order electrostatic and second-order dispersion (MED) scoring model. The extension has allowed applications to a much larger set of biocomplexes than it was possible with the original Das. A reasonable correlation between MED and experimentally determined inhibitory activities was achieved in a number of test cases, including structures featuring nonphysically shortened intermonomer distances, which constitute a particular challenge for binding strength predictions. Since the MED model is also computationally efficient, it can be used for reliable and rapid assessment of the ligand affinity or multidimensional scanning of amino acid side-chain conformations in the process of rational design of novel drugs or biocatalysts.


S-2
The training set used for the determination of D 20 as parameters, shown in Table S1, consisted of 164 dimers (see Table S2). Geometries of the training set dimers with bromine and iodine atoms were taken from the X40 database, S6 and included four geometries with the center-ofmass (COM) separation (R) smaller than at the minimum, and five geometries with the COM separation larger than at the minimum (together yielding ten different configurations for each dimer). Geometries of the remaining dimers were optimized at the second order Moeller-Plesset MP2/aug-cc-pVTZ S5 level of theory (several minimum geometries were obtained from the S22 or NCCE31/05 datasets, as indicated in Table S2). For these dimers, ten different radial geometries corresponding to the same angular configuration were provided.
In particular, the minimum geometry was accompanied by two geometries with the COM separation smaller than at the minimum, and seven geometries with the COM separation larger than at the minimum (up to 10Å). Overall, there were 1640 configurations in the training set, and the MUE and MURE errors associated with the obtained D 20 as values were equal to 0.1 kcal · mol −1 and 5.1%, respectively.
The D 20 as parameters are given in Table S1. The C 6 x and C 8 x coefficients are given in the units of J · nm 6 · mol −1 and J · nm 8 · mol −1 , respectively. The values of β x are given in bohr −1 .
To obtain the values of C 6 x and C 8 x in atomic units, the corresponding values need to be multiplied by 17.34525495 and 6194.102092, respectively.

S-15
EL,MTP and D 20 as contributions a , computed for HB alcohol dimers.
-5.4 -5.5 -7.1 -4.8 -11.8 EtOH -5.6 -5.7 -6.8 -5.3 -12.0 nPrOH -5.9 -5.9 -6.3 -5.4 -11.7 nBuOH -6.0 -5.9 -5.8 -5.5 -11.4 iPrOH -6.5 -6.5 -5.9 -6.9 -12.7 tBuOH -7.2 -7.3 -7.7 -7.6 -15.4 a In units of kcal · mol −1 . b Naming of dimers is consistent with Ref. S24, from which the reference SAPT energy values were taken.  b Numbering of the inhibitors is consistent with Ref. S23. c Correlation coefficient between the energy obtained at a given level of theory and the experimental inhibitory activity (expressed as pIC 50 values) taken from Ref. S23. In the case of GoldScore, ChemScore, ChemPLP and ASP functions, for which higher score indicates greater inhibitory activity, the opposite of the correlation coefficient value is given to facilitate direct comparison with the results of the remaining empirical scoring functions (or the MED model), wherein the more potent inhibitor is associated with a lower score (or the binding energy) value.