Comparison of Docking Scoring Functions by InterCriteria Analysis on a Set of Protein Targets Related to Alzheimer and Parkinson Diseases

Abstract

Nowadays, the pharmaceutical industry extensively uses in silico drug design methods to deal with the enormously wide chemical space of druggable compounds and to reduce the R&D expenses. Having a significant number of tools for structure-based in silico screening, the aim of this study was to assess the putative consonance between the docking scores obtained by different scoring functions, which could provide a basis for reduction of the computational cost and/or for optimal choice of scoring functions. The proteins used in the study were acetylcholine esterase (AChE), histone deacetylase 2 (HDAC2), and monoamine oxidase B (MAO-B), enzymes related to the treatment of symptoms of Alzheimer and Parkinson diseases and potential subjects for inclusion in “single drug – multiple targets” research. The 11085 small molecules (ligands) used in the study were selected from a database of more than 600000 commercially available drug-like compounds which docking scores obtained by a rigid docking were better than the re-docking scores of the co-crystalized reference ligands in the selected enzymes. To assess the differences in performance of the scoring functions implemented in different molecular docking software packages, docking of these ligands in the target proteins was performed using several widely used molecular modelling platforms: (i) rigid- and flexible-protein docking in MOE (v. 2019.01, https://www.chemcomp.com ); (ii) rigid-protein docking in FlexX (v. 4.3) and ligand optimization and rescoring in HYDEscorer (v. 1.0), respectively ( www.biosolveit.de ); and (iii) rigid-protein docking in AutoDock Vina ( http://vina.scripps.edu ). Besides docking scores, the time for docking was also recorded and the computational costs were calculated for each of the studied docking protocol/scoring function pairs. The binding energies estimated by the selected scoring functions were subjected to intercriteria analysis (ICrA). The ICrA approach relies on the formalisms of the intuitionistic fuzzy sets and index matrices, and attempts to uncover similarities in the behavior of criteria applied for evaluation of multiple objects. ICrA was employed as a potential tool to support the selection of an appropriate scoring function for ranking of more than 11 000 ligands identified to interact with all three proteins simultaneously. Further, an analysis of the intersections of the top 1000 ligands for each target ranked by different scoring functions, was performed. ICrA analysis revealed consonances between FlexX and AutoDock Vina scoring functions in all studied proteins, and additionally between MOE flexible-protein docking and AutoDock Vina in AChE. Analysis of intersection results was to a great extent in line with the intercriteria relations. The results indicate that a precise selection of scoring functions and docking protocols, confirmed by the available knowledge of the studied objects, is needed. This analysis suggests also the possibility for optimization of in silico screening campaigns by avoidance of computationally expensive docking protocols that are highly consonant with the less expensive ones.