Development of a ligand-based virtual screening method for preliminary compound selection (part I) : performance evaluation for preliminary compound selection methods

Abstract

Molecular docking is one of the most popular and most promising methods for virtual screening. However the heavy CPU cost prevents from applying it to the large compound library. Therefore, preliminary compound selection typically must be done before using molecular docking. We have planned to develop a new preliminary compound selection method and our unsuccessful feasible study suggested that DUD, typically used benchmark data set for virtual screening, might be unsuitable for our purpose. In this study, we have comprehensively calculated descriptors of DUD ligands and decoys using CDK, the open source cheminformatics tools, and analyzed their distributions. For some targets, the distributions of descriptors of decoys are smaller than those of ligands and sometimes included by ligands. It suggests that there are few decoys to reject by our selection method for the targets and using the whole DUD data set is not suitable for evaluating our compound selection procedures. We propose alternative benchmark data sets for the preliminary compound selection methods.