Target Prediction and Ligand Selectivity Using Inverse Virtual Screening for Structure-Based Drug Design

Inverse virtual screening (IVS) offers an avenue for discovering novel therapeutic indications for existing drugs and identifying new targets for drug development. For docking-based inverse virtual screening, small molecules undergo screening against a database of protein targets to identify true target structures, a process that is reversed from the traditional virtual screening. Despite numerous methods for protein-ligand docking, accurately ranking positive targets within a large database continues to be a challenge. In this work, we present an inverse virtual screening of a subset of nonredundant drugs from DrugBank and the sc-PDB database for systematic performance evaluation using AutoDock Vina and MDock as docking engines. We employed AUC-ROC on our updated benchmark, providing a robust basis for measuring the recognition of true positive targets. This study can serve as a practical guide for researchers in identifying the potential targets of compounds found through high-throughput screening and in predicting ligand selectivity.