Integration of Ligand-Based and Structure-Based Approaches for Virtual Screening of Factor Xa Inhibitors

Abstract

Factor Xa (FXa) is recognized as an attractive target for the design and development of new anticoagulant agents for combating the thromboembolic diseases. Recently, in silico prediction and screening approaches have been adopted as effective paradigms to complement high-throughput screening (HTS) for the identification of novel lead compounds of a specific biological target. In this work, we integrated the ligand-based virtual screening (LBVS) and structure-based virtual screening (SBVS) approaches for prediction and screening of the potentially potent FXa inhibitors from large chemical libraries. The state-of-the-art machine learning methods, including support vector machine and random forest, were firstly employed as the LBVS approach to develop the prediction models to rapidly narrow large chemical libraries to just hundreds of enriched hits or more. The VS performance of the developed models was evaluated by using the annotated MDL Drug Data Report (MDDR) database, achieving substantial yields and comparable hit rates and enrichment factors. The better performing random forest model was subsequently used to perform VS against the “fragment-like” subset of ZINC database to enrich the potential actives. These potentially enriched actives were further docked to the target protein human FXa using AutoDock4 of the SBVS approach, so as to examine their binding affinities, thereby obtaining 27 most potent candidates.