An Interpretable Deep Learning and Molecular Docking Framework for Repurposing Existing Drugs as Inhibitors of SARS-CoV-2 Main Protease

Despite the widespread use of vaccines against SARS-CoV-2, COVID-19 continues to pose global health challenges, requiring efficient drug screening and repurposing strategies. This study presents a novel hybrid framework that integrates deep learning (DL) with molecular docking to accelerate the identification of potential therapeutics. The framework comprises three crucial steps: (1) a previously developed DL model is employed to rapidly screen candidate compounds, selecting those with predicted interaction scores above a cut-off value of 0.8; (2) AutoDock Vina version 1.5.6 and LeDock version 1.0 are used to evaluate binding affinities, with a threshold of <-7.0 kcal·mol^-1; and (3) predicted drug-protein binding sites are evaluated to determine their overlap with known active residues of the target protein. We first validated the framework using four experimentally confirmed COVID-19 drug-target pairs and then applied it to identify potential inhibitors of the SARS-CoV-2 main protease (M^Pro). Among 29 drug candidates selected based on antiviral, anti-inflammatory, or anti-cancer properties, only Enasidenib met all three selection criteria, showing promise as an M^Pro inhibitor. However, further experimental and clinical studies are required to confirm its efficacy against SARS-CoV-2. This work provides an interpretable strategy for virtual screening and drug repurposing, which can be readily adapted to other DL models and docking tools.