Drug Score of target proteins.

The global pandemic, due to the emergence of COVID-19, has created a public health crisis. It has a huge morbidity rate that was never comprehended in the recent decades. Despite numerous efforts, potent antiviral drugs are lacking. Repurposing of drugs presents a low-cost and rapid solution for finding new drugs by exploiting known drugs. In this study, we employed an integrated In-Silico approach using molecular docking and machine learning regression models to explore the potential inhibitors against key proteins of SARS-CoV-2. A library of 5903 drugs from the ZINC database was retrieved and screened against three crucial viral targets: Spike glycoprotein (7LM9), main protease 3CLpro (7JSU), and Nucleocapsid protein (7DE1). Binding affinities were predicted by using molecular docking, and subsequent predictive regression models, Decision Tree Regression (DTR), Gradient Boosting, XGBoost, Extra Trees, KNNR, and MLP, were constructed employing MACCS molecular fingerprints. Among them, the DTR model had better predictive performance, as indicated by the highest R² and lowest RMSE. The highest ranked compounds possessed good binding affinities (−12.6 to −19.7 kcal/mol) and favorable pharmacokinetics. Importantly, five novel candidate compounds, namely ZINC003873365, ZINC085432544, ZINC008214470, ZINC085536956, and ZINC261494640, had multi-target potential and optimal binding interaction. This computational analysis yields useful information for lead prioritization and sets the stage for additional in vitro and in vivo confirmation of these drug candidates to combat COVID-19.