MERS-Mpro Predictor: A Machine Learning-Based Tool for Rapid Screening of Potential MERS-CoV Main Protease Inhibitors

The Middle East Respiratory Syndrome coronavirus (MERS-CoV) remains a significant global health concern due to the absence of approved antiviral therapeutics. In this study, we developed a ligand-based machine learning framework to identify potential inhibitors of the MERS-CoV main protease (Mpro) using molecular representations derived from SMILES strings. Multiple classification algorithms, including logistic regression, support vector machines, random forests, and Extreme Gradient Boosting (XGBoost), were systematically evaluated. Model performance was assessed through both internal validation and an external dataset. While several models exhibited strong performance during validation, the Random Forest classifier demonstrated the most robust and consistent generalization, achieving superior predictive performance on the external dataset. To ensure model reliability, a comprehensive validation strategy was implemented, including strict data partitioning to prevent structural overlap, Y-scrambling analysis to eliminate chance correlations, and applicability domain assessment to define the model's reliable prediction space. The final model was deployed as an interactive web-based application, enabling rapid virtual screening of compounds through single or batch SMILES input, and providing activity predictions along with probability scores and selected physicochemical descriptors. Overall, this study presents a reproducible ligand-based approach for supporting the early-stage identification of potential MERS-CoV Mpro inhibitors.