Designing Potent HIV‑1 Protease Inhibitors Using Machine Learning and QSAR Approaches

Acquired Immune Deficiency Syndrome (AIDS), caused by Human Immunodeficiency Virus type-1 (HIV-1), remains a global health crisis. Despite advances in antiretroviral therapy, drug resistance, particularly to protease inhibitors, persists as a significant challenge. Darunavir, a second-generation protease inhibitor, has reduced efficacy against resistant HIV-1 variants, underscoring the development of new inhibitors. This study combines machine learning (ML) and quantitative structure-activity relationship (QSAR) models to design potent HIV-1 protease inhibitors using phenol-based and polyphenol-based P2 ligands. QSAR models, including genetic function approximation (GFA), multiple linear regression (MLR), random forest (RF), gradient boosting regressor (GBR), and Extreme Gradient Boosting (XGBoost), were developed to analyze molecular descriptors. GBR exhibited the highest accuracy (R ² = 0.911 and 0.994) with GFA and PI-selected descriptors, respectively. SHAP analysis highlighted key contributions to pIC₅₀ predictions, including electronic charge at C53, low dipole moments, and shortened bond length (C53-O54). Five potent inhibitors (B01-B05) were predicted, outperforming HIV-1 protease inhibitors. Furthermore, molecular docking suggested that B03 and B05 exhibit strong binding interactions with wild-type and variants, particularly through hydrophobic and hydrogen bonding interactions, with key residues including D25, G27, D29, D30, D25', and D30'. This integrated QSAR-ML and structure-based analysis offers promising candidates for addressing drug resistance.