A Robust Ensemble Machine Learning Approach for Inhibitor Discovery: Case Study of HIV‐1 NNRTI and Validation Using MD Simulation

The growing demand for new therapeutics highlights the need for intelligent, cost-effective, and scalable drug discovery strategies. Here, we present an artificial intelligence (AI)-based ensemble framework to accelerate the identification of small-molecule inhibitors against therapeutic targets. As a case study, we applied this approach to HIV-1 reverse transcriptase (HIV-1 RT), an essential enzyme in viral replication. Our stacking ensemble model, trained on a curated ChEMBL dataset, achieved high predictive performance (90.3% accuracy, 89.4% ROC-AUC) and was used to screen the Natural Products Atlas (NPA) database. Promising hits were evaluated through physicochemical and ADMET filters, molecular docking, and 1 µs molecular dynamics (MD) simulations. Compound NP1, which exhibited stable binding to the NNRTI binding pocket, outperformed the FDA-approved drug doravirine in post-MD characterizations. Network analysis further suggested potential allosteric regulation via residues N136 and E138. This flexible AI-MD pipeline provides an efficient strategy for discovering and repurposing inhibitors, with broad applicability to other therapeutic targets.