Machine Learning-Driven discovery of mushroom-derived inhibitors targeting InhA of Mycobacterium tuberculosis: An integrated QSAR, molecular docking and molecular dynamic simulation approach

Mycobacterium tuberculosis causes tuberculosis (TB), which remains a significant health problem worldwide. The rise of multidrug-resistant bacteria has worsened the situation, and current treatments are becoming less effective. InhA, a key enzyme involved in mycolic acid biosynthesis, is a validated therapeutic target in anti-TB therapy. This study aimed to explore the chemical diversity of natural substances from mushrooms against TB. Experimentally validated inhibitors from ChEMBL were retrieved to generate machine learning-based QSAR models combining nine chemical fingerprints and rigorous feature selection. The optimal RF-SVM-RFE model displayed high prediction performance (accuracy = 0.953, ROC_AUC = 0.971) and led virtual screening of mushroom metabolites. Six top-ranked compounds, including Inoscavin A and Schizine A, displayed substantial binding affinities (-11.7 to -10.5 kcal/mol) and stable interaction networks in molecular docking and MD simulations. Explainable AI (SHAP and LIME) showed fundamental structural motifs that drive activity and enhance chemical interpretability. These findings suggest promising natural scaffolds for anti-TB drug development and underscore the importance of AI-driven strategies in accelerating natural product-based therapeutics.