Support vector machine classification-guided identification of novel monoamine oxidase-B inhibitors via...

Monoamine oxidase-B (MAO-B) is a mitochondrial enzyme that catalyzes the oxidative deamination of dopamine and other neurotransmitters, contributing to neurodegeneration via increased oxidative stress. Elevated MAO-B activity is implicated in neurodegenerative disorders such as Parkinson's and Alzheimer's diseases. Although several MAO-B inhibitors are clinically available, their therapeutic utility is often limited by poor selectivity, off-target effects, and suboptimal pharmacokinetics. Here, we developed an integrated computational approach to identify novel, selective MAO-B inhibitors by repurposing FDA-approved drugs. Support vector machine (SVM) classification models were trained on curated MAO-B bioactivity data using four molecular fingerprints, with robust feature selection and cross-validation. The models achieved high predictive performance (AUC up to 0.9914, accuracy >96.93 %, MCC > 0.9179) and were used to screen FDA-approved drugs. For benchmarking, Random Forest, XGBoost and Deep Neural Network (DNN) were also evaluated using the same datasets and molecular fingerprints. Although all the three algorithms exhibited strong predictive performance (AUC > 0.85), the SVM model consistently outperformed them across all key metrics, underscoring its robustness and reliability for MAO-B inhibitor prediction. Top-ranked leads were subjected to molecular docking, MM-GBSA binding free energy analysis, and extended 200 ns molecular dynamics simulations using Amber v.24. Z1541760033 (Irinotecan hydrochloride) emerged as the most promising candidate, showing a docking score of -11.86 kcal/mol, strong hydrogen bonding with Gly55, Gln204 and Tyr186, π-π stacking with Tyr433, and a favorable MM-GBSA ΔG_total of -98.3 ± 0.19 kcal/mol. MD analysis confirmed stable conformational behavior and persistent active-site engagement. Additional candidates, including Z1546610486, also displayed favorable interaction profiles and dynamic stability. Leveraging chemically diverse training data and extended MD simulations, our pipeline prioritizes drug candidates with high predicted affinity and dynamic stability. This study highlights the power of combining machine learning classification with atomistic simulations for drug repurposing and provides a set of promising lead compounds for further experimental validation as MAO-B inhibitors targeting neurodegenerative diseases.