Computational design of MMP-13 inhibitors using a combined approach of machine learning, docking, and molecular dynamics

Matrix metalloproteinase-13 (MMP-13) is a zinc-dependent endopeptidase involved in extracellular matrix degradation and inflammation, contributing to the progression of various diseases. This study applied an integrated computational approach encompassing QSAR modeling, machine learning (ML), scaffold analysis, docking, and molecular dynamics (MD) simulations to investigate the structure-activity relationships and binding mechanisms of MMP-13 inhibitors. A curated dataset of 1,741 unique compounds from ChEMBL was used to develop predictive QSAR models based on PubChem fingerprints. Among eight regression models, LGBM, SVR, and RF exhibited superior predictive performance, with LGBM achieving the best generalization (test RMSE = 0.825, R² = 0.646, Q² = 0.628). Similarly, LGBM and SVM classifiers demonstrated high accuracy (0.802) and MCC (0.589) with test data. Docking analysis identified three top candidates (ChEMBL1770157, ChEMBL425020 and ChEMBL5182668) with strong binding affinities of -10.98, -10.93 and -10.80 kcal/mol, respectively. The identified interaction hotspots, particularly Thr245, Ala186, Leu185, Val219, and the highly versatile His222, represent key residues to target for enhancing binding affinity. Subsequent 200 ns MD simulations confirmed their structural stability and favorable binding dynamics within the MMP-13 active site. Scaffold analysis revealed the predominance of sulfonamide and carboxyl-containing polar functional groups, known to be important for solubility and target binding. The findings underscore the importance of physicochemical and structural attributes in MMP-13 inhibitor design and support the therapeutic potential of targeting MMP-13 in diverse pathological contexts.