Computational identification of potential MMP-2 inhibitors in cancer using machine learning, molecular docking, and dynamics simulations

Abstract

Matrix metalloproteinase-2 (MMP-2) is a zinc-dependent endopeptidase which plays a key role in the extracellular matrix-remodeling and cancer metastasis. Nevertheless, despite the vast number of attempts, MMP-2 selective and low-toxicity development is a problematic area because of the insufficient selectivity and the off-target effect of the previous candidates. This work demonstrated that an integrated machine learning-driven virtual screening pipeline can be used to discover better selectivity, and binding stability novel MMP-2 inhibitors. Various models of classification were trained with the help of a set of different molecular fingerprints, and random Forest and radial-basis-function Support Vector Model of classification showed the best predictive results (AUC > 0.97, MCC > 0.86). These models have been used to filter the Maybridge compound library resulting in the selection of the top-ranked ones. Molecular docking and subsequent ADMET profiling of the shortlisted seven potential compounds yielded a list of 1. Molecular dynamics simulations (100 ns) showed that GK03418 and RH00707 had stable binding conformations similar to that of the reference inhibitor. Free energy landscape mapping and principal component analysis was another method that proved thermodynamic stability of GK03418. The energetics of binding free-energy calculations with MM/PBSA and MM/GBSA showed positive results and the most promising inhibitor was GK03418. In general, this paper provides a computationally sound and scalable structure of the discovery of selective MMP-2 inhibitors that have future anticancer applicability.