Discovery of PIM-1 kinase inhibitors from marine natural products through machine learning and structure-based screening

Abstract

PIM-1 kinase, a serine/threonine kinase implicated in several cancers, has emerged as a promising yet underexplored target for anticancer therapy. This study aimed to identify the potential of PIM-1 inhibitors from marine natural products by integrating machine learning (ML)-based quantitative structure-activity relationship (QSAR) modeling with structure-based virtual screening. Eleven ML models were developed and evaluated using experimentally validated ChEMBL data. The best-performing model was used to screen the Chemical Marine Natural Product Database (CMNPD), yielding 15 candidate compounds. Molecular docking prioritized four hits based on similarity of binding interactions at the catalytic site to the co-crystallized reference ligand, rather than on docking score alone. These hits were further assessed using 100 ns molecular dynamics (MD) simulations and MM/GBSA binding free-energy calculations. ADMET profiling was also performed to evaluate pharmacokinetic and toxicity properties. Among the screened compounds, Hit-1 (CMNPD11687) emerged as the lead candidate. It showed stable binding throughout MD simulations, with an average RMSD of 4.10 ± 1.67 Å, low atomic fluctuations (RMSF), and favorable binding free energy (ΔG- Total = -18.45 ± 2.70 kcal/mol). Hit-1 also maintained key protein-ligand interactions more consistently than the other hits. ADMET analysis predicted favorable drug-like and non-toxic properties. These findings identified CMNPD11687 as a promising marine-derived scaffold for PIM-1 inhibition and demonstrated the value of combining ML-driven screening with molecular simulation to accelerate early-stage anticancer drug discovery.