Combining machine learning with molecular docking and steered molecular dynamics to identify potent ALK inhibitors from the EMNPD database

Abstract

Anaplastic lymphoma kinase (ALK) is a key oncogenic driver in cancers such as non-small cell lung cancer (NSCLC), where gene fusions and mutations lead to constitutive kinase activation. Despite the clinical success of current ALK inhibitors, acquired resistance remains a major limitation, underscoring the need for novel and structurally diverse therapeutic agents. In this study, we applied an integrated computational approach combining machine learning (ML), molecular docking, molecular dynamics (MD), and steered molecular dynamics (SMD) to identify potential ALK inhibitors from the Endophytic Microorganism Natural Product Database (EMNPD). CatBoost was selected as the optimal ML model (RMSE = 0.933, MAE = 0.719, and R = 0.808) and used to prioritize candidates with predicted binding free energies below −11 kcal mol⁻¹. Molecular docking of 369 selected compounds identified five top hits (IDs 100, 248, 254, 277, and 307) with ΔG_dock < −9.0 kcal mol⁻¹. MD simulations confirmed stable protein–ligand interactions, with RMSD < 0.25 nm and persistent hydrogen bonding, especially for compounds 248, 254, and 277. SMD simulations indicated that compounds 254 and 277 exhibited the highest rupture forces, pulling work, and ΔG_SMD values of –9.582 and −9.649 kcal mol⁻¹, respectively, suggesting strong mechanical stability. Pharmacokinetic and toxicity analyses highlighted compounds 248 and 254 as the most balanced candidates. This study demonstrates that integrating ML with docking and dynamic simulations is an effective strategy for discovering natural product-derived ALK inhibitors, offering promising leads for the development of next-generation targeted therapies.