Discovery of potential ALK inhibitors from Glycosmis phytochemicals via integrated machine learning and umbrella sampling approaches

Abstract Anaplastic lymphoma kinase (ALK) is a key therapeutic target for various cancers because of its central role in tumor growth and survival signaling. This study employed an integrated computational workflow combining machine learning (ML), molecular docking, molecular dynamics (MD) simulations, umbrella sampling (US), and toxicity prediction to identify potential ALK inhibitors from 307 phytochemicals of the Glycosmis genus. A previously validated CatBoost regression model predicted binding free energies, leading to the selection of 26 compounds with ΔGML < −11 kcal mol−1. Molecular docking identified six top candidates with favorable binding affinities ranging from −5.856 to −9.667 kcal mol⁻¹, which subsequently underwent 200 ns MD simulations. Structural analyses confirmed the stability of all ALK–ligand complexes, while interaction studies revealed extensive hydrophobic contacts and conserved hydrogen bonds involving key ALK residues. US calculations estimated binding free energies from −8.354 to −13.272 kcal mol−1, with compound 54 exhibiting the strongest affinity (ΔGUS = −13.272 kcal mol−1). Several alkaloids, including acrifoline, heptazoline, and atalaphyllidine, also demonstrated promisingbinding properties and reported anticancer activities. Toxicity prediction indicated acceptable safety profiles, highlighting Glycosmis-derived phytochemicals as promising potential ALK inhibitor candidates and supporting the utility of ML-guided virtual screening combined with advanced molecular simulations for natural product-based anticancer drug discovery. Nevertheless, biochemical and cellular experiments are required to confirm their ALK inhibitory activity, selectivity, and therapeutic potential.