In silico identification of NPACT-derived PAK1 inhibitors using machine learning, molecular docking, and dynamic...

P21-activated kinase 1 (PAK1) is a key serine/threonine kinase involved in cytoskeletal remodeling, cell proliferation, and survival, and its aberrant activation has been strongly associated with tumorigenesis in multiple cancer types. Owing to its central role in oncogenic signaling and its more favorable safety profile compared with other PAK isoforms, PAK1 has emerged as an attractive therapeutic target. However, the development of selective and non-toxic PAK1 inhibitors remains challenging, particularly due to the conserved nature of kinase ATP-binding sites. In this study, an integrated in silico strategy was employed to identify potential PAK1 inhibitors from the Naturally occurring Plant-based Anti-cancer Compound-Activity-Target database (NPACT). Machine learning models were first constructed to predict binding free energies, with the K-Nearest Neighbors model showing the best performance. Six top-ranked compounds were subsequently subjected to molecular docking, revealing favorable binding affinities and consistent interactions with key active-site residues of PAK1. Binding affinity was further refined using steered molecular dynamics simulations, which showed a strong correlation between maximum rupture force and experimental inhibitory activity, supporting the reliability of the steered molecular dynamics (SMD)-derived binding free energies. Toxicity prediction suggested moderate to low toxicity profiles for all candidates, with no predicted cardiotoxicity. Density functional theory calculations further elucidated the electronic properties, frontier molecular orbitals, and molecular electrostatic potential features relevant to protein-ligand interactions. Overall, this comprehensive computational investigation highlights several NPACT-derived phytochemicals as promising PAK1 inhibitors, providing valuable insights for future experimental validation and anticancer drug development.