Abstract
Background: Aurora kinase B (AurB) is a pivotal regulator of mitosis, making it a compelling target for cancer therapy. Despite significant advances in protein kinase inhibitor development, there are currently no AurB inhibitors readily available for therapeutic use. Methods: This study introduces a machine learning-assisted drug repurposing framework integrating quantitative structure-activity relationship (QSAR) modeling, molecular fingerprints-based classification, molecular docking, and molecular dynamics (MD) simulations. Using this pipeline, we analyzed 4680 investigational and approved drugs from DrugBank database. Results: The machine learning models trained for drug repurposing showed satisfying performance and yielded the identification of saredutant, montelukast, and canertinib as potential AurB inhibitors. The candidates demonstrated strong binding energies, key molecular interactions with critical residues (e.g., Phe88, Glu161), and stable MD trajectories, particularly saredutant, a neurokinin-2 (NK2) antagonist. Conclusions: Beyond identifying potential AurB inhibitors, this study highlights an integrated methodology that can be applied to other challenging drug targets.