Discovery of Novel AURKA Inhibitors for Triple-Negative Breast Cancer (TNBC) Therapy via a Hybrid Virtual Screening Pipeline, Biological Evaluation and Molecular Dynamics Simulation

Abstract

Aurora kinase A (AURKA) is a pivotal driver of malignant progression and poor prognosis in triple-negative breast cancer (TNBC). In this study, we developed a cascaded AI-driven virtual screening pipeline, integrating sequence-based affinity prediction (PSICHIC), equivariant deep learning docking (KarmaDock), and geometric rescoring (DeepDock) to identify novel AURKA inhibitor candidates. From an in-house 160,000-compound screening library assembled from commercially available collections, three leads (compounds 3, 5, and 8) were selected and subsequently validated via HTRF biochemical assays, exhibiting potent enzymatic inhibition with IC50 values of 157 nM, 21.64 nM, and 46.03 nM, respectively. Cell-based assays demonstrated that compound 3 produced stronger short-term cell-growth inhibition in MDA-MB-231 (TNBC) cells compared to clinical benchmarks MLN8237 and CCT241736, whereas compounds 3 and 5 showed cell-growth inhibition in NIH/3T3 cells within the same concentration range as the reference inhibitors. Triplicate 500 ns molecular dynamics simulations supported stable binding modes of the identified leads in the AURKA binding pocket. Additional computational analyses further provided supportive information for subsequent lead optimization. This study provides a transparent and open-source workflow for AI-assisted identification of AURKA-active chemotypes.