Machine Learning-Guided Virtual Screening Identifies a Novel MELK-Targeted Antiproliferative Lead Compound with G2/M Arrest Activity in Triple-Negative Breast Cancer and Lung Adenocarcinoma Cells

Maternal embryonic leucine zipper kinase (MELK) is a cell cycle-regulated serine/threonine kinase frequently overexpressed in human malignancies that has been controversially pursued as an anticancer target. Here we applied an integrated computational workflow combining four machine learning regressors (Random Forest, Gradient Boosting, XGBoost, LightGBM), hierarchical Glide docking (HTVS to SP to XP), MM/GBSA rescoring, ADMET prediction, and 100 ns molecular dynamics simulation to a ~15.58 million-compound commercial library, yielding 12 prioritized candidates. Three commercially available compounds (CPD4, CPD6, CPD10) were evaluated in an antiproliferative panel. CPD6 emerged as the most active hit, with IC50 values of 1.44 μM in MDA-MB-468 and 1.66 μM in H1299, modestly exceeding the reference compound MELK-8a under the same assay conditions (2.48 and 2.51 μM, respectively). Cellular thermal shift assay in cell lysates provided biochemical evidence of MELK engagement, and 100 ns molecular dynamics simulations supported a plausible computationally predicted binding mode in the MELK kinase domain. Colony formation, EdU incorporation, and cell cycle analyses demonstrated dose-dependent antiproliferative effects and G2/M arrest, with concomitant downregulation of CDC25C, CDK1, and Cyclin B1. These findings support CPD6 as a structurally novel, cell-active antiproliferative lead with a chemically distinct scaffold, evidence of MELK target engagement, preferential activity against cancer cells, and a clear G2/M arrest phenotype; reduced MELK substrate phosphorylation and partial rescue by MELK overexpression further support MELK-dependent activity, warranting further hit-to-lead optimization and selectivity profiling.