An Integrative Computational Pipeline for CK2 Inhibitor Discovery in Triple-Negative Breast Cancer Using Virtual Screening, Molecular Dynamics, Machine Learning, and Density Functional Theory

Background: Triple-negative breast cancer (TNBC) remains among the most aggressive and therapeutically unresponsive subtypes due to the absence of ER, PR, and HER2 targets. Casein Kinase II (CK2), a pleiotropic serine/threonine kinase overexpressed in TNBC, represents a compelling target for rational drug design. Methods: Here, we present an AI-integrated benchmarking framework combining virtual drug discovery, molecular dynamics simulations, machine learning-driven QSAR modeling, and quantum-mechanical electronic structure analysis to identify potent CK2 inhibitors from natural product chemical space. Results: A validated XP docking protocol (ROC–AUC = 0.748) screened ~480,000 compounds, yielding seven hits, with superior binding to the reference inhibitor CX-4945. Among these, Anastatin B, 3,4,8,9,10-pentahydroxy-dibenzo-[b,d]pyran-6-one, Rhein, and aloe emodin acetate exhibited highly favorable docking scores (−11.6 to −13.1 kcal mol−1) and stable 200 ns binding dynamics, reflected by RMSD ≤ 2 Å and compact Rg trajectories. MM-PBSA/MM-GBSA analyses confirmed robust thermodynamic stability, while DFT-derived HOMO–LUMO gaps (3.8–4.3 eV) suggested optimal electronic reactivity for kinase inhibition. Machine learning QSAR models demonstrated strong predictive performance, with the best stacking models achieving test R2 ≈ 0.69 and consistent cross-validation performance (CV R2 ≈ 0.66–0.69), supporting reliable prediction of pIC50 values and prioritization of top-ranked scaffolds. Conclusions: Collectively, this integrative framework bridges AI-based learning and biophysical validation, establishing a reproducible paradigm for de novo CK2 inhibitor discovery in TNBC.