Machine learning-based virtual screening integrating pharmacophores, docking and molecular descriptors for discovery of CDK2 inhibitors

Cyclin-dependent kinase 2 (CDK2) serves as a key regulator of cell cycle progression. Studies have confirmed that aberrant CDK2 activity is closely associated with tumor initiation and progression, exerting a critical impact on ovarian cancer, solid tumors, small-cell lung cancer, and metastatic breast cancer. In addition, CDK2/4/6 inhibitors can suppress the proliferation of drug-resistant tumors. Accordingly, the development of CDK2 kinase inhibitors represents a promising therapeutic strategy for oncology treatment. In recent years, machine learning-driven virtual screening has been increasingly employed in novel drug discovery, providing promising paradigms for the rational development of CDK2 inhibitors. In this study, a naïve Bayesian classification (NBC) model was established by integrating molecular descriptors, molecular fingerprints, molecular docking and pharmacophore models for the virtual screening of potential CDK2 inhibitors. The verification results of the test set revealed that the optimal model exhibited superior performance in discriminating active from inactive small molecules. Moreover, the optimal mode clarifies critical structural fragments that are beneficial or detrimental to the inhibition of CDK2 activity, which facilitates the rational design and discovery of novel CDK2 inhibitors. Subsequently, the optimal model was employed to screen ChemDiv's representative diversity libraries, leading to the identification of four potential CDK2 inhibitors. Furthermore, 100 ns molecular dynamics (MD) simulations verified that the complexes formed by the four hit compounds with CDK2 exhibited favorable stability. In conclusion, the current research provides valuable theoretical guidance and practical references for the further development of novel CDK2 inhibitors.