Identification of Potent CDK9 Inhibitors with Novel Skeletons via Virtual Screening, Biological Evaluation, and Molecular Dynamics Simulation

Cyclin-dependent kinase 9 (CDK9) is a pivotal therapeutic target for acute myeloid leukemia (AML), a hematologic malignancy characterized by limited effective treatments. In this study, we introduced an innovative virtual screening strategy combines machine learning models with molecular docking techniques. We constructed a comprehensive CDK9 data set, followed by applying molecular fingerprints and selecting machine learning models, particularly random forest models, for virtual screening. Molecular docking, combined with similarity and drug-likeness screening was used to ensure scaffold diversity from existing CDK9 inhibitors. This integrative strategy led to the identification of a new class of CDK9 inhibitors with scaffolds distinct from traditional CDK9 molecules. Here, Compound 2 demonstrated significant inhibitory activity against CDK9 and reduced cell viability effectively in hematologic tumor cell lines. Additional validation using molecular dynamics simulation clarified the binding mechanism of the inhibitor-protein complex. These findings highlight the effectiveness of our hybrid virtual screening approach in drug discovery, offering a promising new scaffold for CDK9 inhibitors.