Machine Learning and Virtual Screening Methods to Discover Potential Cyclin-Dependent Kinase 2 (CDK2) Inhibitors

Area of Science:

• Computational chemistry
• Drug discovery
• Molecular biology

Background:

• Cyclin-dependent kinase 2 (CDK2) is crucial for cell cycle regulation.
• CDK2 is a significant therapeutic target for cancer treatment.
• Novel inhibitors are needed to advance cancer therapy.

Purpose of the Study:

• To identify novel Cyclin-dependent kinase 2 (CDK2) inhibitors.
• To utilize an integrated computational strategy combining machine learning and structure-based methods.
• To discover potential anticancer drug candidates.

Main Methods:

• Developed a computational pipeline including Lipinski's Rule of Five, machine learning (ML) prediction, molecular docking, and molecular dynamics simulations (MDs).
• Trained and validated ML models using a CDK2 inhibitor dataset from ChEMBL, with random forest showing optimal performance (AUC-ROC 0.90, accuracy 0.84).
• Screened the InterBioScreen database, docked promising compounds against CDK2 (PDB ID: 2FVD), and performed 100 ns MDs for stability analysis.

Main Results:

• Identified 187 potential CDK2 inhibitors through ML prediction.
• Two compounds, STOCK4S-00019 and STOCK4S-00025, exhibited docking scores comparable to the reference ligand.
• Molecular dynamics simulations confirmed stable binding and favorable interactions for the identified compounds.

Conclusions:

• STOCK4S-00019 (hit1) and STOCK4S-00025 (hit2) demonstrate significant potential as CDK2 inhibitors.
• These findings validate the efficacy of integrated computational approaches in anticancer drug discovery.
• Further experimental validation is recommended for the identified hit compounds.