kinCSM: Using graph-based signatures to predict small molecule CDK2 inhibitors.

Area of Science:

• Computational chemistry
• Molecular modeling
• Drug discovery

Background:

• Protein phosphorylation is crucial in cellular signaling, making kinases therapeutic targets.
• Computer-aided drug discovery aids in prioritizing screening libraries.
• Limited understanding exists regarding potent kinase inhibitor characteristics.

Purpose of the Study:

• To develop kinCSM, an integrative computational tool for identifying potent cyclin-dependent kinase 2 (CDK2) inhibitors.
• To quantitatively predict CDK2 ligand-kinase inhibition constants (pK_i).
• To classify inhibitor types based on binding modes and provide insights into ligand-kinase interactions.

Main Methods:

• Development of kinCSM using supervised learning and graph-based signatures.
• Capturing physicochemical and geometric properties of small molecules.
• Validation using cross-validation and a nonredundant blind test.

Main Results:

• Accurate identification of CDK2 inhibitors (MCC up to 0.74) and prediction of inhibition constants (Pearson's correlation up to 0.76).
• Consistent performance on blind test sets (MCC 0.66, Pearson's correlation 0.68).
• Successful classification of inhibition types (MCC up to 0.80) and identification of enriched chemical fragments.

Conclusions:

• kinCSM is a valuable tool for guiding kinase drug discovery.
• The tool provides insights into molecular mechanisms of ligand-kinase interactions.
• kinCSM facilitates fast and accurate screening of potential CDK2 inhibitors.