kinCSM

kinCSM predicts CDK2 inhibitor potency and classifies inhibition mode from small-molecule graph-based signatures to facilitate identification and characterization of potent CDK2 kinase inhibitors.

Key Features:

• Graph-Based Signatures: Employs graph-based signatures to capture physicochemical and geometric properties of small molecules.
• Predictive Accuracy: Identifies potent CDK2 inhibitors with Matthew’s Correlation Coefficients up to 0.74 and predicts CDK2 ligand-kinase inhibition constants (pKi) with Pearson’s correlation coefficient up to 0.76.
• Inhibition Mode Classification: Classifies inhibition mode with Matthew’s Correlation Coefficient up to 0.80 in cross-validation and 0.73 in blind tests.
• Consistent Performance: Demonstrates consistent performance on non-redundant blind tests with correlation coefficients of 0.66 for inhibitor identification and 0.68 for pKi prediction.

Scientific Applications:

• Kinase drug discovery: Prioritizes and enriches screening libraries for CDK2 by predicting pKi and inhibition mode for small-molecule ligands.
• Mechanistic insight: Identifies enriched chemical fragments and molecular composition features associated with potent CDK2 inhibitors to inform ligand-kinase interaction hypotheses.

Methodology:

Uses supervised learning on graph-based signatures to build predictive models that do not require prior kinase information and analyzes molecular composition to identify enriched chemical fragments associated with potent CDK2 inhibitors.