Balaguru Ravikumar

The paper introduces CCGM, a compound representation model that improves chemotype exploration and screening by using a simplified graph structure to capture pharmacophore and chemotype features.

The paper proposes a two-stage machine learning method that combines single-dose and dose-response data to improve compound-kinase activity prediction, achieving high accuracy and hit rates in experimental validation.

This study uses AlphaFold2 Steering and machine learning to model kinase conformations and identify state-specific inhibitors, achieving a 53% hit rate in virtual screening and validating conformation-specificity for FLT3 ligands.