CSM-Potential

CSM-Potential identifies interaction-prone regions on protein surfaces using geometric deep learning applied to predicted holo 3D protein structures to connect structural models with protein interaction function.

Key Features:

• Geometric deep learning: Uses a geometric deep learning approach to analyze protein surface geometry.
• Interaction prediction: Predicts regions on protein surfaces likely to mediate protein-protein and protein-ligand interactions.
• Holo 3D structural input: Leverages predicted holo 3D structures from recent protein structural modeling across diverse proteins.
• Protein-protein performance: Achieves ROC AUC values up to 0.81 for identifying potential protein-protein binding sites.
• Ligand classification performance: Demonstrates biological ligand classification accuracy up to 96%.
• Validation: Evaluated using independent blind tests to assess generalizability.

Scientific Applications:

• Interaction hotspot identification: Locates potential interaction hotspots on protein surfaces to inform functional hypotheses.
• Understanding molecular mechanisms: Aids interpretation of how protein interactions contribute to molecular mechanisms.
• Therapeutic design: Informs design and prioritization of therapeutic interventions by identifying candidate binding sites.
• Structure-function linkage: Links predicted holo 3D structural models to biological function through predicted interaction sites.

Methodology:

Applies geometric deep learning to holo 3D protein structures predicted by protein structural modeling and evaluates performance via independent blind tests.