Deciphering interaction fingerprints from protein molecular surfaces using geometric deep learning.

Area of Science:

• Structural biology
• Computational biology
• Biophysics

Background:

• Predicting biomolecular interactions from protein structure is challenging.
• The molecular surface contains geometric and chemical features (fingerprints) that dictate interaction modes.
• These fingerprints may be conserved across proteins with similar interaction functions, irrespective of evolutionary relationships.

Purpose of the Study:

• To introduce MaSIF (molecular surface interaction fingerprinting), a novel framework for capturing protein interaction fingerprints.
• To leverage geometric deep learning for learning these fingerprints from large-scale data.
• To demonstrate MaSIF's utility in predicting protein-ligand and protein-protein interactions.

Main Methods:

• Developed MaSIF, a geometric deep learning approach to analyze molecular surfaces.
• Trained MaSIF on datasets to learn interaction-relevant fingerprints.
• Applied MaSIF to three distinct prediction tasks: protein pocket-ligand prediction, protein-protein interaction site prediction, and protein-protein complex prediction.

Main Results:

• MaSIF successfully captures interaction fingerprints from protein molecular surfaces.
• The framework demonstrates efficacy across diverse prediction challenges, including identifying binding sites and predicting complexes.
• Achieved accurate predictions for protein-ligand and protein-protein interactions, showcasing the power of learned surface fingerprints.

Conclusions:

• MaSIF provides a powerful conceptual framework for understanding and predicting biomolecular interactions based on molecular surfaces.
• Geometric deep learning can effectively learn complex interaction patterns from structural data.
• This approach holds promise for advancing our understanding of protein function and facilitating protein design.