Integrating structural homology with deep learning to achieve highly accurate protein-protein interface prediction for the human interactome

PIONEER2 outperforms AlphaFold3 in predicting protein-protein interfaces, especially for uncertain and unstructured complexes, enabling comprehensive human interactome analysis.

• Why it matters: Accurate identification of protein-protein interfaces is crucial for understanding disease mutations and developing personalized medicine, but structural data for many complexes is lacking.
• What they did: The team developed PIONEER2, integrating 3D structural similarity with geometric deep learning, and applied it to predict interfaces across all 352,124 human binary PPIs, validating with experimental mutations.
• The result: PIONEER2's predictions match experimental mutation impacts and better explain mutation patterns than AlphaFold3, providing a valuable resource for disease research and a user-friendly web platform.