About

PIONEER2 is an equivariant geometric deep learning pipeline to generate precise partner-specific protein-protein interface residue predictions with high accuracy. PIONEER2 integrates a comprehensive set of biophysical, evolutionary, structural, and sequence features, together with PrePPI-predicted interface information based on structural alignment of query proteins to homologous PDB templates. Extensive benchmark tests show that PIONEER2 is significantly better than PIONEER and other similar algorithms. Even compared with the latest AlphaFold3 method, PIONEER2 achieves superior performance when AlphaFold3 produces low confidence structures. We estimate below that these cases correspond to nearly half of all human protein-protein interactions. Together with atomic resolution structures of complexes in the PDB, we established a comprehensive multiscale structurally informed human protein-protein interactome consisting of 352,124 experimentally-detected binary interactions with interface information for each. Our PIONEER2-generated structurally informed human interactome will serve as an invaluable resource for investigating disease mechanisms and advancing personalized therapeutic strategies.

Version

Version 1.1 January 2024

Contact

The PIONEER2 interactive browser is provided by the Yu lab at Cornell University. Please contact us at weill-pioneer@cornell.edu for any issues.