SAKE-PP: A Spatial-Attention Equivariant Network for Accurate Ranking of Protein–Protein Interaction Models

Accurate prioritization of near-native protein–protein interaction (PPI) models remains a major bottleneck in structural biology. Here, we present SAKE-PP, a physics-inspired, spatial-attention equivariant graph neural network that directly regresses interface RMSD (iRMSD) without native references. Trained with a hierarchical iRMSD-guided sampling strategy on PDBBind, SAKE-PP integrates force-field-like attention with Laplacian-eigenvector orientation to couple local interaction forces with global topology. On the 2024PDB benchmark of 176 heterodimers, SAKE-PP improves AF3-decoy selection by 13.75% (iRMSD) and 12.5% (DockQ) and consistently outperforms the AF3 ranking score in overlap, hit-rate, and correlation metrics. In zero-shot evaluation on 139 antibody–antigen complexes, SAKE-PP increases correlation by 0.4. By promoting geometrically near-native, energetically plausible interfaces to the top ranks, SAKE-PP reduces wasted MD trajectories and improves refinement reliability. Overall, SAKE-PP provides a robust, plug-and-play scoring function that streamlines PPI evaluation and accelerates downstream structure-guided drug-design workflows.