Predicting interface residues for protein polymers based on module division

In the field of computational biology, AlphaFold has facilitated remarkable progress in predicting protein structures. However, for some multimeric complexes, the accuracy of its predictions still needs improvement. Enhancing the prediction of binding sites in protein oligomers contributes to the prediction of overall structures, and the details of binding sites are crucial for understanding the functional mechanism, affinity, and specificity of complexes. The approach presented here is independent of black-box modeling, thereby facilitating a clearer elucidation of the underlying biological mechanisms of protein-protein interaction interfaces. The study investigates protein interaction sites, mainly for trimers and tetramers, in the Protein Data Bank (PDB) database, using a modular division method. By ranking modules based on the linear combination of the square of the solvent-accessible surface area and the internal contact area (SSAIA), it was found that interface modules are more likely to appear in surface modules with lower values. Among 44 964 monomers, 92.25% of the monomer interface residues are found in the three modules with the smallest SSAIA values. In 5240 trimers, 98.63% of the complexes had at least one chain correctly predicted, and in 7311 tetramers, 98.92% of the complexes had at least one chain correctly predicted, with an average of approximately 3 out of 4 residues in the top-ranked interface modules being interface residues. The results of comparison with other methods in the Benchmark 3.0 dataset and 12 targets of the CASP15 competition validated the effectiveness of our approach, providing a new perspective for predicting binding sites in multimer interactions.