The identification of protein-protein interaction sites is crucial for mutant design and prediction of the protein-protein network. We analyzed the physicochemical characteristic of protein interaction sites and developed the prediction method using sequence profile of neighboring residues, physicochemical parameters, and support vector machines (SVM). In both homo and hetero complexes, the complexes which have low hydrophobic (higher charged) interacting sites and a broader gap between protomers, increase with decreasing interfacial surface ratio. It is assumed that complexes whose binding energy is hydrophobic energy are major in higher interfacial surface ratio complexes, while in lower interfacial surface ratio complexes, complexes whose major binding energy is polar energy including hydrogen bonds increase.
The prediction accuracy of interacting sites is the highest when sequence profiles and accessible surface area of neighboring residues are used as input vectors for the SVM. The effects of surface flatness, sequence conservation, and hydrophobic moment are not remarkable. The prediction accuracy is low in complexes which have polar interacting sites and low interacting sites ratios, while it is higher in complexes which have hydrophobic interacting sites and high interacting sites ratios. The comparison with other methods is also discussed.