Effective identification of kinase-specific phosphorylation sites based on domain-domain interactions

Deciphering interactions between protein kinases (PKs) and the target substrates are fundamental for understanding the molecular mechanisms of phosphorylation. Although all PKs have been identified in eukaryotes, the sites that they phosphorylate are only partially elucidated. Experimental identification of phosphorylation sites is labor and resource intensive, so developing an effective method to computationally predict potential sites is increasingly important. Here, a novel method was proposed for the identification of kinase-specific phosphorylation sites based on domain-domain interactions (DDIs). Using difference analysis between phosphorylation sites and non-phosphorylation sites, the distinct neighbor residues around the phosphorylation sites were firstly identified in our study. The results of difference analysis by rank sum test indicate that 19, 26, 26 and 10 neighbor residues are distinctive for the phosphorylation site prediction of four major serine (S)/threonine (T) protein kinase families-CDK, CK2, PKA and PKC respectively. Then the correlation coefficients were computed to represent the interaction between PK domains and phosphorylation domains of the substrate proteins. Four random forest models (RF) were constructed to predict the potential sites, the CDK, CK2, PKA and PKC RF models yield an accuracy of 86.57%, 91.44%, 87.02% and 80.11% on the test sets respectively. Finally, the new substrate proteins in protein data bank (PDB) were extracted to verify the distinct residues around the phosphorylation sites at 3D-structural level and the results further demonstrate the reliability of our models, which indicate that our method will be a useful tool for elucidating dynamic interactions between PKs and their substrates.