(A) weighted domain combination based protein-protein interaction prediction method

Abstract

Recently, many domain based protein-protein interaction prediction methods have been developed. However, the methods seldom consider the collaborations of domains due to the lack of appropriate computational model or complexity. In this paper, we propose a weighted domain combi-nation based protein interaction prediction method. Unlike conventional domain combination methods, the proposed method develops a model to quantify the influence of a domain combination pair in a protein interaction. The results of the quantification are summarized and stored in a matrix called Interaction Significance (IS) matrix. IS matrix contains information on weighted domain combinations and domain combination pair’s coupling power, which denote possibilities of domain collaboration and being the main body of a protein interaction respectively. In this paper, we also ex-tracted primary interacting domain combination pair in a PPI; it can give a clue whether the assigned weights are reliable or not by comparing the re-sults with real physical interactions. In S. cerevisiae, approximately 63% of sensitivity and 86% of specificity were achieved when we predict PPI by using PPI data from DIP and IntAct and domain data from Pfam-A; and prediction accuracy gradually increased as the size of the learning set grows. In addition, 83% of accuracy was achieved when we extracted primary interacting DC pair of PPI from iPfam and verified the results using PDB crystal structure. Our computational model provides several benefits against conventional domain or domain combination based PPI prediction methods. The PPI prediction accuracy can be improved, and we can provide addi-tional information relevant to PPI in our prediction model. Biologists can prepare their plan for experiments investigating a protein interaction more efficiently by focusing more on the primary DC-pair than the other DC-pairs. In addition, information of primary interacting DC pair is useful in predicting domain interface d...