PreSPI: Prediction System for Protein Interaction

The accumulation of protein and its associated data on the Internet gives us the chance to computationally identify protein structures and functions using the data. More specifically, the accumulation of Protein-Protein Interaction (PPI) and domain data enables us to computationally predict protein interactions for experimentally unidentified protein interactions. The benefits of computational prediction of PPI are obvious. First and foremost, mass prediction of PPI is possible at low cost. If a large-scale protein interaction network is constructed from the massive PPI information, biologists can try to predict the functions of unknown proteins [3], from the PPI network. The prediction can also help in finding critical proteins out of PPI information. Besides, biologists can have some hints in assigning priorities to the proteins or domains to be tested. PreSPI is unique software in that it uses domain combination pair information for the prediction. The PPI prediction method of PreSPI has originated from the domain based PPI prediction [1], eliminating some of drawbacks. Previous domain based PPI researches usually considered interactions of a pair of domains only in the prediction, that is, the researchers assume that an interaction of a domain pair is independent of another pair for computational simplicity. In contrast, domain combination based approach interprets the protein interaction as the result of interactions of multi-domain pairs or interactions of domain groups. The PPI prediction algorithm of the system is well-studied by Han's research group and prediction accuracy is revealed superior to other conventional domain based prediction method [2]. With 80% of the set of interacting protein pairs in the DIP as the learning set, on average, 77% sensitivity and 95% specificity were achieved for the test groups containing common domains with the learning set of proteins within our system.