PICUPP: Protein Interaction Classification by Unlikely Profile Pair

A computational approach that infers protein-protein interactions from genome sequences is proposed in this paper. It is based on our recent observation that protein-protein interactions can be identified by a set of “unusual” protein-profile pairs in experimentally determined protein interactions. A pair of proteinprofiles is considered to be unusual if its occurrence in the given data is statistically unusual from what is expected at random. The proposed method, called PICUPP, sifts out unusual protein-profile pairs by comparing frequency distributions of their occurrences in the given data to what may result from random appearances. It is demonstrated that such unusual protein-profile pairs with statistically assessed confidences can be learned efficiently from the DIP database using a bootstrapping approach. We particularly illustrate that unusual (or, significant) protein-profile pairs can be characterized as pair wise interactions between the Pfam domains, Blocks protein families, or InterPro signatures. Such statistically significant protein-profile pairs can be used for predicting putative pairs of interacting proteins. Their prediction accuracy is around 86% and 90% when InterPro and Pfam profiles are used, respectively at 75% confidence level.