Efficient discovery of binding motif pairs from protein-protein interactions

Abstract

Protein interaction sites mediate protein interactions. Current methods to determine protein interaction sites are still preliminary. This dissertation introduces a new motif concept, called binding motif pairs, to model regular patterns located at protein interaction sites. We propose two different approaches to mining binding motif pairs from protein interaction data. The first method is based on a fixed-point theory where a motif pair is defined as a point of the transformation function. The fixed points of the transformation function that are also statistically significant are the binding motif pairs. The second method is built on an all-versus-all interaction relationship between two protein groups in protein interaction networks. This idea reflects an inherent binding mechanism between proteins. We call those groups with all-versus-all relationship interacting protein group pairs and use a data mining method to efficiently discover them. The binding motif pairs are generated from the group pairs using standard motif discovery algorithms. The effectiveness of the two methods is demonstrated by comprehensive validation results.