Predicting conformation of protein complexes by determining statistically significant domain-domain interactions

Abstract

The present study proposes a method to predict the conformation of protein complexes by using statistically significant domain-domain interactions (DDIs). High-throughput methods for detecting protein interactions generate a significant number of false-positives, and especially the combinatorial method of protein-complex purification and mass spectrometry detect both direct and non-direct interactions i.e. "bait-prey" and "prey-prey" interactions making it difficult to predict the conformation of complexes. Therefore in this work we utilized the DDIs as a means to support the interactions and subsequently to predict the conformation of complexes. As the first step, we extracted 312 statistically significant DDIs out of 1,162 DDIs underlying 3, 118 protein-protein interactions (PPIs) of Arabidopsis thaliana by using Fisher's exact test. Next, 67 protein complexes were obtained by applying a graph clustering algorithm to the PPI network. Finally, we discussed the conformation of protein complexes based on DDI information extracted in the first step. Information on significant DDIs can also be utilized to annotate unknown function proteins and to predict localization of proteins with confidence. © 2009 The Japanese Society for Plant Cell and Molecular Biology.