Predicting and Validating Protein Interactions Using Network Structure

Pao-Yang Chen, Charlotte M. Deane, Gesine Reinert

Abstract: Author Summary For understanding the complex activities within an organism, a complete and error-free network of protein interactions which occur in the organism would be a significant step forward. The large amount of experimentally derived data now available has provided us with a chance to study the complicated behaviour of protein interactions. The power of such studies, however, has been limited due to the high false positive and false negative rates in the datasets. We propose a network-based method, taking advantage of the tendency of clustering in protein interaction networks, to validate experimental data and to predict unknown interactions. The integration of multiple protein characteristics (i.e., structure, function, etc.) allows our predictive method to significantly outperform two other approaches based on homology and protein-domain relationships on datasets which contain a large amount of interactions, but not much detailed information on the proteins involved in the interactions. In addition, our predictive score based on triadic interaction patterns improves over a pair-wise approach, suggesting the importance of network structure. Moreover, using pooled interactions as prior information, we find evidence for fundamental differences in protein interaction networks between eukaryotes and prokaryotes.