Semantic Labeling to Identify Reliable Interactions in the Protein Interaction Network

Recent high-throughput techniques (e.g. yeast-two- hybrid) have generated large-scale protein interaction data, which are believed to be highly noisy. Assessing the reliability of these interactions becomes an important task in proteomics. In this paper, we first label the protein interaction network by Gene Ontology (GO) annotations, and then identify reliable interactions based on the statistical significance of these semantic labels. Experimental results show that our proposed reliability measure can be used to detect reliable interactions with a better performance.