Effectively integrating information content and structural relationship to improve the GO-based similarity measure between proteins
The Gene Ontology (GO) provides a knowledge base to effectively describe proteins. However, measuring similarity between proteins based on GO remains a challenge. In this paper, we propose a new similarity measure, information coefficient similarity measure (SimIC), to effectively integrate both the information content (IC) of GO terms and the structural information of GO hierarchy to determine the similarity between proteins. Testing on yeast proteins, our results show that SimIC efficiently addresses the shallow annotation issue in GO, thus improves the correlations between GO similarities of yeast proteins and their expression similarities as well as between GO similarities of yeast proteins and their sequence similarities. Furthermore, we demonstrate that the proposed SimIC is superior in predicting yeast protein interactions. We predict 20484 yeast protein-protein interactions (PPIs) between 2462 proteins based on the high SimIC values of biological process (BP) and cellular component (CC). Examining the 214 MIPS complexes in our predicted PPIs shows that all members of 159 MIPS complexes can be found in our PPI predictions, which is more than those (120/214) found in PPIs predicted by relative specificity similarity (RSS). Integrating IC and structural information of GO hierarchy can improve the effectiveness of the semantic similarity measure of GO terms. The new SimIC can effectively correct the effect of shallow annotation, and then provide an effective way to measure similarity between proteins based on Gene Ontology.