RGRank: Ranking Semantically Related Genes

Biologists often need to know the set of genes that are semantically related to a given set of genes. However, most current similarity measures determine the semantic (biological meaning) similarities between the set rather than identifying the set. Moreover, these similarity measures determine the semantic similarities between the genes based solely on the proximity of their GO term annotations to each other in GO graph while overlooking the structural dependencies between these terms, which may lead to lower recall and precision of results. We propose in this paper a search engine called RGRank, which overcomes the limitations of current similarity measures outlined above as follows: (1) Given a set S of genes, RGRank would return a set of genes, where each gene in is semantically related to each gene in, (2) RGRank ranks gene results by relevance to their semantic similarities to the input genes, and (3) RGRank employs the concept of existence dependency to determine the structural dependencies among the GO terms annotating a given set of gene. We evaluated RGRank experimentally and compared it with three existing methods. Results showed marked improvement.