Application of an algorithm for the computation of functional similarity in the research of protein interaction

The similarity among gene products based on gene ontology terminology is mainly measured through semantic similarity, which first computes the semantic similarity among gene ontology terminologies and then computes the similarity among gene products related to GO terminologies. Methods for computing the semantic similarity in the first step are either information-quantity-based or semantic-distance-based. The former category considers the information quantity of the nodes themselves while neglecting the structure of gene ontology; the later category neglects the semantic relations among the nodes in gene ontology and the uneven distribution of connections between nodes. This paper proposes a comprehensive approach where the similarity is first computed on the terminological level, inducing density-constraining and depth-constraining variables based on the microscopic structure (density and depth of the nodes) of the gene ontology nodes, then based on node properties and node-based instances, the comprehensive semantic similarity is computed. Published researches regarding the functional similarity in the second step are rare. Considering the defect of computing the functional similarity by averaging, a new weighted functional similarity algorithm is proposed. The effects of 4 algorithms: average-based, maximum-based, Wang and WSim are evaluated by means of ROC curve using the interaction data of human protein and fruit fly protein from the DIP database. Result of comparison proves that the proposed weighted algorithm is favorable.