GOcSim: GO context-driven similarity

We present in this paper novel context-driven search techniques that compute the semantic similarities among different GO ontology terms. We implemented these techniques in a middleware search engine called GOcSim, which resides between user application and GO database. Most current research is focused on determining semantic similarities between GO ontology terms based solely on their IDs and proximity to one another in GO graph structure, while overlooking the contexts of the terms, which may lead to erroneous results. The context of a term T is determined by the set of other terms, whose existence is dependent on T. We propose novel techniques that determine the contexts of terms based on the concept of existence dependency. We present a stack-based sort-merge algorithm employing these techniques for determining the semantic similarities between GO terms. We evaluated GOcSim experimentally and compared it with four other methods. The results showed marked improvement.