A NOVEL SEMANTIC SIMILARITY SCORE FOR PROTEIN DATA ANALYSIS

Abstract

Background: Semantic similarity between Gene Ontology (GO) terms underpins numerous applications in genomics, yet existing measures typically rely on a single strategy — topology, information content, or node/edge structure — and no individual approach consistently outperforms the others.
Objective: This study develops a similarity evaluation measure for GO terms that jointly incorporates information content, term membership, and topology-based similarity.
Methods: The proposed method computes a topological depth factor from the number of descendants of a term, an information-content-based shortest-path distance to the Most Informative Common Ancestor, and a Gaussian membership function; these are combined via Best Match Average and evaluated on positive and negative UniProt interaction datasets and Pfam protein family clans.
Results: The proposed method achieved higher Pfam similarity and Pearson correlation than Resnik, Lin, Wang, GOGO, Lord, Al-Mubaid, and TopoICSim overall, with the strongest correlation observed for the Molecular Function (MF) ontology.
Conclusion: The experimental results demonstrate a clear advantage of the proposed method over existing semantic similarity measures for GO term and protein-clan analysis.

Keywords: Annotation, Information Content, Membership, Topology

HIGHLIGHTS:
1. An improved approach for semantic similarity evaluation of Gene Ontology (GO) terms is developed by combining information content and topological factors.
2. A multi-factor similarity measure integrating fuzzy clustering, term membership, and a weight function is proposed.
3. The proposed method achieves the highest Pfam similarity and the highest correlation for the Molecular Function (MF) ontology among all compared approaches.