Ranked Gene Ontology Based Protein Function Prediction by Analysis of Protein-Protein Interactions

Computational function prediction of unknown protein is a challenging task in proteomics. As protein-protein interactions directly contribute to the protein function, recent efforts attempt to infer about proteins ' functional group by studying their interactions. Recently, use of hierarchical relationship between functional groups characterized by Gene Ontology improves prediction ability compared to hierarchy unaware "flat" prediction methods. As a protein may have multiple functional groups with different degrees of evidences, function prediction is viewed as a complex multi-class classification problem. In this paper, we propose a method which assigns multiple Gene Ontology terms to unknown protein from its neighborhood topology using a ranking methodology showing different levels of association. This work achieves precision of 0.74, recall of 0.67, and F-score of 0.73, respectively, on 19,247 human proteins having 8,548,002 interactions in between themselves.