Using gnome wide data for protein function prediction by exploiting gene ontology relationships

Many new therapeutic techniques depend not only on the knowledge of the molecules participating in the biological phenomena but also their biochemical function. Advancements in prediction of new proteins are immense if compared with the annotation of functionally unknown proteins. To accelerate the personalized medicine effort, computational techniques should be used in a smart way to accurately predict protein function. In this paper, we propose and evaluate a technique that utilizes integrated biological data from different online databases. We use this information along-with Gene Ontology (GO) relationships of functional annotations in a wide-ranging way to accurately predict protein function. We integrate PPI (Protein Protein Interactions) data, protein motifs information, and protein homology data, with a semantic similarity measure based on Gene Ontology to infer functional information for unannotated proteins. Our method is applied to predict function of a subset of homo sapiens species proteins. The integrated approach with GO relationships provides substantial improvement in precision and accuracy when compared to functional links without GO relationships. We provide a comprehensive assignment of annotated GO terms to many proteins that currently are not assigned any function.