Protein Enrichment in PPI Prediction 1 Enrichment of Protein Utilization in Protein-Protein Interaction Prediction by Adjusting e-value of InterProScan and Using Gene Ontology

Among the computational prediction methods, domain based protein interaction prediction approaches are getting popular. However, since conventional domain based approaches use only domain to identify proteins, two different proteins cannot be distinguished if they share same domains. Moreover, proteins which have no domain information cannot be used in the domain approaches, so not all interaction pairs are available. In this paper, two different ideas are proposed; first one is an increasing domains by adjusting e-value of InterProScan which extracts domains from a raw sequence, and the other is a utilizing gene ontology (GO) which describes structural and functional information of genes. In Yeast proteins, newly generated domains cover 49.84% more by adjusting e-values and 77.06 % more by utilizing GO. In addition, an average number of proteins which share same patterns are reduced to 1.27 with e-value adjusting, 1.55 with utilizing GO. To reduce an average number of GO for one protein, an abstraction rule for GO acyclic graph is also proposed in this paper. Taking all results we have done into consideration, we found that both approaches improve determinability and utilization of proteins, and they can be used complementary