Research On Algorithm Of Identifying Protein Complexes Based On Gene Ontology In Protein-protein Interaction Network

Mining protein complexes in protein-protein interaction networks, which canpredict unknown protein interactions, explain specific biological process and revealthe function of protein, understand protein interaction network evolution and theresearch on complex disease is of great significance. With the development ofhigh-throughput experimental technologies, there are a lot of large-scale proteininteraction network data, which makes it possible for us to identify protein complex inprotein-protein interaction networks. However, most of the existing PPI networkswhich have a lot of noise data and are sparse, how to study a highly efficientalgorithm of identification protein complexes still remains challenging.From the biological function of the protein itself, gene ontology can be used toevaluate the reliability of the interaction between proteins for protein interaction datawith high false positive and false negative. Therefore, in this paper, we put forward anew method called MCGO, which is based on gene ontology to identify proteincomplex in protein networks. Using three yeast protein data sets (DIP, Gavin, Krogan),the algorithm is compared with some classical algorithms to analyze its performance.The experimental results show that the algorithm can identify more complexes withbiological significance, from the perspective of Precision, F-measure, Coverage rateon the indicators of performance is superior to other algorithms, especially on theDIP.The algorithm just considers functional information of proteins, which has somelimitations, ignoring the topological characteristics of the protein-protein network.Therefore, a new method called MGOTC, based on gene ontology and topologicalcharacteristics to extract protein complexes is proposed in this paper. Using threeyeast protein data sets and two standard protein complexes, the algorithm is comparedwith some classical algorithms to analyze its performance. Experimental results showthat the algorithm can discover more protein complexes of the standard proteincomplexes, especially significantly better than the other algorithms in terms of recalland coverage rate.