Identification of Protein Complex Based on Multi-label Learning Algorithm

Abstract

Protein complexes are key molecular entities that play an integral role in human life activities. Systematic identification of protein complexes is an important application of data mining in the biological sciences. Existing multi-label learning algorithms can effectively label nodes belonging to different complexes in protein-protein interaction network to identify overlapping complexes. However, the protein complexes formed by the stochastic strategy may have unstable results and insufficient community quality. To solve these problems, this paper proposes a novel protein complex identification method based on multisource fused data and the multi-label learning algorithm. The descending order of the potential influence of the nodes is used as the node selection order to solve the problem of unstable partitioning of the composite results. The comprehensive similarity obtained by the link correlation and the similarity of the gene annotations is used as the node label update strategy to improve the quality of the composite. The experimental results show that the new proposed method is much more effective and feasible, and has higher precision and biological significance.