Research And Application On Protein Complexes Identification Based On Semantic Similarity

In recent years,with the continuous development of biology technology,it promote the succession of human genome project.Meanwhile,the systematic biology and proteomics research are dose to perfection.Therefore,how to study the quality complex and its function has become a hot research topic based on the structure and biological characteristics of the known protein interaction network.The identification of protein complexes is the method used to solve this problem by which we can study the nutritious protein complexes and predict the function of unknown proteins.The basic research methods of protein complex recognition are introduced in detail,including the method based on graph division,the hierarchical approach and the method based on biological information fusion.On this basis,this paper presents two new recognition algorithms based on the structure of the protein interaction network and the characteristics of the protein complexes.(1)A clustering algorithm based on semantic similarity is proposed.Because the recognition algorithm of most complexes is acting on the non-protein network,the biological characteristics inherent in the protein are not taken into account,and the accuracy of the recognition is greatly affected.Therefore,the author proposes a clustering algorithm based on semantic similarity,the DSC algorithm,which first constructs a protein-weighted network based on semantic similarity,and then identifies the protein complex based on the edge aggregation factor.Experiments show that the algorithm achieves good results.(2)A protein complex recognition algorithm based on hierarchical expansion of key nodes is proposed.According to the traditional method focus on the whole network structure and neglect the topology of complexes.In this paper,we use the key node selection and multi-level expansion to identify the protein complex.We propose a method based on the hierarchical expansion of the key nodes---KNHE,and apply it in the protein-weighted network.Because this algorithm fully consider the importance of known key protein and the structure of the complex,the experimental results show that the algorithm has great improvement in sensitivity and specificity,and the experiment has achieved good results.The clustering algorithms proposed in this paper start off from different sights and solve the problem of low identification recognition rate.The proposed algorithms have good clustering performance,it can predict unknown protein functions.