Research On Algorithms For Detecting Functional Module In Protein-Protein Interaction Network

With the completion of human genome sequencing,in the postgenome era,proteomics has become the main research field in life sciences.The basis of life activities is constituted by a huge amount of interactions between proteins.However,the specific biological function is not exerted by a single protein.Generally,multiple interacted densely proteins form a molecule complex to perform biological function in a cell.In recent years,a large number of protein interaction data has been produced with the rapid development of high-throughput technology,and protein protein interaction networks are constructed by these interacted data.The interacted proteins are generally considered to have the same or similar biological functions then those noninteracted.Therefore,it has been attracted a plenty of researchers to study how to develop novel computational approaches which can discover protein modules with remarkable biological functions from protein-protein interaction networks.Since the number of protein-protein interactions have been discovered is relatively small so far and there are some noise interactions(false positives)in these few interactions.Then it is a hard work to identify protein functional modules directly from these sparse and noisy protein-protein interaction networks.Therefore,the protein functional modules detected by protein functional module detection algorithms which are completely based on the network topology structures may be unsatisfactory.However,it is fortunate that some human curated protein complexes with high-quality have been produced by human.Therefore,it is necessary to study and design some protein functional module detection algorithms that can consider protein-protein interaction data and protein complex data simultaneously.At present,most functional module detection algorithms are unsupervised,then the main content of this dissertation is to analyze the relationship between protein topology modules and functional modules,and to design novel semi-supervised protein functional module detection algorithms that use protein complex data as priori information.The main contribution of this paper is described as follows:(1)The relationship between protein functional modules and topological modules is not yet clear,in this manuscript,we systematically analyzes the relationship between protein topology modules and functional modules.We first utilize five non-overlapping module detection algorithms and two overlapping module detection algorithms to detect protein topological modules from human related protein-protein interaction networks.Then the physical properties of these topological modules are analyzed by us.Finally,the biological function analysis of the protein topological modules are carried out in four aspects:gene ontology enrichment analysis,gene ontology homogeneity,pathway homogeneity,symptom similarities between protein modules.The experimental results show that the protein topological modules have a diversity of biological functions.Therefore,the other biological information should be considered when detecting protein modules with high homogeneity.(2)Aiming at the problem that human protein-protein interaction data is less and some noise interactions also existed,a pairwise constrained protein functional module detection algorithm based on Non-negative Matrix Tri-Factorization(PCNMTF)is proposed in this paper.The proposed PCNMTF algorithm extracts prior information from currently known reliable protein complexes,and then use these prior information to guide the learning process of the protein module membership matrix.At the same time,a method for detecting overlapping protein functional modules utilizing the relationship between protein modules was also designed in this manuscript.The experimental results confirm that using the protein complex data as prior information can improve the detection accuracy of protein functional modules.(3)The prior information is generally only used to constrain the protein module membership matrix and is rarely used to constrain the relationship matrix between protein modules.We propose a novel semi-supervised protein functional module detection algorithm based on NMTF(SSNMTF).The proposed algorithm can use prior information to guide the learning process of the protein module membership matrix and the relationship matrix between modules simultaneously.At the same time,we propose a parameter-free overlapping protein functional module detection algorithm based on protein module membership matrix.Experimental results show that the use of prior information to guide the learning processes of protein module membership matrix and module relationship matrix simultaneously can improve the efficiency of prior information.(4)The prior information of must-link is mainly constrained by the graph regular-ization but ignores positions of the two corresponding proteins in module.Then a novel semi-supervised protein functional module detection algorithm based on vector inner product similarity(Semi-Supervised protein functional module detection algorithm based on NMTF with Inner Product,NMTFIP)is proposed in this paper.The mustlink constraint is used to minimize the distance between the module membership vectors of two corresponding proteins in PCNMTF and SSNMTF rather than considering the fact that the importance of protein in module,this may lead to inaccurate module detection result.NMTFIP constrains the must-link information by maximizing the similarity between the two protein module membership vectors,and this method can use prior information more reasonably to guide the protein module detection process to obtain more accurate protein functional modules.