Research On Topological Reconstruction And Optimization Algorithm Of Gene Regulatory Network Based On Information Theory

Gene regulatory network plays an important role in cellular activities,and the accurate construction of gene regulatory network is helpful to understand the pathogenic mechanism and cure mechanism of disease at the molecular level.With the increase of gene expression data,mining gene regulatory network from huge gene data has become an important problem currently faced.Although a large number of methods have been proposed to construct gene regulatory network,the network structure they construct often contain a large number of false positive edges because of the scarce gene expression samples and the sparse network structure.It is the current primary goal to seek a gene regulatory network topology reconstruction and optimization algorithm that can reduce false positives and improve true positives.On the basis of information theory,this paper proposes some network structure construction algorithms to optimize the topology of gene regulatory network,aiming at the shortcomings of current gene regulatory network construction methods.The work of this paper mainly focuses on the following two parts:(1)Aiming at the low network construction accuracy caused by the inability of existing network construction methods to effectively adapt to the structural characteristic of "global sparse,local tight",a gene regulatory network construction method based on an improved random walk algorithm with restart is proposed to realize the unified inference of global structure and local structure of network.In this method,on the basis of assumption of gene function modularity,we improve the calculation of restart probability,initial probability vector and transition probability matrix,and realize the combination of global topology and local topology through the local module control of global random walk algorithm.Then,a gene regulatory network structure with “global sparse,local tight” is inferenced.Finally,the isolated genes in the inferred result are processed by the Markov blanket discovery algorithm to obtain an optimized network structure.The method was tested on the dataset of DREAM3 yeast network,the chain chemical reaction network and the Escherichia coli SOS DNA repair system network.The results show the advantages of our method in the structure construction of gene regulatory network,especially in the reduction of false positive rate.(2)Aiming at the problem that some methods blindly pursue reducing the false positive rate of network construction,which leads to the sacrifice of true positive rate,a gene regulation network construction method based on network structure control strategy is proposed.On the basis of the overall upstream and downstream gene regulation levels of the network and the local-scale network motif structure,the method refines the specific forms and cooperation patterns of the network global and local topology.Specifically,genes are first globally ranked according to network topological centrality to determine the global topological features(such as sparsity and hierarchy)to determine intergenic regulation from a broader scale.Then,a common association patterns(i.e.,network motifs)in complex networks are used for the inference of gene regulatory relationships to determine the local topology of the network(i.e.,local regulatory tightness and functional modularity).The method is compared with method(1)and other classical methods on the same dataset,and the comparison shows that the optimization effect of this method on network topology is further than method(1),and the true positive rate of network construction is significantly improved.