Study On Algorithms For Reconstruction Of Gene Regulatory Networks

With the development of biological and genetic technology,the post genome era is coming after the completion of genome sequencing.More and more studies have focused on gene structure and biological function in addition to gene sequences research.It can provide more valuable knowledge of life science and theoretical basis for revealing the mysteries of life by digging deeper into the gene expression data.The goal of the construction of gene regulatory network is to infer the potential relationship between genes from gene expression data.In this thesis,two improved algorithms for reconstruction of gene regulatory network are proposed after analysis of the advantages and disadvantages of some existing algorithms.The experimental results show that the improved algorithms have better performances.The following specific research work was completed:We proposed a Bayesian network algorithm MI_K2SA based on K2 and simulated annealing algorithms in order to solve the problem that K2 algorithm is difficult to determine the initial node order and easily fall into the local optimal solution.Mutual information was used to measure the association between two nodes in the algorithm MI_K2SA.The initial node order obtained by maximum weight spanning tree was used to the initial input of the algorithm.In the process of searching optimal network,the simulated annealing algorithm was introduced in order to jump out of the local optimal network to search the global optimal network.The experimental results showed that MI_K2SA algorithm had better performance on obtaining the initial node order,jumping out of local optimal solution and reconstructing the gene regulatory network with higher accuracyFurthermore,a novel algorithm MICRAT based on maximal information coefficient was proposed.It addressed the problem of small sample size with large number of genes and not making full use of the timing series gene data in reconstructing gene regulatory network.The maximal information coefficient was used to measure the association between two genes on even a small data sets with high dimensions and small sample size.The algorithm MICRAT first employed maximal information coefficient to measure the degree of correlation between the two genes for reconstructing an undirected gene regulatory graph,followed by removing redundant edges avoiding cycles based on conditional mutual information.Finally both time series mutual and conditional relative average entropy(CRAE)were used to orient the edges in the undirected graph.Experimental results showed that MICRAT algorithm has better accuracy in comparison to other relative methods,and it also has stable performance even the datasets with different size.