Research On Gene Regulatory Network Reconstruction Algorithms Based On Information Theory

With the continuous advancement of biological sequencing technology,microarray technologies have produced tremendous amounts of gene expression data,which provide opportunity for understanding the underlying regulatory mechanism.The in-depth and careful studies and analysis on gene regulatory networks will help to reveal the pathogenesis of various diseases and understanding of the internal fundamentals of life.Therefore,the construction of GRNs has already become one of the research hotspots and keystones in the field of bioinformatics.Recently,numerous approaches have been proposed to infer GRNs from gene expression data.Information theory,which is known for its calculation simplicity and effectiveness in capturing the nonlinear correlation between genes,has become a powerful tool in the construction of GRNs and has been increasingly applied in GRNs.In this work,methods based on information theory have been developed to infer GRNs from static and time series gene expression data with the motivation of improving the accuracy of network inference.The specific work includes the following aspect:In allusion to combinational regulation existed in biological networks,this article proposed an algorithm based on information theoretic criteria to infer combinatorial transcriptional dynamics by using time series gene expression data.The schematic of the algorithm includes three steps.First,test time delay between genes pairs and calculate the mutual information matrix.Second,use each MI value as a threshold to determine a model,the one with the minimum description length is chosen as the optimal model after obtaining all the models.And third,compact the optimal model.Unlike the conventional method in using information theory,this algorithm incorporates CMI and DPI to handle the influence of the migration and the dispersion effects,and also stores the conditional mutual information to prevent deleting the combinational regulation by mistake.The mistake is caused by the lower the conditional mutual information value.Results from the synthesized and real biological networks verify that the proposed algorithm can effectively reveal the combinational regulation and can infer more edges with biological significance.In allusion to disadvantage of unable to describe the regulatory dynamics by using static gene expression data.An algorithm combining nonlinear differential equation and novel PCA-CMI is presented to improve the accuracy of GRN inference,and therefore has the advantages of both differential equation and information theory methods.The schematic of the algorithm includes two stages,namely,causal model selection stage and optimization of process stage.In the first stage,the improved path consistency algorithm is used.This algorithm take the advantages of high efficiency of the path consistency algorithm,and overcome the error which occurs at the calculation of high order conditional mutual information,it can also conquer the differences caused by different input sequence.In the second phase,the direction of regulations is determined by differential equation model.With this strategy,the searching space of the differential equation model is reduced,and the regulation direction between genes pairs by using information theoretic criteria can be determined.The results on benchmark datasets from DREAM challenge confirmed the effectiveness of this method...