Key Genes Selection Based On Mutual Information Network

In the post-genome era，the emphasis of life science is no longer on a single gene,but in the structure and function of the genome level to study the running mechanismof biological systems. This thesis screened out some differentially expressed genesusing microarray gene expression profile, then built mutual information network andanalyzed the structure of the network. The strength, betweenness and fault influenceof the nodes in the weighted network and their difference between the normal andcancer networks were analyzed. The corresponding gene rankings were furtherobtained, then aggregated these rankings into a final gene ranking by severalalgorithms of social choice model. Main results are as follows:First of all, two groups of breast cancer data, one group is paired and the otherone is unpaired, were downloaded form GEO database. After data processing, about300differentially expressed genes were screened out by SAM, and normal and cancermutual information networks were built based on these differentially expressed genes.After analyzed6parameters(average degree, average core,etc.) of the network, it isshow that the structure of normal and cancer network were significantly different forboth two groups of data, and particularly, the difference of the unpaired data weremore significantly. Structure determines function, structure difference is the directreason of function differences.Compared strength, betweenness and fault influence of the nodes in normal andcancer weighted mutual information network of the unpaired data,3correspondinggene rankings were obtained. Aggregated these rankings by Borda and Max-Diffalgorithm, and the results of these two algorithms were almost the same, they bothobtained8key genes from front15genes of their final ranking. So, Borda andMax-Diff algorithm are both good algorithm for the selection of key genes. Inaddition, another famous algorithm-Footrule algorithm focus too much on the wholeranking, ignoring the individual specific genes, so Footrule algorithm is unsuitable forthe selection of key genes.Analysis of network structure and selection of key gene can help us to find thecause of disease and disease genes for gene-related diseases, are beneficial to thediagnosis and treatment of disease.