| Summary: | 碩士 === 國立高雄應用科技大學 === 資訊工程系 === 106 === Biological networks carry information that may help to identify critical genes associated with underlying phenotypes. In such biological networks, each gene is represented as a node, and links between nodes indicate a concordant gene co-expression pattern. Many methods for predicting disease-related genes often ignore the dynamic changes of the edges in the networks, thus some important information is absent. We propose two methods H_ij^n (to detect pairs of genes with the highest edge changes) and H ̂_ij (to detect pairs of genes with the lowest entropy by edge change rate) to identify genes associated with complex disease. The test dataset was expression profiles from bladder cancer research download from NCBI GEO database(GDS4456). There were five stages of gene expression data (pTa, pT1, pT2, pT3 and pT4) . After constructing five gene co-expression networks, we identified top-ranked gene pairs by applying H_ij^n and H ̂_ij. Finally we compared these genes with COSMIC and found several known mutated genes: MPHOSPH9、TNRC6A、ZNF236、CDH10、SV2B、CR2 and RELN. In the future, we plan to improve H_ij^n and H ̂_ij. First, H_ij^n will ignore gene pairs of same co-expression pattern, thus missing important information. Second, the rate of association change by H ̂_ij depends on the number of graphs. In small number of graphs, many of the gene pairs will have same entropy, therefore our scoring function will not be able to distinguished these pairs.
|
|---|
