| Summary: | 碩士 === 國立陽明大學 === 生物資訊研究所 === 95 === In the past decade, DNA microarray technologies have been extensively used to detect cancer differentially expressed genes and coexpressed genes. They found that cancer will lead to a cluster of genes expressed differently from normal cells, and perturb regular metabolism or signal transduction pathway. But currently, statistic analyses are just devoted to compute significance of differential expression of single gene, add some interaction network or pathway information, “see” the correlation of them, and then discover cancer related “candidate genes”. In my study, I have tried to discover “candidate networks” directly by statistic and computational analysis. This method is called “Automatically Search Candidate Network” (AutoSCAN). It needs two kinds of information: the situation of network and node weight. By using Monte Carlo
Test to simulate the random distribution of node weight in the network and computing the significance of the target subnetwork, this method determines if the subnetwork is
differentially expressed in entire network between cancer and normal samples. AutoSCAN uses all genes to compute the significance of perturbed pathways or network to eliminate the loss of any important information about insignificant genes in microarray. The results show that I used the “first level interaction complex analysis” of AutoSCAN to find some lung cancer related genes that T-test can not find, and use the “candidate networks analysis” to successfully discover perturbed pathways in lung cancer.
|
|---|
