| Microarrays have been widely used to study differential gene expression at the genomic level. They can also provide genome-wide co-expression information. Robust approaches are needed for integration and validation of independently-collected datasets which may contribute to a common hypothesis. Previously, attempts at meta-analysis have contributed to solutions to this problem. As an alternative, for microarray data from multiple highly similar biological experimental designs, a more direct combined approach is possible. In this thesis, a novel approach is described for microarray combined analysis, including gene-level unification into a virtual platform followed by normalization and a method for ranking candidate genes based on co-expression information---called Coex-Rank. We applied this approach to our Sppar (a PPARgamma mutant) dataset, which illustrated an improvement in statistical power and a complementary advantage of the Coex-Rank method from a biological perspective.;We also performed analysis to other PPARgamma-related microarray datasets. From the perspective of gene sets, we observed that up-regulated genes from mice treated with the PPARgamma ligand rosiglitazone were significantly down-regulated in mice with a global knock-in dominant-negative mutation of PPARgamma. Integrated with publicly available PPRE (PPAR Response Element) datasets, we found that the genes which were most upregulated by rosiglitazone treatment and which were also down-regulated by the global knock-in mutation of PPARgamma were robustly enriched in PPREs near transcription start sites. In addition, we identified several potential PPARgamma targets in the aorta and mesenteric artery for further experimental validation, such as Rhobtb1 and Rgs5. |
