A Ranksum Statistics Based Framework to Decipher Transcription Regulation

The unbiased generation of specific and meaningful hypotheses from the deluge of data generated by modern genomic methods remains a challenge. These datasets require increasing level of expertise to analyze fully, and are often underutilized even in the originating lab. It would be desirable to have a computational strategy that is easy to implement, robust against outliers and missing data, and broadly applicable to diverse experimental designs. In this dissertation, I present a set of ranksum statistics-based analytical methods as a framework to extract testable hypotheses from large and complex datasets. To illustrate its utility, this framework was applied to two clinically important biological questions. In both instances, my method yielded novel molecular mechanisms that were subsequently validated with both in vitro and in vivo experiments. In the first study, gene expression profiles from multiple mouse models of cardiac hypertrophy were analyzed to reveal a novel interaction between transcription factors Nkx2-5 and Egr1, providing mechanistic insight into how Nkx2-5 haploinsufficiency leads to exacerbated cardiac hypertrophy and poor survival in these mice. In the second study, thousands of microarray samples acquired from public data repositories were analyzed to quantitatively define tissue-specific expression pattern for every gene represented on a microarray platform. The tissue-specific expression data was then used to identify novel transcriptional regulators of brown fat gene expression program in adipocytes. The successful application of the analytical framework in these examples, regardless of their differing experimental design, highlights its adaptability in facilitating discoveries in a wide array of biological problems