New analysis methods for microarray data

UnrestrictedMicroarray technology allows the simultaneous measurement of RNA levels in a cell culture or tissue sample. This has resulted in the generation of vast amounts of data, and the methods to analyze this onslaught are only just now catching up. In this thesis, we present three novel methods for the algorithmic and statistical analysis of microarray data.; First, in joint work with Michael R. Mehan, we systematically annotated over 300 microarray datasets with phenotype information, then used these annotations to find gene coexpression modules that are specific to particular phenotypes. Though previous methods existed to find recurrent coexpression modules, ours is the first to find phenotype-specific ones, avoiding the problem past methods faced of finding mostly general, cell-cycle related modules. Second, we studied the statistics of gene networks. Although the algorithms to study gene coexpression networks have become increasingly sophisticated, the statistics have failed to keep up. As a result, it is currently impossible to estimate the statistical significance of coexpression modules, or other network features. Our statistical investigations did not solve that problem, but they contribute to that end by demonstrating the deficiencies of past models and suggesting further avenues of research. Finally, we developed a new method to jointly study microRNA and messenger RNA (mRNA) microarray data. MicroRNAs are a novel class of RNA that regulates the translation and turnover of target mRNAs. Using a non-parametric permutation approach, we were able to show that levels of miRNAs are positively correlated with their targets in brain tissue. Thanks to the non-parametric approach, we can be confident that this is true independent of any potential biases in the sample data. Additionally, we found that specific miRNA-target regulatory pairs are altered in Alzheimer’s disease brain.; In all, the work presented in this thesis advances the field of microarray analysis, and in particular demonstrates how data from multiple arrays, multiple genes, and multiple datasets can be aggregated to enhance potentially subtle signals in microarray data