Computational analysis of genome-wide RNA expression and its regulation by microRNAs.

摘要

Eukaryotic complexity relies on correspondingly elaborate modes of gene regulation, much of it orchestrated by non-protein coding RNAs. However, we still have a very limited understanding of the non-protein coding portions of the genome, the RNA transcripts arising from it, and their role in gene regulation. This thesis applies computational approaches towards answering questions related to these topics.;To explore transcription from the unannotated "dark matter" of the genome, we developed a novel algorithm for analyzing the results of genome-wide tiling array experiments. The method explicitly models non-specific hybridization and combines replicate experiments in a way that makes minimal assumptions about the data. Applying this tool to genome-wide microarray data measuring RNA extracted from a developmental time course of D. melanogaster, and from adult mosquitoes (Anopheles gambiae ) yields the surprising finding that a large fraction of these genomes are transcribed.;We also apply computational methods towards studying the mechanism by which microRNAs recognize their target transcripts. Our results confirm that microRNAs exert an effect on mRNA stability that is detectable directly from microarray studies. We confirm also the primacy of the seed sequence in accounting for miRNA-mediated transcriptional down-regulation, but note that there is considerable heterogeneity among miRNAs in the degree of tolerance for mismatches within the seed, and in the extent to which different seed types are utilized. In addition, we find motifs pairing to the downstream portion of the microRNA that affect the efficacy of targeting. Last, we find sequences unrelated to the transfected constructs, including an AU-rich element and other microRNA seeds.;Interpretation of the array data is aided by our observation that sequence features of the human 3' untranslated region (UTR) are highly correlated with signal intensities. Correcting for these artifactual correlations removes a potential source of confounding for any array-based method that studies the effects of 3' UTR sequence on gene expression.