Condition-dependent coexpression in Arabidopsis and the bootstrap estimation of false discovery rate
Year: 2015
Type: dissertation
Abstract: Conventional coexpression analysis looks at gene expression levels over diverse conditions. Although such ``condition independent'' analyses provide useful generalized information, the coexpression of genes can depend on biological context. Looking at gene coexpression in a condition-dependent fashion identifies gene interactions relevant to more-specific biological questions. This thesis addresses condition-dependent coexpression in the context of seed germination in Arabidopsis thaliana. We study two broad aspects of the problem: (i) detecting pairwise interactions, and (ii) detecting three-way interactions. First, using gene expression data gathered exclusively from mature seeds, we develop a coexpression network, SeedNet. We analyze its statistical properties and find, for example, that SeedNet bifurcates into two previously undescribed components, one corresponding to germination and one to non-germination. Second, we develop a detector for three-way interactions in gene expression data and demonstrate its efficacy. In this approach, the coexpression of a pair of genes can depend on the expression level of a third gene. That is, coexpression is conditioned on the transcriptional environment. Because of the vast number of possible three-way interactions, there is an extremely large potential for false positives, so it is crucial to accurately estimate the False Discovery Rate (FDR). We show that extending approaches used for two-way interactions can significantly underestimate the FDR of three-way interactions. The thesis develops and evaluates a new approach for estimating FDR based on the bootstrap. We show that the bootstrap approach produces reasonable estimates over a wide range of statistical conditions, whereas conventional approaches rapidly break down. Finally, we use our detector and FDR estimates to add over 64,000 new, highly-confident edges to SeedNet.
Source: TSpace (University of Toronto)
Author: Hui Lan
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: green
Funders: Strong, Hospital for Sick Children
OpenAlex ID: W2278563175