Statistical methods for the analysis of expression quantitative traits

Advances in high throughput molecular profiling technologies to quantify gene expressions and identify genetic variations have ushered in a new era of biology. For the first time, we have the ability to globally survey naturally occurring genetic perturbations to identify those variants that alter gene expressions. Yet to fully realize the potential of the new biology these datasets promise, one must go beyond the standard expression quantitative trait (eQTL) mapping study in order to understand the relationships between gene expressions and genetic variations in the contexts of biological pathways, tissue types and environments. This thesis introduces novel statistical methods that extend the standard eQTL analysis to identify eQTLs in a biological context that are both statistically robust and biologically relevant.;First, I address the problem of discovering eQTLs across multiple tissues using a procedure that captures similarities between tissues. I introduce both a general statistical framework for performing multiple tests in multiple dimensions and provide a few strategies for deriving a practical procedure that improves the over all power to detect cross tissue and tissue specific eQTLs. I show in simulation and replication studies that the performance of our method is better than standard likelihood ratio based approaches. I also show across four tissues the ability of this method to identify functionally consistent cross tissue and tissue specific eQTLs.;Second, I present an application of mixed models to adjust for unmodeled heterogeneous sample structure resulting from experimental and technical biases in eQTL analysis. After showing that many previous eQTL studies suffer from spurious linkage and association signals that do not duplicate between replicates, I show that our approach performs significantly better by recovering eQTLs that are more concordant between biological replicates and increasing the number of cis eQTLs identified.;Third, I present an approach for integrating eQTL analysis with chIP-Chip data to understand the effects of genetic variations on the dynamics of transcription regulation. I developed new statistical tests based on a network component analysis framework where the possible transcription regulatory networks are constrained to known relationships between transcription factors and their targets inferred from chIP-Chip data. Using yeast expression and genotyping data, I show that global trans linkage patterns can often be explained by regulatory eQTLs perturbing these transcription regulatory networks to induce large-scale differential expression.;Fourth, I present a novel multivariate-based approach for interpreting differential expression and association studies in the context of annotated gene sets. I devised a new aggregate statistic that introduces the notion of a "tightly regulated" gene set that complements the notion of a differentially expressed gene set. This is motivated by the idea that sets of genes whose expression levels are coordinately differentially expressed with respect to a genetic variation or an experimental condition are more suggestive of real biological pathways. I identified several interesting gene sets in the context of an eQTL study conducted in murine hematopoietic stem cells.;Finally, I present a novel model selection method for identifying the presence and absence of causal relationships between genes leveraging expression and variation data. We applied our method to identify "causal regulators" in a yeast eQTL study where global linkage patterns have been observed and attributed to "master regulators" that control the gene expression of a large number of genes. In addition to confirming several previously suggested regulators, we also identified several novel ones.