Statistical methods for vQTL mapping and Mendelian randomization analysis with a time-varying exposure
Complex diseases are affected by genetic factors, environmental factors, and their interactions. Traditional genetic studies for complex diseases focus on identifying loci associated with mean heterogeneity of a phenotype. A new class of genetic loci that are associated with phenotype variance heterogeneity (vQTL) has been suggested for identifying gene-gene and gene-environment interactions. While several tests have been proposed to detect vQTL for unrelated individuals, there are no tests for related individuals, commonly seen in family-based genetic studies. The first part of this dissertation introduces a likelihood ratio test (LRT) for vQTL identification using a linear mixed model framework, adjusting for covariates and family relatedness. The LRT test statistic approximately follows chi-squared distributions for normally distributed quantitative traits. A parametric bootstrap based LRT was proposed for non-normally distributed quantitative traits. Simulation studies show that the family-based test controls Type I error and has good power. We demonstrate the utility and efficiency gains of the proposed method using the Framingham Heart Study (FHS) data to detect loci associated with body mass index (BMI) variability. The second part of this dissertation introduces Mendelian randomization (MR) analysis of a time-varying exposure using functional data analysis techniques. MR analysis is a method to analyze the causal effect of an environmental exposure variable on an outcome variable from observational studies by using genetic variants as instrumental variables. Many exposures of interest are time-varying, for example, BMI. However, current MR studies only use a single measurement of a time-varying exposure variable given that longitudinal measurements have been collected in many cohort studies. One measurement cannot adequately capture information of a time-varying exposure variable. Thus, we propose to use the functional principal component analysis method to recover the underlying individual trajectories of the time-varying exposure variable from the sparsely and irregularly observed longitudinal data, and then conduct MR analysis using the recovered trajectories. We focused on statistical testing for a causal effect. Different MR analysis methods have been proposed for continuous outcome variables and binary outcome variables to analyze the recovered functional exposure data. Simulation studies show that the functional data analysis-based methods incorporating longitudinal data have substantial power gain as compared with standard MR analysis. We used the FHS data to demonstrate the promising performance of the new methods.
Cao, Ying, "Statistical methods for vQTL mapping and Mendelian randomization analysis with a time-varying exposure" (2015). Texas Medical Center Dissertations (via ProQuest). AAI3731983.