| Identifying causal susceptibility alleles for asthma poses many challenges. These include the multigenic nature of the disease, the lack of reliable assessment of individual exposures, and complex interactions with environmental factors. The completion of the Human Genome Project has greatly accelerated the technology development for linkage analysis, candidate gene association studies, and, more recently, genome-wide association studies (GWAS). At the same time, high-throughput technologies for profiling gene expression, DNA methylation, and protein abundances have advanced our understanding of the molecular basis of disease etiology, disease heterogeneity and classification. Accordingly, the best statistical method to use for elucidating the genetic basis of a complex disease depends on the questions of interest, the design of the study, assumptions made about genetic models (i.e. recessive, additive, dominant), how a disease is defined, and the type of genetic markers under investigation. Conventional analyses of gene-environment (G x E) interaction require much larger sample size than for studying main effects. This can lead to false discoveries or false negatives due to lack of power. Hence, there is a need to develop statistical methods aimed at detecting interactions between factors. The observational epidemiology Southern California Children's Health Study (CHS) and the experimental UCLA Challenge Study together offer a unique opportunity to investigate the interactions between genetic variation and exposure to particulates on the risk of asthma through joint modeling of effects attributable to differential responses of asthma-related immune phenotypes.;In this dissertation, I first review the asthma-associated risk factors, the background of the CHS and UCLA Challenge Study, and my research goals. The second section summarizes currently used statistical approaches to the discovery of genetic determinants for complex diseases including conventional regression models, exploratory data mining techniques, biology-driven methods, and prior hypothesis-driven approaches. Next, I propose two distinct Bayesian hierarchical modeling approaches to inform the analysis of one study with information derived from another through joint analysis. This is illustrated using data from the CHS and the UCLA Challenge Study for estimating G x E effects. I describe the simulation design to evaluate the performance of my proposed method, and then illustrate the application for genetic association studies using real candidate gene data from the two studies. Finally, I conclude with a discussion of implications and future challenges for this statistical methodology framework. |
