Bayesian Analysis for Studies of Gene-Environment Interaction

摘要在基因-环境与疾病相关的病例对照研究中，当遗传和环境暴露可以假设在基础人群中是独立的时，人们可以利用独立性假设来推导出比传统的Logistic回归分析更有效的估计方法。病例对照分析的许多经典结果假定协变量分布为非参数分布，但在暴露分布的约束空间下不成立。然而，现代回溯方法效率的提高是以缺乏稳健性为代价的，因为在违反基因-环境独立性假设的情况下，回溯估计中会引入较大的偏差。本研究建议的主要目的是在一些常用的流行病学设计下，寻找自然的分析工具来解决现代基因-环境相互作用研究的回溯性分析的模型规范困境。利用Chatterjee和Carroll(2005)开发的轮廓-似然框架，研究人员提出了一种贝叶斯方法，该方法以自然数据自适应的方式结合了关于假设的基因-环境独立性约束的不确定性。所提出的收缩估计器是从贝叶斯的角度构思的，旨在保持有吸引力的效率性质，而不依赖于不可验证的模型约束。在基因-环境关联的不同场景下，研究了所提出的估计器的理论性质。研究者同时考虑了经验贝叶斯方法和层次贝叶斯方法来放宽基因-环境独立性假设。该提案探讨了贝叶斯方法与另一种随机效果模型的联系。这些方法从通常使用的不匹配的病例对照研究设计扩展到基因-环境相互作用的两阶段和基于家庭的研究。目前有两个科学流派在临床医学和公共卫生中发挥着极其重要的作用：侧重于遗传学的分子生物学方法和侧重于流行病学的定量方法。这些领域的发展共同为复杂疾病的病因、诊断、预后和治疗的研究做出了基础性贡献。医学和基因技术的突飞猛进带来了许多统计学家和流行病学家以前从未遇到过的复杂的设计和分析问题。这一建议在于人类遗传学、流行病学和统计学的新接口。病例对照研究越来越多地被用于研究疾病和候选基因之间的联系。然而，除了一些罕见的疾病，如亨廷顿病或泰萨克斯病，可能是单一基因产物缺乏的结果，大多数常见的人类疾病都有一个多因素的病因，涉及许多遗传和环境因素的复杂相互作用。通过临床和流行病学研究识别和表征这种复杂的基因-环境相互作用，人们有更多的机会了解复杂疾病的起源和病因，并为高危个体制定有针对性的干预策略。该提案提出了稳健和有效的统计技术，以研究复杂疾病中基因和环境之间的协同作用。提案中开发的高性能计算工具使该方法在大规模应用中使用是可行的，例如全基因组关联研究。

ABSTRACT In case-control studies of gene-environment association with disease, when genetic and environmental exposures can be assumed to be independent in the underlying population, one may exploit theindependence assumption in order to derive more efficient estimation techniques than the traditional logistic regression analysis. Many of the classical results for case-control analysis, which assume the covariate distribution to be non-parametric, do not hold under a constrained space of exposure distributions. However, the gain in efficiency of modern retrospective methods comes at the cost of lack of robustness, since large biases are introduced in the retrospective estimates under violation of the gene-environment independence assumption. The main goal of this research proposal is to find natural analytical tools to solve the model specification dilemma of modern retrospective analysis of studies of gene-environment interaction, under some commonly used epidemiological designs. Using the profile-likelihood framework developed by Chatterjee and Carroll (2005, Biometrika), the investigator proposes a Bayesian approach that incorporates uncertainty regarding the assumed constraint of gene-environment independence in a natural data adaptive way. The proposed shrinkage estimator, conceived from a Bayesian standpoint, is designed to maintain attractive efficiency properties, without relying on unverifiable model constraints. Theoretical properties of the proposed estimator are studied under varying scenarios of gene-environment association. The investigator considers both empirical Bayes and hierarchical Bayes methods to relax gene-environment independence assumption. The proposal explores the connection of the Bayesian approaches to an alternative random-effects model. The methods are extended beyond the commonly used unmatched case-control study design to two-phase and family-based studies of gene-environment interaction.Two scientific streams are currently playing extremely important roles in clinical medicine and public health: the molecular biology approach with an emphasis on genetics, and the quantitative approach with an emphasis on epidemiology. The developments in these areas jointly are making fundamental contributions to the study of etiology, diagnosis, prognosis and treatment of complex diseases. Phenomenal advancement of medical science and genetic technology is giving rise to many complex design and analysis issues which statisticians and epidemiologists have never confronted before. This proposal lies in that new interface of human genetics, epidemiology and statistics. Case-control studies are being increasingly used for studying theassociation between a disease and a candidate gene. However, except for some rare diseases, such as Huntington or Tay Sachs disease which may be the result of a deficiency of a single gene product,most common human diseases have a multifactorial etiology involving complex interplay of many genetic and environmental factors. By identifying and characterizing such complicated gene-environment interactions through clinical and epidemiological studies, one has more opportunities to understand the genesis and etiology of complex diseases and to develop targeted intervention strategies for high-risk individuals. The proposal presents robust and efficient statistical techniques to investigate the synergism between gene and environment in studying complex diseases. The high-performance computing tools developed in the proposal makes it feasible to use the methods in large-scale applications such as genome-wide association studies.