Efficient Design and Analytic Strategies for Enhancing the Power of Detecting G X

描述（由申请人提供）：本提案是对关于人类疾病中基因-环境相互作用检测方法和途径的资助机会公告PAR- 11-032（R21）的回应。该提案将由多名PI领导，密歇根大学生物统计学系的Bhramar Mukherjee博士和宾夕法尼亚大学生物统计学和流行病学系的Jinbo Chen博士。史蒂芬博士B。来自密歇根大学的Gruber、Sung Kyun Park博士和Naisyin Wang博士是该项目的主要临床和方法学顾问。在本提案中，我们将有两个具体目标：（i）在后全基因组关联研究（GWAS）时代评估有效的两阶段设计和分析选择，其中在现有研究基础上优先选择研究受试者的子样本收集额外的基因分型或生物标志物数据。这包括可能使用关于病例的补充数据和仅具有遗传或环境数据的对照。这些方法以现代回顾性可能性框架为指导。(ii)使用PI开发的称为“主要相互作用分析”的新技术，开发用于筛选队列研究中相互作用的方法。该方法是基于一个简约的低秩表示的互作矩阵后，拟合基因和环境的加性主效应。该提案计划将这种方法扩展到纵向研究，以捕捉相互作用的时变效应。作为副产品，将开发视觉诊断，以确定至关重要的时间窗口。这一重要提案中的计划工作将为FOA的使命做出有意义的贡献，并推进研究G x E效应的研究设计和分析技术。该提案将涉及陈博士和慕克吉博士，他们的博士/博士后学员之间的积极合作，并促进两个同行机构之间的合作：宾夕法尼亚大学和密歇根大学。该提案涉及统计学、医学、流行病学和人类遗传学在方法学发展方面的交叉。更广泛的影响是更好地了解疾病的病因，并确定有针对性的干预策略的潜力。 公共卫生相关性：本提案是为了响应关于人类疾病中基因-环境相互作用检测方法和途径的资助机会公告PAR- 11-032（R21）而提交的。该提案由多个PD/PI领导，密歇根大学的Bhramar Mukherjee博士和宾夕法尼亚大学的Jinbo Chen博士。基因和环境的协同作用在复杂疾病的病因学中起着重要作用。该提案针对流行病学研究中有效检测基因-环境相互作用的重要研究设计和分析挑战。在第一个具体目标中，我们考虑了与选择病例/对照的策略相关的设计和分析方法，用于在现有研究基础上进行额外的基因分型或收集生物标志物数据。在第二个具体的目标，我们考虑了一个非常新颖的策略，探索在纵向研究中的相互作用，基于奇异值分解的残余相互作用的对比矩阵后，去除添加剂的影响。我们称这种分析为主相互作用分析，因为它与主成分分析相似。该建议预计将大大有助于现有的文献G x E的研究。

DESCRIPTION (provided by applicant): This proposal is in response to the Funding Opportunity Announcement PAR- 11-032 on Methods and Approaches for Detection of Gene- Environment Interactions in Human Disease (R21). The proposal will be led by multiple PIs, Dr. Bhramar Mukherjee at the Department of Biostatistics, University of Michigan and Dr. Jinbo Chen at the Department of Biostatistics and Epidemiology, University of Pennsylvania. Dr. Stephen B. Gruber, Dr. Sung Kyun Park and Dr. Naisyin Wang from the University of Michigan are key clinical and methodological consultants on the project. In this proposal, we will have two specific aims: (i) Evaluate efficient two-phase design and analysis choices in the post genomewide association studies (GWAS) era where additional genotyping or biomarker data is collected on a prioritized selection of a sub-sample of study subjects in an existing study base. This includes the possibility of using supplementary data on cases and controls with only genetic or environmental data. The methods are guided by modern retrospective likelihood framework. (ii) Develop methods for screening of interaction in cohort studies using a novel technique developed by the PIs called "Principal Interactions Analysis". This method is based on a parsimonious low rank representation of the interaction matrix after fitting additive main effects of gene and environment. The proposal plans to extend this method to longitudinal studies to capture time- varying effects of interaction. Visual diagnostics to identify time-windows of critical importance will be developed as a byproduct. The planned work in this important proposal will meaningfully contribute to the mission of this FOA, and advance study design and analytical techniques for studying G x E effects. The proposal will involve active collaboration between Dr. Chen and Dr. Mukherjee, their doctoral/post-doctoral trainees and foster collaboration between two peer institutions: University of Pennsylvania and University of Michigan. The proposal lies in the intersection of statistics, medicine, epidemiology and human genetics in terms of methodology development. The broader impact is better understanding of disease etiology and identify potentials for targeted intervention strategies. PUBLIC HEALTH RELEVANCE: This proposal is submitted in response to the Funding Opportunity Announcement PAR- 11-032 on Methods and Approaches for Detection of Gene-Environment Interactions in Human Disease (R21). The proposal is lead by multiple PD/PI, Dr. Bhramar Mukherjee at the University of Michigan and Dr. Jinbo Chen at the University of Pennsylvania. Synergism of genes and environment plays an important role in the etiology of complex diseases. This proposal addresses important study design and analytical challenges for efficient detection of gene-environment interactions in epidemiological studies. In the first specific aim, we consider design and analytical methods associated with strategies for selection of cases/controls for additional genotyping or collection of biomarker data in existing study bases. In the second specific aim, we consider a highly novel strategy for exploring interactions in longitudinal studies, based on a singular value decomposition of the residual interaction contrast matrix after removing additive effects. We call this analysis Principal Interactions Analysis, due to its similarity with Principal Components Analysis. The proposal is expected to contribute significantly to the existing literature on G x E studies.