Multivariate Modelling and Inference in Genetic Studies

全基因组测序正成为人类复杂疾病基因组扫描的标准手段。因此，几乎每个人的所有遗传变异性都将被测量。然而，由于许多基因变异的数量和频率，检测感兴趣基因的统计能力是一个挑战。许多人类复杂疾病通过相关的数量表型(性状)表现出来。例如，肥胖与体重、腰围和腰臀比有关。然而，直接测量的性状(例如体重、身高)可能不足以检测出易感变异。例如，体重指数(BMI)，定义为体重除以身高的平方，是一个信息量很大的复合表型的典型例子，它被证明与与体重或身高无关的遗传变异有关。因此，鉴于全基因组测序将提供有关DNA变异的全面数据，我们需要优化研究设计的其他方面，例如要研究的表型的选择。这一“表型优化”问题被认为是基因类型映射到复杂疾病表型的方法的一部分。利用表型优化的统计遗传学方法可能会提高定位感兴趣基因的能力，方法是利用相关结构来识别共同影响相关表型组的基因，或者使用更接近基因作用的新表型，这些新表型旨在突出一系列相关路径之间的相关性。我提出的研究计划涉及表型优化，这是现代复杂疾病遗传/基因组分析中最有趣的问题之一。在这项建议中，我建议新的多变量方法，利用多种相关的表型来提高在整个基因组测序时代识别易感基因的统计能力。

Whole genome sequencing is becoming a standard means of genome scanning for human complex diseases. Thus, almost all genetic variability in each person will be measured. However, statistical power to detect genes of interest is a challenge due to the number and frequency of much of the genetic variation. Many human complex diseases manifest themselves through correlated quantitative phenotypes (traits). For instance, obesity is associated with weight, waist circumference and waist to hip ratio. However, direct measured trait along (e. g. weight, height) may not be informative enough to detect susceptibility variants. For instance, body mass index (BMI), defined as weight divided by squared height, is a typical example of a very informative composite phenotype that has shown to be linked with genetic variants not associated with weight nor height. Consequently, given the comprehensive data on DNA variation that will be provided by whole genome sequencing, we need to optimize other aspects of the study design, such as the choice of phenotypes to be studied. This ''phenotype optimization'' problem is believed to be part of the way that genotypes map to the phenotypes for complex diseases. Statistical genetic methods that take advantage of optimizing phenotypes may improve power to localize genes of interest by exploiting correlational structures to identify genes jointly influencing suites of related phenotypes or by using new phenotypes more proximal to gene action, which are designed to highlight independent measures underlying the correlations among a set of related traits.The research program I propose deals with the phenotype optimization, one of the most interesting problems in modern genetic/genomic analysis of complex diseases. Within this proposal, I suggest new multivariate approaches that take advantage of multiple related phenotypes to improve statistical power to identify susceptible genes in the whole genome sequencing era.