Statistical Methods For Gene/environment Interaction And Genetic Susceptibility

我们早先对倍增基因-环境相互作用的病例-父母数据的分析方法在两个关键假设下导致了对致因SNP的有效推断。第一个假设是，根据父母的基因类型，儿童的基因分布反映了孟德尔式的分类。第二点是，根据父母的基因型别，孩子的基因型别和接触情况是独立的。 最近，我们一直在考虑一项研究设计，包括一个受影响的后代和一个未受影响的后代及其父母。我们称这种结构为四分体。我们的建议是对受影响的后代和父母进行基因分型，并从两个后代那里收集暴露信息，在这种想法下，我们可以使用嵌入的病例-父母-三联体设计来测试遗传和基因-环境交互作用的影响，我们可以使用嵌入的兄弟姐妹对设计来研究暴露。在研究这一设计时，我们了解到，以前提出的基于家庭的基因-环境相互作用测试，当亚群在等位基因频率和暴露流行率方面存在差异，并且研究中的SNP是一个标记而不是导致SNP本身时，可能是有偏见的。这一发现既令人惊讶又令人担忧，因为研究人员此前认为，即使在研究标记时，基于家庭的基因与环境相互作用的研究也会对群体结构的偏差产生强烈的影响。我们已经发表了一篇手稿，详细描述了这些问题，并提出了一种稳健的四分体设计或不协调同胞对的分析方法，当研究中的暴露是二分的时，该方法可以为标记提供关于基因-环境相互作用的有效推断。在一篇相关的手稿中，我们提出了一种仅兄弟姐妹扩增病例的设计，并表明，通过适当的分析，它提供了与我们为四分体和疾病不一致的同胞对设计提出的分析相同的关于基因-环境相互作用的稳健推断。 我们还发现，对于有一个受影响的兄弟姐妹和两个或更多未受影响的兄弟姐妹的匹配设计，将兄弟姐妹关系视为一个缺少父母的核心家庭比传统的条件Logistic回归更有效，而且它也允许包含不匹配的受试者。 我们正在研究当基因分型分析计算混合样本中变异等位基因的数量时，样本池对来自病例-父母三合一的DNA的应用。我们的程序将一个三联体样本划分为一个小的子集，每个子集有两个三联体，并为每个子集构建三个汇集的DNA样本：分别来自母亲、父亲和后代。我们的对数线性建模方法将组成每个池的单个基因类型视为缺失数据，使用期望最大化算法来估计遗传等位基因、母系等位基因或父母起源效应的相对风险参数，这是其他DNA池方法无法做到的。当没有错误地测量基因型别时，我们看到与基因分型个体相比，能力损失很小，但随着基因分型错误率的增加，能力下降。对于足够准确的分析，我们的方法承诺以最小的功率损失降低基因分型成本。 (另见Z01 ES040007 BB；Pi Clare Weinberg。)

Our earlier approach to analyzing case-parents data for multiplicative gene-environment interaction leads to valid inference for a causative SNP under two crucial assumptions. The first assumption is that, conditional on parents' genotypes, the genotype distributions of children reflect Mendelian assortment. The second is that, conditional on parents' genotypes, a child's genotype and exposure are independent. Recently, we have been considering a study design that involves one affected and one unaffected offspring and their parents. We call this structure a tetrad. Our proposal is to genotype the affected offspring and the parents and to collect exposure information from both offspring under that idea that we could test genetic and gene-environment interaction effects using the embedded case-parent-triad design and we could study exposure using the embedded sibling-pair design. In studying this design, we learned that previously proposed family-based tests of gene-environment interaction can be biased when subpopulations differ in both allele frequency and exposure prevalence and when the SNP under study is a marker and not itself the causative SNP. This finding was both surprising and troubling, as researchers had previously believed that family-based studies of gene-environment interaction would robust to bias from population structure even when studying markers. We have published a manuscript that describes these issues in detail and proposes a robust method of analysis for tetrad designs or discordant sib-pairs that can provide valid inference about gene-environment interactions for markers when the exposure under study is dichotomous. In a related manuscript, we have proposed a sibling-augmented case-only design and showed that, with appropriate analysis, it provides the same robust inferences for gene-environment interactions as the analyses we proposed for the tetrad and disease-discordant sib-pair designs. We also found that, for matched designs with one affected and two or more unaffected sibling, treating the sibship as a nuclear family with missing parents is more efficient than traditional conditional logistic regression, and it allows inclusion of unmatched subjects as well. We are studying the application of specimen pooling to DNA from case-parent triads when the genotyping assay counts the number of variant alleles in a pooled specimen. Our procedure partitions a sample of triads into small subsets of, say, two triads each, and, for each subset, constructs three pooled DNA specimens: one each from mothers, from fathers, and from offspring. Treating the individual genotypes that comprise each pool as missing data, our log-linear-modeling approach uses the expectation-maximization algorithm to estimate relative risk parameters for inherited alleles, maternal alleles, or parent-of-origin effects, something other DNA pooling approaches cannot do. We see little loss of power compared to genotyping individuals when genotypes are measured without error, but power declines as genotyping error rates increase. For sufficiently accurate assays, our approach promises to reduce genotyping costs with minimal loss of power. (see also Z01 ES040007 BB; PI Clare Weinberg.)