Coarse-to-fine Discovery for Genetic Association

孟德尔性状是由单个基因控制的，识别这些基因的方法已经非常成功。与之形成鲜明对比的是，复杂性状是由多个基因变异产生的，这些基因变异既不是必要的，也不是充分的，它们经常相互作用，还与环境相互作用。事实上，总的来说，迄今为止鉴定出的复杂性状的遗传变异通常只能解释不到10%的表型变异。目前还没有提出统一的计算和统计框架来组织发现过程。迄今为止，一种单一的策略占主导地位：静态变量对变量的分析。相比之下，研究人员提出了新的从粗到精的统计框架，其动机是生物医学假设，即在特定途径和这些途径内的基因中导致特定疾病的突变聚集。仿真结果表明，在此假设下，多尺度、分层的粗到细序列测试比传统方法具有更大的有效性。研究人员将这些启发转化为数学，并提供了一个全面的分析，从经验和理论两方面，对引入关于基因和途径中活跃变异分布的精心选择的偏差所产生的权衡进行了分析。新方法被应用于具有大队列的真实全基因组关联研究（GWAS）的数据，以验证其实用性。了解导致心血管疾病、糖尿病、自闭症和其他流行疾病的基因变异，将对确定药物靶点、预测高危人群和建议个性化治疗具有重要价值。然而，这些疾病不是由单个基因的突变引起的，而是由多个突变联合起来破坏多基因生物学途径引起的。因此，研究人员开发了一种新的统计框架，从途径水平开始搜索疾病风险基因，然后依次将搜索范围缩小到途径内的基因和基因内的等位基因。正在进行的涉及数万至数十万人的人类基因研究成功应用于识别导致心血管疾病的基因。更普遍的是，在当前“大数据”时代，从粗到精的统计框架具有巨大的价值，越来越大的数据量需要创新的统计方法。

Mendelian traits are governed by single genes, and methods to identifythese genes have been remarkably successful. In stark contrast,complex traits result from multiple genetic variants that areindividually neither necessary nor sufficient, often interacting witheach other and the environment. Indeed, collectively, geneticvariants identified to date for complex traits typically explain lessthan 10% of the phenotype variance. No unified computational andstatistical framework has been advanced for organizing the discoveryprocess. To date, a single strategy has dominated: staticvariant-by-variant analysis. In contrast, the investigators propose anew coarse-to-fine statistical framework motivated by the biomedicalhypothesis that mutations contributing to a specific disease clusterin specific pathways, and in genes within these pathways. Simulationsdemonstrate that multi-scale, hierarchical coarse-to-fine sequentialtests have greater power than conventional methods under thishypothesis. The researchers convert these heuristics into mathematicsand provide a comprehensive analysis, both empirical and theoretical,of the trade-offs resulting from the introduction of carefully chosenbiases about the distribution of active variants within genes andpathways. The new methods are applied to data from real genome-wideassociation studies (GWAS) with large cohorts to validate theirutility.Knowing the genetic variants that contribute to cardiovasculardisease, diabetes, autism, and other prevalent disorders would havegreat value in identifying drug targets, predicting people at risk,and suggesting personalized therapies. These diseases are not causedby mutations in single genes, however, but by multiple mutations thatcombine to disrupt multi-gene biological pathways. The investigatorstherefore develop a new statistical framework that begins the searchfor disease-risk genes at the pathway level, then sequentially narrowsthe search to genes within pathways and alleles within genes.Successful applications to ongoing human genetic studies involvingtens to hundreds of thousands of people identify genes contributing tocardiovascular disease. More generally, the coarse-to-finestatistical framework has great value in the current era of "bigdata", with increasingly large data volumes calling for innovativestatistical methods.