The pursuit of genetic causal mechanisms

项目摘要 近年来见证了大型研究项目的发展，这些项目涉及 对数十万人进行基因分型，我们有详细的医疗记录。 例子包括我们所有人的研究项目，百万退伍军人计划和英国生物库 资源。通常，全基因组测序数据也可以用于很大一部分 个人。这些大样本，加上它们精确的基因和表型信息，给出了 我们有机会了解基因变异和 医学上有兴趣的特征达到了下一个水平。 虽然最初的小样本量可用于全基因组关联研究(GWAS) 有动机的分析本质上是近似的，我们现在可以探索更多 与医学相关表型背后的遗传因果机制密切相关。我们可以立志 区分有因果关系的变量和因连锁而关联的变量 不平衡或人口结构。事实上，我们需要更加关注 隐藏混杂因素的影响：当样本大小为 足够大了。 增加我们描述因果机制的分辨率将导致 确定更明确的药物开发目标。它还将提高数据处理的精度 基于基因类型的个性化风险评估：如果我们可以使用变量构建风险分数 他们的表现将在不同种族和环境中保持稳定 曝光。 为了放大具有因果关系的遗传变异，这个项目将利用一套新的 调查人员最近介绍的统计方法。这些新方法 是非常灵活的，因为它们不依赖于如何表现型的特定假设 都与基因变异有关。事实上，它们允许研究人员利用强大的机器 学习算法，并为其结果提供精确的可复制性保证，这一点至关重要。 我们组建了一个多样化和互补性的团队，其中包括统计方面的专家 基因组学、方法统计学和计算机科学，在软件方面都有很好的记录 发育和遗传数据分析。一名博士后学者和两名研究生将 为研究计划做出贡献，以及他们将在 统计、计算和遗传学将增加另一个实质性的好处。

Project Summary Recent years have witnessed the development of large research projects that involve genotyping hundreds of thousands of individuals, on which we have available detailed medical records. Examples include the All of us research project, the Million Veteran Program, and the UKBiobank resource. Often, whole-genome sequencing data is also available for a substantial fraction of the individuals. These large samples, with their precise genotypic and phenotypic information, give us the opportunity to bring our understanding of the relations between genetic variation and traits of medical interest to the next level. While the initial small sample sizes available for genome wide association studies (GWAS) motivated analyses that were approximative in nature, we are now in the position to probe more closely the genetic causal mechanisms underlying medically relevant phenotypes. We can aspire to distinguish variants that have causal effects from those that are associated because of linkage disequilibrium or population structure. Indeed, we need to pay even greater attention to the implications of hidden confounders: even small effects become significant when sample sizes are large enough. Increasing the resolution with which we can describe causal mechanisms will result in the identification of clearer targets for drug development. It will also improve the precision of personalized risk evaluations based on genotypes: if we can construct risk scores using variants that are truly causal, their performance will remain solid across ethnicities and environmental exposures. To zoom in on genetic variants with causal effects, this project will leverage a set of new statistical methodologies that the investigators have recently introduced. These new approaches are remarkably flexible, in that they do not rely on specific assumptions of how phenotypes are linked to genetic variants. Indeed, they allow researchers to capitalize on powerful machine learning algorithms and, crucially, equip their results with precise replicability guarantees. We have assembled a diverse and complementary team, including experts in statistical genomics, methodological statistics and computer science, with a strong record both of software development and genetic data analysis. A postdoctoral scholar and two graduate students will contribute to the research program, and the interdisciplinary training they will acquire in statistics, computation and genetics will add another substantial benefit.