Improving Polygenic Prediction using Next-Generation Data Sets

了解基因型和表型之间的关系是遗传学的中心目标。可用 对许多与医学相关的人类性状的遗传力估计表明，30%-80%的表型变异 是由于潜在的遗传变异。根据基因类型预测表型的能力是终极考验。 我们对复杂性状遗传学的理解。自20世纪初复杂性状遗传学问世以来 在21世纪，进展一直受到表型良好的人群中遗传数据可获得性的限制。现在，到期了 由于技术的非凡进步，微阵列基因分型数据集，外显子组测序数据集和 靶向测序数据集可用于大量临床表型人群，功能数据是 变得有空。未来全基因组测序数据的爆炸性增长也被广泛预期。这种情况发生了转变 重点从数据采集到数据解释以及计算和统计方法的发展 用于根据基因类型和功能信息预测表型。我们建议开发新的方法来 从基因类型预测表型并将这些方法应用于新收集的关于人类复合体的数据 具有直接医学意义的性状，包括数量性状和疾病性状。我们在表型预测方面的工作 将提供有关复杂性状的等位基因结构的信息，并将为未来的遗传学提供指导 学习。从实践的角度来看，关于基因诊断的潜力正在进行辩论 在生命早期识别出特定复杂疾病的高风险个体。如果成功，基因 诊断可以为选择患者进行早期治疗干预提供信息。然而，它的实际效用 遗传学在评估复杂疾病风险中的作用尚未得到证实，而且还存在广泛的争论。我们将严格执行 检验基于基因型的表型预测效用的假设。 在具体目标1中，我们将开发和测试新的统计方法来预测微阵列的表型 基因分型数据。我们将研究几种型号选择和收缩策略。我们将评估 单独估计单个标记的贡献率还是拟合所有标记更有效 同时。在具体目标2中，我们将改进不同血统人群的多基因预测。它 重要的是，医学进步不限于欧洲人口。我们的方法将产生 跨人群的预测，考虑到等位基因频率的人群差异， 等位基因变异和连锁不平衡模式。在具体目标3中，我们将开发和测试统计学 根据测序数据预测表型的方法。测序数据提供了一组不同的统计数据 挑战，因为它们含有低频和罕见的等位基因变异，而且往往是个别罕见的影响 变种是无法估计的。在具体目标4中，我们将把函数数据合并到方法中 表型预测。我们将调查合并功能数据是否可以改善表型 来自基因数据的预测。

Understanding the relationship between genotype and phenotype is the central goal of genetics. Available heritability estimates for many human traits of medical relevance suggest that 30-80% of phenotypic variation is due to underlying genetic variation. The ability to predict phenotypes based on genotypes is the ultimate test of our understanding of complex trait genetics. Since the dawn of complex trait genetics in the early 20th century, progress has been limited by the availability of genetic data in well-phenotyped populations. Now, due to the extraordinary progress in technology, microarray genotyping datasets, exome sequencing datasets and targeted sequencing datasets are available for large clinically phenotyped populations, and functional data is becoming available. A future explosion of whole-genome sequencing data is also widely anticipated. This shifts the focus from data acquisition to data interpretation and development of computational and statistical methods for predicting phenotypes from genotypes and functional information. We propose to develop new methods for predicting phenotypes from genotypes and apply these methods to newly collected data on human complex traits of direct medical interest, including both quantitative and disease traits. Our work on phenotype prediction will be informative about the allelic architecture of complex traits and will provide guidance for future genetic studies. From a practical perspective, there is an ongoing debate on the potential of genetic diagnostics in identification of individuals at elevated risk for specific complex diseases early in life. If successful, genetic diagnostics may inform selection of patients for early therapeutic intervention. However, the practical utility of genetics in evaluating risk of complex diseases has not been proven and is widely debated. We will rigorously test the hypothesis of the utility of genotype-based phenotypic predictions. In Specific Aim 1 we will develop and test new statistical methods for predicting phenotypes from microarray genotyping data. We will investigate several model selection and shrinkage strategies. We will evaluate whether it is more efficient to estimate contributions of individual markers independently or to fit all markers simultaneously. In Specific Aim 2 we will improve polygenic prediction in populations of diverse ancestry. It is important that medical progress not be limited to European populations. Our methods will generate predictions across human populations, accounting for population differences in allele frequencies, rates of allelic variation and patterns of linkage disequilibrium. In Specific Aim 3 we will develop and test statistical methods for predicting phenotypes from sequencing data. Sequencing data provide a distinct set of statistical challenges because they contain low-frequency and rare allelic variants, and often the effects of individual rare variants cannot be estimated. In Specific Aim 4 we will incorporate functional data into methods for phenotype prediction. We will investigate whether incorporation of functional data can improve phenotype predictions from genetic data.