CAREER: Flexible and efficient mixed models to infer the genetic architecture of complex phenotypes

复杂的表型，即可观察到的特征，是生物学和医学的中心焦点。越来越多的数据收集调查了基因和表型在不同个体之间的差异，这为系统地了解复杂表型的遗传结构提供了诱人的机会。从这些大量的高维基因组数据中得出关于遗传结构的推论，将阐明预测有机体表型的一套规则。基于混合模型的方法，即对大量遗传变异的联合影响进行建模，已成为这一努力中的重要工具。然而，混合模型方法依赖于许多简化的建模假设，这可能会导致有偏见的推断。此外，将这些方法应用于大规模基因数据集在计算上是不切实际的。该项目将开发新的、可扩展的方法，可以表征复杂特征的遗传结构。将这些方法应用于现有的大型遗传数据集，将导致对复杂表型变异背后的基因的新见解，从而进一步揭示生命规律。该项目的跨学科方面将把来自计算机科学、统计学、生物信息学和人类遗传学的研究人员和学生聚集在一起，并将导致这些社区的交叉受精和更密切的互动。该项目将开发灵活和高效的新的计算混合模型方法，并应用这些方法来获得对遗传结构的新见解。具体地说，该项目将开发1)线性混合模型，提供对广泛遗传结构的遗传力的准确估计，2)非线性混合模型，估计基因-基因和基因-环境交互作用的贡献，以及3)多性状混合模型，估计性状间共有的遗传成分。重要的是，建议的方法旨在扩展到包含数百万个人的数据集。为了证明它们的实用性，我们将把这些方法应用于大型遗传数据集，以获得关于遗传力的新见解，它在基因组中的分布，它与其他特征的相关性，基因-基因和基因-环境相互作用的贡献，以及自然选择的影响。这一奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Complex phenotypes, i.e., observable characteristics, are a central focus of biology and medicine. Growing collections of data that survey how genes and phenotypes vary across individuals present the tantalizing opportunity to systematically understand the genetic architecture of complex phenotypes. Drawing inferences about genetic architecture from these large collections of high-dimensional genomic data will elucidate the sets of rules that predict an organism’s phenotype. Methods based on mixed models, that model the joint effects of large numbers of genetic variants, have emerged as an important tool in this endeavor. Mixed model methods, however, rely on a number of simplifying modeling assumptions that can lead to biased inferences. Further, applying these methods to large-scale genetic datasets is computationally impractical. The project will develop novel, scalable methods that can characterize genetic architecture of complex traits. The application of these methods to large genetic datasets available will lead to novel insights into genes that underlie variation in complex phenotypes, which leads to further uncover rules of life. The inter-disciplinary aspect of the project will bring together researchers and students from computer science, statistics, bioinformatics and human genetics and will lead to cross-fertilization and closer interactions across these communities. The project will develop new computational mixed model methods that are flexible and efficient and to apply these methods to obtain novel insights into genetic architecture. Specifically, the project will develop 1) linear mixed models that provide accurate estimates of heritability across a wide range of genetic architectures, 2) non-linear mixed models that estimate the contribution of gene-gene and gene-environment interactions, and 3) multi-trait mixed models that estimate the genetic component shared across traits. Importantly, the proposed methods are designed to scale to datasets that contain millions of individuals. To demonstrate their utility, we will apply these methods to large genetic datasets to obtain novel insights into heritability, its distribution across the genome, its correlation with other traits, the contribution of gene-gene and gene-environment interactions, and the impact of natural selection.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.