人类复杂性状“消失的遗传率”统计遗传学分析方法研究

Heritability is the central concept for statistical genetics. For human complex traits, estimating heritability is a necessary step for genetic analyses. Uncertainty in the estimation of heritability leads to “missing heritability”, which has unveiled mechanism. In his previous investigation, the applicant has already found that “missing heritability” is driven by experimental design and statistical methods, and consequently has established, based on the principles of population genetics, a framework built on Haseman-Elston regression (HE). The new tool is called Haseman-Elston variance component (HE-VC), which has merits in dissecting the mechanism underlying “missing heritability”. HE had been known as a tool for linkage analysis, and the applicant has already converted it to a tool for estimating variance component using whole genome-wide markers. In this proposal, the applicant tries to further investigate the HE-VC in the following aspects. I) Theory and algorithm for HE-VC. Based on the established framework, the applicant is going to reconcile the controversy often observed between linkage and association studies, between population-based design and nuclear family design; furthermore, the HE-VC framework will be extended to bivariate analysis. II) Real data analyses. HE-VC will be applied to various GWAS data, and estimate heritability for human complex traits. This proposal will bring out an empowered HE-VC, which is efficient for large GWAS data analysis for human complex traits, and establish the foundation for future genetic investigation for even larger genetic data.

遗传率是统计遗传学的核心，估算遗传率是人类复杂性状研究的必要步骤。遗传率估算的不准确导致“消失的遗传率”（missing heritability）问题出现。申请人在前期工作中发现“消失的遗传率”是由实验设计、分析方法共同作用产生，此后申请者重构了Haseman-Elston（HE）模型，提出了可以解释“消失的遗传率”问题的初步理论和技术框架–基于HE的方差分量方法HE-VC。本项目拟在之前HE-VC工作基础上，I）在理论和算法层面拓展HE-VC框架，破解广泛存在于连锁分析和关联分析、家系设计和群体设计之间的“消失的遗传率”问题，并推广到成对性状遗传相关性估算；借助HE-VC的计算性能优势，开发高性能计算软件。II）在应用层面，HE-VC方法将应用到人类复杂性状遗传率估算。相比混合线性模型框架，HE-VC统计遗传方法运算优势明显，将为人类遗传大数据分析提供有力的理论基础分析工具。

本研究围绕复杂性状“消失的遗传率”主题展开了一系列研究。1）大样本数据计算，复杂性状的遗传率估算需要大样本才能达到比较高的精度。基于随机矩阵计算理论，我们开发了针对遗传学大数据的计算方法—随机Haseman-Elston回归（RHE），解决了在30万样本量的UKBiobank的数据上计算问题，且获得比较精准的计算结果。2）改进了我们之前提出的非监督Fst扫描方法，探索了群体遗传结构对于遗传率的影响，找到遗传位如何受到自然选择和人工选择的影响，并且在各种测序平台和物种上得到应用。3) 结合课题组开发的RHE算法，在预测精度持平情况下大大提高了GBLUP的计算速度。