西南财经大学统计研究中心系列讲座(第255期)

Duke-NUS医学院刘瑾助理教授: Accounting for correlated horizontal pleiotropy in two-sample Mendelian randomization using correlated instrumental variants



主题:Accounting for correlated horizontal pleiotropy in two-sample Mendelian randomization using correlated instrumental variants

主讲人:Duke-NUS医学院刘瑾助理教授

主持人:统计学院统计研究中心 林华珍教授

时间:2020年7月13日上午11:00-12:00

地点:腾讯会议:373 394 304

主办单位:统计研究中心 统计学院 科研处


主讲人简介:

刘瑾,现为Duke-NUS医学院计量医学中心助理教授,2011年从爱荷华大学取得统计学博士学位。2014年获伊利诺伊大学芝加哥分校Faculty Scholarship Support Fund。 其研究兴趣为统计遗传学/基因组学,生存分析,计算统计学,生物信息学。已发表高质量学术论文近50篇,合著书籍2本;主持科研项目10余项。详情请见其个人主页:http://blog.nus.edu.sg/jinliu/


主要内容:

Mendelian randomization (MR) is a powerful approach to examine the causal relationships between health risk factors (exposures) and outcomes from observational studies. Due to the proliferation of genome-wide association studies (GWASs) and abundant fully accessible GWAS summary statistics, a variety of two-sample MR methods for summary data have been developed to either detect or account for horizontal pleiotropy, primarily based on the instrument strength independent of direct effect (InSIDE) condition, requiring the independence between the effects of variants on exposure (γ) and horizontal pleiotropy (α). However, in many cases, we observe heteroscedasticity in a linear regression for the observed associations between exposures and outcomes. This heteroscedasticity is essentially caused by the correlations between γ and α, which is referred as correlated horizontal pleiotropy (CHP). To account for this CHP, we propose a Bayesian approach, MR-〖"Corr" 〗^2,  that uses the orthogonal projection to reparameterize the bivariate Gaussian distribution for γ and α, and a spike-slab prior to remove the impact of CHP. The proposed strategy cannot only be used to account for CHP in MR, but also can be applied in the case that there exist associations between genetic variants and unobserved confounders. We develop an efficient algorithm with paralleled Gibbs sampling. To demonstrate the advantages of MR-〖"Corr" 〗^2 over existing method, we conducted comprehensive simulation studies to make comparisons for both type-I error control and point estimates in various scenarios. By applying MR-〖"Corr" 〗^2 to study the relationships between pairs in a spectrum of complex traits, we invalidate the previously arguably causal relationship between HDL-c and CAD. Moreover, the results provide a perspective of causal network among complex traits.

孟德尔随机(MR)是检查健康风险因素(暴露)与观察性研究结果之间因果关系的有效方法。由于全基因组关联研究(GWAS)的激增和大量可完全访问的GWAS摘要统计信息,主要基于工具变量的强度,已开发出多种用于汇总数据的双样本MR方法来检测或解释水平多效性与直接效应(InSIDE)条件无关,要求变体对暴露()的影响与水平多效性()之间的独立性。但是,在许多情况下,我们通过线性回归观察到异方差性,从而观察到暴露与结果之间的关联。 这种异方差本质上是由和之间的相关性引起的,这被称为相关水平多向性(CHP)。 为了消除相关水平多向性的影响,我们提出了一种贝叶斯方法MR-,该方法使用正交投影对和的双变量高斯分布重新参数化,并在此基础上使用spike-slab先验消除相关水平多向性的影响。所提出的策略不仅可以用于解释MR中的CHP,而且可以在遗传变异与未观察到的混杂因素之间存在关联的情况下应用。 我们开发了一种具有并行吉布斯采样的高效算法。 为了证明MR-相对于现有方法的优势,我们进行了全面的仿真研究,以比较各种情况下的I型错误控制和点估计。通过应用MR-研究一系列复杂性状中的相互因果关系,我们使HDL-c和CAD之间先前可能存在的因果关系无效。 此外,结果为复杂性因果网络提供了一个视角。



上一条:香港科技大学杨灿副教授:MR-APSS: a unified approach to Mendelian Randomization accounting for pleiotropy, sample overlap and selection bias using the genome-wide summary data

下一条:密西根大学宋学坤教授: Regularized estimation with L-zero penalty in the causal mediation analysis