光华讲坛——社会名流与企业家论坛第 5680 期

主题：MR-LDP: a two-sample Mendelian randomization for GWAS summary statistics accountinglinkage disequilibrium and horizontal pleiotropy.

主讲人：Duke-NUS 医学院成青博士

主持人：统计学院统计研究中心 林华珍教授

时间：2020年01月02日（星期四）下午3:30-4:30

地点：西南财经大学柳林校区弘远楼408会议室

主办单位：统计研究中心 统计学院 科研处

主讲人简介：

Qing received her Ph.D. in Statistics from the Shanghai University of Finance and Economics.  She is now a research fellow at Duke-NUS Medical School. Her current research interests are on the functional regression model, interaction detection, Empirical Bayes, Variational inference and Bayesian variable selection.

成青在上海财经大学获得统计学博士学位，现在是Duke-NUS医学院的研究员。目前主要研究方向为函数回归模型、交互检测、经验贝叶斯、变分推论和贝叶斯变量选择。

主要内容：

The proliferation of genome-wide association studies (GWAS) has prompted the use of two-sample Mendelian randomization (MR) with genetic variants as instrumental variables (IV) for drawing reliable causal relationships between health risk factors and disease outcomes. However, the unique features of GWAS demand MR methods account for both linkage disequilibrium (LD) and ubiquitously existing horizontal pleiotropy among complex traits, which is a phenomenon that a variant affects the outcome other than exclusively through the exposure. Therefore, statistical methods that fail to consider LD and horizontal pleiotropy can lead to biased estimates and false-positive causal relationships. To overcome these limitations, we propose a probabilistic model for MR analysis to identify causal effect between risk factors and disease outcomes by using GWAS summary statistics in the presence of LD, as well as properly accounts for horizontal Pleiotropy among genetic variants (MR-LDP). MR-LDP utilizes a computationally efficient parameter-expanded variational Bayes expectation-maximization (PX-VBEM) algorithm, calibrating the evidence lower bound (ELBO) for a likelihood ratio test. We further conducted comprehensive simulation studies to demonstrate the advantages of MR-LDP over existing methods in terms of both type-I error control and point estimates. Moreover, we used two real exposure-outcome pairs (CAD-CAD and BMI-BMI; CAD for coronary artery disease and BMI for body mass index) to validate results from MR-LDP in comparison with alternative methods, particularly showing that our method is more efficient using all instrumental variants in LD. By further applying MR-LDP to lipid traits and BMI as risk factors on complex diseases, we identified multiple pairs of significant causal relationships, including protective effect of high-density lipoprotein cholesterol (HDL-C) on peripheral vascular disease (PVD), and positive causal effect of body mass index (BMI) on hemorrhoids.

全基因组关联研究（GWAS）的兴起促使人们采用两样本孟德尔随机化（MR）和遗传变异作为工具变量（IV），在健康风险因素和疾病结局之间建立可靠的因果关系。然而，GWAS所需的MR方法的独特特征同时导致了连锁不平衡（LD）和复杂性状中普遍存在的水平多效性，这是一种不只是通过暴露来影响结果的变异的现象。因此，不能考虑LD和水平多效性的统计方法可能导致有偏估计和假阳性因果关系。为了克服这些局限性，我们提出了一个用于MR分析的概率模型，通过在LD存在的情况下使用GWAS汇总统计来确定风险因素和疾病结果之间的因果关系，并适当地解释了遗传变异之间的水平多效性（MR-LDP）。 MR-LDP利用一种计算有效的参数展开的变分贝叶斯期望最大化（PX-VBEM）算法，校准似然比检验的证据下界（ELBO）。我们进一步进行了全面的模拟研究，以证明MR-LDP相对于现有的方法在I型差错控制和点估计方面的优点。此外，我们使用了两个真实的暴露结果对（CAD-CAD和BMI-BMI；冠心病CAD和体重指数BMI）来验证MR-LDP与其他方法相比的结果，突出表明我们的方法更有效地使用了LD中的所有工具变量。通过进一步将MR-LDP应用于血脂特征和BMI作为复杂疾病的危险因素，我们确定了多对显著的因果关系，包括高密度脂蛋白胆固醇（HDL-C）对周围血管疾病（PVD）的保护作用和体重指数（BMI）对痔疮的正因果作用。