A summary-data-based Mendelian randomization method with application to correlated lipidomic data

项目摘要 近年来，最先进的技术已经允许对组学标记物进行高通量分析， 这些新兴数据已成为了解复杂生物系统的强大工具。同时，基因组- 广泛关联研究（GWAS）与组学标记已经进行，并正在出现的总结 统计学提供了可评估的机会来检查组学标记物和疾病之间的因果关系 结果使用孟德尔随机化（MR）。然而，大多数时候，这些新的组学数据是高- 三维和相关的，在使用MR的重大挑战。我们将制定一个总结数据， 基于 用于高维相关数据的MR方法。我们的模型是一个多变量线性混合模型， 创新之处在于，我们将考虑自变量和因变量中的随机误差， 固定和随机效应模型下的变量。我们的模型允许多组学标记作为独立的 变量，并通过包括具有较不严格的P值的变量作为IV并逐步实施 选择保留最重要的组学标记。我们将按组织进行分层分析 表达及其生物学功能 基因变异来测试因果关系的稳健性。W e将 开发一个R脚本，并在Github上与整个研究社区分享（目标1）。我们将进行 模拟研究，以比较我们的MR模型的性能，以前的摘要数据为基础的MR方法 (aim 2）。高密度脂蛋白胆固醇（HDL-c）水平长期以来被认为是一个重要的逆转 冠心病的预测因子。然而，临床试验和孟德尔随机化表明， HDL-c是HDL功能的不良生物标志物，并提出了一个重要的问题，即其他新的测量方法是否 高密度脂蛋白（即不同颗粒大小的高密度脂蛋白中的脂质）可能对心脏代谢疾病有因果关系。HDL 是具有不同尺寸的颗粒的集合，在脂质组成中高度不均匀。小HDL发挥 胆固醇流出能力越强， 磷脂和较低的甘油三酯和胆固醇， 而大HDL具有较高的甘油三酯和胆固醇含量以及较低的磷脂。照经上所 必须了解HDL脂蛋白中的脂质是否与心脏代谢疾病有因果关系。 我们 将应用我们的MR方法来检查脂质（胆固醇，磷脂和甘油三酯）在 HDL颗粒（非常大、大、中、小、非常小）对心脏代谢疾病（冠心病）风险的影响 疾病，中风和2型糖尿病）使用最近发布的脂质组学数据在英国生物银行和现有的 脂质组分的汇总统计量， MAGNETIC财团（目标3）。

PROJECT SUMMARY In recent years, state-of-the-art technologies have allowed for high-throughput profilings of -omics markers, and these emerging data have been powerful tools to understand complex biological systems. Meanwhile, genome- wide association studies (GWAS) with -omics markers have been conducted, and the emerging summary statistics has provided evaluable opportunity to examine causality between -omics markers and disease outcomes using Mendelian randomization (MR). However, most of the time, these novel -omics data are high- dimensional and correlated, imposing significant challenges in use of MR. We will develop a summary-data- based MR method for use with high-dimensional correlated data. Our mode is a multivariable linear mixed model, and the innovation lies in that we will account for random errors in both independent variables and dependent variable under fixed- and random-effects models. Our model allows for multiple -omics markers as independent variables, and boosts power by including variants with less stringent P value as IVs and implementing stepwise selection which retains the most important -omics markers. We will conduct stratified analysis by tissue expression and biological function of genetic variants to test for robustness of causal associations. W e will develop an R script and share it with the whole research community in Github (aim 1). We will conduct a simulation study to compare the performance of our MR model to previous summary-data-based-MR methods (aim 2). High-density lipoprotein cholesterol (HDL-c) levels have long been recognized an important inverse predictor of coronary heart disease. However, clinical trials and Mendelian randomization suggest that level of HDL-c is a poor biomarker of HDL function, and raise an important question about whether other novel measures of HDL (i.e. lipids in HDLs of different particle sizes) may have causal effects on cardiometabolic disease. HDL is a collection of particles with different sizes highly heterogeneous in lipid composition. Small HDL that exerts stronger cholesterol efflux capacity has higher contents of phospholipids and lower triglycerides and cholesterols, while large HDL has higher contents of triglycerides and cholesterols and lower phospholipids. Thus, it is imperative to understand whether lipids in HDL lipoproteins are causally related to cardiometabolic disease. We will apply our MR method to examine causal effects of lipids (cholesterol, phospholipids, and triglycerides) in HDL particles (very large, large, medium, small, very small) on risk of cardiometabolic disease (coronary heart disease, stroke, and type 2 diabetes) using the recently released lipidomics data in the UK Biobank and existing summary statistics with lipid fractions from the MAGNETIC consortium (aim 3).