Selecting likely causal risk factors from high-throughput experiments using multivariable Mendelian randomization

Verena Zuber,Caroline Klaver,Stephen Burgess,Johanna Maria Colijn

doi:10.1038/s41467-019-13870-3

Abstract

Modern high-throughput experiments provide a rich resource to investigate causal determinants of disease risk. Mendelian randomization (MR) is the use of genetic variants as instrumental variables to infer the causal effect of a specific risk factor on an outcome. Multivariable MR is an extension of the standard MR framework to consider multiple potential risk factors in a single model. However, current implementations of multivariable MR use standard linear regression and hence perform poorly with many risk factors. Here, we propose a two-sample multivariable MR approach based on Bayesian model averaging (MR-BMA) that scales to high-throughput experiments. In a realistic simulation study, we show that MR-BMA can detect true causal risk factors even when the candidate risk factors are highly correlated. We illustrate MR-BMA by analysing publicly-available summarized data on metabolites to prioritise likely causal biomarkers for age-related macular degeneration.

Highlights

Modern high-throughput experiments provide a rich resource to investigate causal determinants of disease risk
In this context we propose a Bayesian model averaging approach (MR-BMA) that scales to the dimension of high-throughput experiments and enables risk factor selection from a large number of candidate risk factors
This nuclear magnetic resonance (NMR) platform provides a detailed characterisation of lipid subfractions, including 14 size categories of lipoprotein particles ranging from extra small (XS) high density lipoprotein (HDL) to extra-extra-large (XXL) very low density lipoprotein (VLDL)

Summary

Introduction

Modern high-throughput experiments provide a rich resource to investigate causal determinants of disease risk. High-throughput experiments provide ideal data resources for conducting MR investigations in conjunction with case-control data sets providing genetic associations with disease outcomes (such as from CARDIoGRAMplusC4D for coronary artery disease[7], DIAGRAM for type 2 diabetes[8], or the International Age-related Macular Degeneration Genomics Consortium [IAMDGC] for age-related macular degeneration[9]). In addition to their untargeted scope, one specific feature of high-throughput experiments is a distinctive correlation pattern between the candidate risk factors shaped by latent biological processes. We assess the performance of our proposed method in a simulation study with scenarios motivated by the metabolite GWAS and by publicly available summary data on blood cell traits measured on nearly 175,000 participants[4]

Objectives

Methods

Results

Conclusion