MR-Base: an online platform for Mendelian randomization using summary data

Kaitlin WADE, University of Bristol, United Kingdom
DAVEY SMITH G. 1,2 , RELTON C. 1,2 , GAUNT T. 1,2 , MARTIN R. 2 , HAYCOCK P. 1,2 , TANSEY K. 1,2 , LAURIN C. 1,2 , SHIHAB H. 1,2 , HEMANI G. 1,2 , ZHENG J. 1,2 , TAN V. 1,2 , LANGDON R. 1,2

1 Integrative Epidemiology Unit (IEU), University of Bristol, Bristol, UK
2 School of Social and Community Medicine, University of Bristol, Bristol, UK

Background: Mendelian randomization (MR) is an increasingly important tool for appraising causality in hypothesized exposure-disease pathways but requires specialist knowledge and access to large genetic datasets. We have created an online platform called MR-Base ( that greatly simplifies the implementation of MR for non-specialists.
Methods: We have collated and harmonized summary genetic data from 34 genome-wide association study (GWAS) consortia, dbGAP and the GWAS catalog, corresponding to >1,200,000 individuals, 141 diseases and 1946 biomarkers, and have automated the MR pipeline through a web-based interface and R package. Implementation of MR through MR-Base involves the following stages: i) selecting instruments for target exposures; ii) choosing statistical methods and iii) specifying outcome traits. In the first stage, the user defines their genetic proxies either through manual upload or selecting exposure from the MR-Base repository. In the second stage, the researcher selects their statistical methods for implementing MR, which currently allows 10 different methods including standard approaches and sensitivity analyses. The user also specifies their preferred method for dealing with correlated genetic proxies as well as strand ambiguity arising from palindromic SNPs. In the third stage, the user selects the outcome of interest from the MR-Base outcome repository. In the fourth and final stage, the analysis is implemented and results are returned in the form of text files and a variety of plots. MR-Base also allows hypothesis-free techniques, such as phenome-wide MR.
Results: MR-Base enables fast and automated application of MR to assess the causal relationship between any known exposure and outcome/disease of interest by combining summary results from hundreds of GWASs and consortia. MR-Base is currently being utilised in a hypothesis-driven context to assess associations of hypothesised risk factors with a variety of outcomes, including lung, prostate and renal cancers, in addition to investigating other risk factors with a hypothesis-free approach.