In jwr-git/mrpipeline: Implements a Ready-for-Use Mendelian Randomisation Pipeline

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

library(mrpipeline)

Introduction

The pipeline contains the ability to run Mendelian randomisation analyses, including relevant sensitivity analyses and visualisation of results from these analyses.

Performing MR

The pipeline expects a harmonised dataset to conduct MR.

exp_dat <- read_exposure("eqtl-a-ENSG00000167207")
out_dat <- read_outcome("ieu-a-7", rsids = exp_dat$SNP)
dat <- harmonise(exp_dat, out_dat)

res <- do_mr(dat)

head(data.frame(res))

The do_mr function conducts the following analyses in this order:

1. If the F-statistic has not been calculated for the exposure data, and an F cutoff threshold is given (default: 10) then the F-statistic will be calculated here.

   • There are two methods which are used to calculate the F-statistic, depending on the data available.
   • If allele frequencies and sample sizes are available, the full F-statistic will be calculated.
     • This function will estimate the portion of variance explained (PVE) and calculate the F-statistic using the formulae presented in this paper.
   • If these data are not available, the approximate F-statistic will be used instead.
     • This is calculated as $F = b^2 / SE^2$.
   • N.B. the F-statistic can be run after loading the exposure data and does not have to be calculated here: ```r exp_dat <- calc_f_stat(exp_dat)

   head(data.frame(exp_dat)) ```

2. The Wald ratio (WR) method is used for single-SNP instruments.

   • If the all_wr argument is TRUE, as it is by default, then the WR will be calculated for all SNPs, even if there are multiple SNPs present for a single exposure.
   • The WR method herein uses the second-term Taylor series expansion for the approximation of the standard error (more information on this, including the derivation, can be found in this paper, supplementary).

3. The inverse variance weighted (IVW) method is used for multi-SNP instruments.

   • This method uses the delta method as detailed in the MendelianRandomization package (see: https://rdrr.io/cran/MendelianRandomization/man/mr_ivw.html).
   • The Egger intercept term is also generated for multi-SNP instruments.

4. Finally, if sample sizes are present for both exposure and outcome datasets, then a test for directionality (Steiger filtering) will be conducted. The pipeline directly calls the TwoSampleMR package for this analysis, so we advise reading those vignettes.

Visualising MR results

Under construction.

Forest plot:

forest_plot(res)

jwr-git/mrpipeline documentation built on Oct. 2, 2022, 3:41 p.m.