READMEmr.md

In SharonLutz/reverseCT: This package contains methods to run mediation analysis and MR with reverse causality

reverseC for Mendelian Randomization

The reverseMRsim function in the reverseC R package examines the performance of Mendelian Randomization (MR) methods in the presence of reverse causality. Through simulation studies, this R function examines the type 1 error rate and power for 3 popular MR methods when the role of the intermediate phenotype and outcome were correctly specified and when they were reversed (i.e. reverse causality).

Installation

Requirements: R v3.4 or higher You will need the proper compiling tools for your platform. * For Windows (Rtools installer): https://cran.r-project.org/bin/windows/Rtools/ * For MacOSX (clang and gfortran): https://cran.r-project.org/bin/macosx/tools/

install.packages("devtools") # devtools must be installed first

# these will fail to install when already loaded, and install_github will sometimes 
# load these as part of its activity, and will then try to install them if they need 
# an update for one of the package dependencies
install.packages(c("Rcpp","RcppEigen", "curl"), quiet=T) 

#this package does not install automatically, but is needed by TwoSampleMR
install.packages("psych")

devtools::install_github("MRCIEU/TwoSampleMR") # this is a dependency not present in R CRAN, it should be installed before reverseC

devtools::install_github("SharonLutz/reverseC",quiet=T)

The install process will involve compiling source code. If you are on MacOSX, this may involve the clang compiler issuing warnings about unknown pragmas similar to the text below. Do not be alarmed if you see these. If there is actually an error, it will be present among the last several messages issued by the compiler.

warning: pragma diagnostic pop could not pop, no matching push [-Wunknown-pragmas]

Input

nSNP is the number of SNPs generated from a binomial distribution for n subjects (input n) for a given minor allele frequency (input vector MAF).

For the SNPs Xi, the mediator/ exposure M is generated from a normal distribution with the variance (input varM) and the mean as follows:

E[M] = γo + ∑ γX Xi

All of these values are inputted by the user (i.e. the intercept gamma0, and the genetic effect size as a vector gammaX).

The outcome Y is generated from a normal distribution with the variance (input varY) and the mean as follows:

E[Y] = βo + βM M

All of these values are inputted by the user (i.e. the intercept beta0 and the effect of the mediator directly on the outcome as betaM).

After the SNPs X, mediator M, and outcome Y are generated, then the reverseMRsim function compares the power and type 1 error rate of the following 3 methods to detect the path from M to Y: Egger Regression, the Median Weighted Approach, and the Inverse Variance Weighted (IVW) Approach.

Example 1:

For 1,000 subjects (n=1000), we generated 10 SNPs (nSNP=10) with a minor allele frequency of 20% (specified by MAF) that have a genetic effect size of 0.4 (specified by gammaX) on the normally distributed mediator and the mediator has an effect size varying from 0, 0.2 to 0.3 (specified by betaM) on the normally distributed outcome. We considered 3 MR approaches: Egger Regression, the Median Weighted Approach, and the Inverse Variance Weighted (IVW) Approach.

library(reverseCT)
?reverseMRsim # For details on this function

reverseMRsim(n = 1000, nSNP = 10, MAF = rep(0.2, 10), gamma0 = 0, gammaX = rep(0.4, 10), 
varM = 1, beta0 = 0, betaM = c(0, 0.2,0.3), varY = 1, nSim = 500, plot.pdf = T, 
plot.name = "reverseMRplot.pdf", alpha_level = 0.05, SEED = 1001)

Output 1:

For the example, we get corresponding plot. In the plot below, the methods ending in NR have the true outcome as the outcome where as the methods ending in R have the true outcome reversed with the mediator. When the mediator and outcome are reversed, the Egger regression and the Median Weighted Approach have an inflated type 1 error rate. While the IVW approach does not have an inflated type 1 error rate, there is very little difference in the IVW approach if the mediator and outcome are reversed, which implies that this approach cannot easily distinguish the causal relationship between the mediator and outcome.

Example 2:

library(reverseCT)
?reverseDirection # For details on this function

reverseMRsim(n = 1000)

Output 2:

Warning: Do not try to access package internals directly or do so at your own risk!

If you try to run methods/functions that are not exported and intended for end users, and feed these functions environments, parameters, or values that are not correctly formed, it could result in an uncaught or uncatchable C++ exception or segmentation fault. If this occurs, it will kill your R session/terminal and if you were working within RStudio it will probably crash too.

References

MR.Egger is the Egger Regression approach to MR.

Bowden J., Davey Smith G., & Burgess S. (2015). Mendelian Randomization 
with invalid instruments: effect estimation and bias detection through 
Egger regression. International Journal of Epidemiology, 44(2), 512-525. 

MR.IVW is the Inverse Variant Weighted approach to MR.

Burgess, S., Butterworth, A., & Thompson, S. G. (2013). Mendelian 
Randomization Analysis With Multiple Genetic Variants Using Summarized 
Data. Genetic Epidemiology, 37(7), 658-665.

MR. Median is the Median Weighted approach to MR.

Bowden, J., Davey Smith, G., Haycock, P. C., & Burgess, S. (2016). Consistent 
Estimation in Mendelian Randomization with Some Invalid Instruments Using a 
Weighted Median Estimator. Genetic Epidemiology, 40(4), 304-314. 

SharonLutz/reverseCT documentation built on Oct. 30, 2019, 11:54 p.m.