README.md

In jalabrecque/MRchecks: Tools for detecting genetic variants whose effect changes with age and associated bias

MRchecks: Mendelian randomization biases analyses for time-varying genetic relationships in R

Overview

MRchecks is a package that performs all the analyses in a forthcoming paper on how time-varying exposures can bias Mendelian randomization (MR) analyses. The theory behind this paper can be found in: Interpretation and Potential Biases of Mendelian Randomization Estimates With Time-Varying Exposures. With the functions in this package you can graph and estimate how the association of a genetic variant with a phenotype changes with age. If you’re concerned that the association changes enough that it may bias your MR estimate, there is also a function that will estimate the potential bias under a number of exposure windows.

Installation

Installation requires devtools which can be installed with install.packages("devtools"). Once you have devtools installed you can install MRchecks with:

devtools::install_github("jalabrecque/MRchecks")

Usage

The package comes with a small test data set called test_data:

library(MRchecks)
data(test_data)

The data set contains the minimal variables required:

head(test_data)

##        age SNP           y           c
## 1 54.52100   0 -0.07573056  0.04497838
## 2 52.78986   0  0.10594687 -0.02224538
## 3 57.25049   2  0.23805736  0.05457459
## 4 62.60263   0  1.33247060  0.10780435
## 5 65.62173   1  1.08230059  0.22915236
## 6 69.62285   1  0.74436273 -0.24542123

An age variable (age), a factor variable with the genetic data (SNP) a phenotype that will be used as the exposure in the MR analysis (y) and here, optionally, a covariate is included (c).

Run model

We can first use the SNPxAGE_model function to fit the model we will use as the input for the other functions:

SNPxAGE_model_output <- SNPxAGE_model(data = test_data,
                                      SNP = "SNP",
                                      phenotype = "y",
                                      age = "age",
                                      k = 3,
                                      covars="c")

The argument k specifies how many internal knots should be used.

Plotting age-varying genetic associations

We can plot the relationship between age and the phenotype by genetic variant as well as a plot of additive genetic effects by age:

SNPxAGE_plot(SNPxAGE_model_output)

## NULL

Plasmode simulation for potential bias

Looking at the output of this plot we might be concerned that the genetic association with the phenotype varies enough that bias is a real concern. Using SNPxAGE_bias we can check what the bias would be if all the MR assumptions hold:

SNPxAGE_bias(SNPxAGE_model_output,rep = 10,age_set = 66)

##                     est          se      q025     q975
## y5            1.3036687 0.056756502 1.2207076 1.376539
## y10           1.6274113 0.119992995 1.4524768 1.783534
## y25           2.0274966 0.232921724 1.6884855 2.353241
## y5_gauss      1.5655560 0.106128111 1.4105040 1.702180
## y10_gauss     2.0138771 0.197217929 1.7269802 2.272895
## y25_gauss     2.2528607 0.319949751 1.7883166 2.715701
## y5_obs        0.9930974 0.009312311 0.9800308 1.006238
## y10_obs       1.0224379 0.019739780 0.9935809 1.050915
## y5_gauss_obs  0.9835989 0.016827172 0.9601988 1.007307
## y10_gauss_obs 1.0134841 0.032293326 0.9672633 1.060024

The output lists an estimate for the bias, standard error as well as 2.5% and 97.5% percentile of the bias over the amount of runs listed in rep. The row names indicate which exposure window was used. The number indicates the length of the exposure window. Rows with gauss use a gaussian exposure window and those without use a linear, increasing exposure window. The suffix obs indicates that the exposure was set relative to the age at which the individual was observed in the dataset. NOTE: the true estimate is always set to 1 and the bias should be interpreted multiplicatively. For example, if the estimate is 1.5, this means that we would expect that the age-varying genetic association in combination with that exposure window would lead to a 50% bias in the MR estimate.

jalabrecque/MRchecks documentation built on Dec. 20, 2021, 9:01 p.m.