Skip to content

MR-PL is a one-sample multivariable Mendelian Randomization method based on partial least squares and Lasso regression (MR-PL). MR-PL is capable of considering the correlation among exposures (e.g., imaging features) when the number of exposures is extremely upscaled, while also correcting for winner's curse bias.

Notifications You must be signed in to change notification settings

sldrcyang/MR-PL

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 

Repository files navigation

MR-PL

MR-PL is a one-sample multivariable Mendelian Randomization method, which could take into consideration the correlation among numerous exposures (e.g., brain imaging features) followed by correcting for the winner’s curse bias.

Prepare dependencies

Make sure you have the R package dependencies below installed and accessible in your $PATH.

BiocManager::install('pls')
BiocManager::install('glmnet')
BiocManager::install('hdi')

Input files

  • g_matrix.txt: the genotype matrix of n samples and m snps.
  • exposure_matrix.txt: the exposure matrix of n samples and k exposures.
  • outcome_matrix.txt: the outcome vector of n samples.
  • gwas_assoc.txt: the gwas summary statistics of m snps and k exposures, which is usually derived from a previously published study.

Where the genotype matrix, exposure matrix and outcome are from the same dataset.

Example data are listed in the example/ folder.

Usage

Detail information describes how to perform MR-PL can be found in example.R in the example/ folder.

Briefly, if the gwas summary statistics used to selected instrumental snps is based on the same dataset used for MR analyses, the following can be run

mr_result = mr_pl(g_matrix0, 
                  exposure_matrix0, 
                  outcome, 
                  gwas_assoc, 
                  cutoff = 5e-8, 
                  wcc = TRUE, 
                  c = 20,  #c = 15/25
                  pleiotropy_test = TRUE)

If the gwas summary statistics used to selected IVs is based on another independent dataset, the following can be run

mr_result = mr_pl(g_matrix0, 
                  exposure_matrix0, 
                  outcome, 
                  gwas_assoc, 
                  cutoff = 5e-8, 
                  wcc = FALSE, 
                  c = 0, 
                  pleiotropy_test = TRUE)

Output

mr_result$main_results: exposures with non-zero causal estimate and its corresponding p-value from lasso projection method.

   exposure_name causal_estimate lasso_proj_p
   exposure_1      -0.1908863 1.319257e-15
   exposure_10     -0.1690168 1.085111e-12
   exposure_11     -0.2010710 2.174622e-19
   exposure_12     -0.2086834 5.944588e-30
   exposure_14      0.1851574 3.035465e-18
   exposure_18     -0.1917043 1.215523e-14
   exposure_2       0.1868070 9.948648e-18
   exposure_4      -0.1706325 1.148394e-13
   exposure_6      -0.1796765 1.577770e-16
   exposure_9      -0.1832399 5.285689e-13

mr_result$pleiotropy_test.p: p value of pleiotropy test; if p < 0.05, there exists horizontal pleiotropy, then this MR result should be discarded.

1

mr_result$iv_include: the instrumental snps used for MR analysis (after winner's curse correction).

"snp_256" "snp_208" "snp_795" "snp_76"  "snp_962" ...

mr_result$exposure_include: the exposures used for MR analysis (after winner's curse correction).

"exposure_1"  "exposure_10" "exposure_11" ...

Reproduce

The the reproduce/ folder contains all the codes to reproduce our simulation results, including the baseline simulation and supplementary simulation.

About

MR-PL is a one-sample multivariable Mendelian Randomization method based on partial least squares and Lasso regression (MR-PL). MR-PL is capable of considering the correlation among exposures (e.g., imaging features) when the number of exposures is extremely upscaled, while also correcting for winner's curse bias.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published