GenomicSEM

R-package for structural equation modeling based on GWAS summary data

203

GenomicSEM

R-package which allows the user to fit structural equation models
based on the summary statistics obtained from genome wide association studies (GWAS). Note that a recent update includes the ability to automatically obtain Q_SNP results without running a separate follow-up model. Until explicitly stated otherwise the code on this github is an alpha version (now on version 0.0.5c) and under active development. The code may thus produce undesired results on certain operating systems or when run concurrently with specific packages or R versions. Feel free to raise issues if (or when…) the package produces undesired results, we will attempt to swiftly deal with known issues. Please visit the wiki to get started, or check out the paper. If you are having issues and not finding the answers anywhere on the wiki or FAQs page, we encourage you to post your question on the google group.

For Linux users Please read 1.1 For Linux users in the wiki.
v0.0.5c Patch notes: This update contains several bug fixes.
v0.0.5 Patch notes: This update contains some bug fixes as well as optimizations making userGWAS() and commonfactorGWAS() 5-20% faster while using slightly less RAM, greater improvement at higher core counts in parallel (see patchnotes).

v0.0.4 Patch notes: This update is the start of a large code restructure and contains minimal changes for users (see below), but large changes for developers (see patchnotes).
v0.0.4 Feature update: userGWAS(), commonfactorGWAS() and sumstats() can now be run in parallel on Windows systems, additionally munge() can now be run in parallel as well (on both Linux/Mac and Windows). A number of functions now contain checks on input (ideally) preventing the function from stopping halway through analysis due to incorrect input. Input to files argument for munge() and sumstats() should now be a vector (a list will still work for now for backwards compatibility).

v0.0.3d Feature update: GenomicSEM can now run HDL a novel method for estimating heritability and genetic correlation that can in some cases outperform LDSC. See out tutorial HERE
v0.0.3d Code update: The most recent code update is dated April 12th, 2021. This code update includes a statistically equivalent, but far more efficient, way of calculating model chi-square. In the context of a userGWAS model where model chi-square was requested this may decrease run-times by ~50%. We recommend reinstalling to the most recent update of GenomicSEM. However, please note that changes in Genomic SEM defaults are likely to produce slight changes in results relative to previous versions. For further details, see patchnotes

PGC worldwide lab meeting on genomicSEM

Click below for a video which provides a very clear introduction to the method/paper:

Contents of the wiki:

Learn how to install GenomicSEM

Consider some of the nuances of summary data, and know where to find summary data.

Fit SEM models to GWAS summary data without a SNP

Run a GWAS where the SNP is included in the structural equation model.

Estimate functional enrichment for any parameter in a Genomic SEM model (e.g., factor variances).

Run multivariate TWAS using T-SEM

Installation:

We assume you are running R 3.4.1 or newer. We guarantee no backward or forward comparability. If something breaks please raise the issue on GitHub and we will try and fix it ASAP.

First, you need to install the devtools package. You can do this from CRAN, launch R and then type

install.packages("devtools")

Load the devtools package.

library(devtools)

Now you are ready to install the latest version of GenomicSEM. Note that this will often raise 24 warnings about replacing previous imports; these warnings are safe to ignore.

install_github("GenomicSEM/GenomicSEM")

That’s it! You are ready to start using GenomicSEM

Parallel performance on Linux

In some instances of R on Linux a parallel backend is automatically configured which by default uses the maximum number of cores (i.e. 1 thread per core).
It is unclear to me if this is base R or some other package that’s doing this in the background.
In GenomicSEM parallel and foreach are used to manage parallel operation which seem to work outside this scope.
Combined, this causes the creation of far too many threads, precisely: cores argument * number of actual CPU cores.
For example, a 16-core machine, with cores=15 would spawn 16*15=240 R threads.
This in turn causes CPU congestion, and causes a very significant drop in performance, especially in high core-count machines. In testing on a 256-core machine, 100K SNP userGWAS took 1,5 hours without limiting this backend, compared to under 10 minutes with limits.

How to change this may depend on your Linux and/or R build, but as a catch-all you can use the following prior to running R:

export OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 MKL_NUM_THREADS=1 NUMEXPR_NUM_THREADS=1 VECLIB_MAXIMUM_THREADS=1

Note that this may change behavior of other programs or packages as well, so it is recommended to do this in a separate session, or from a separate script, e.g. create RunGSEMAnalyses.sh:

#!/bin/bash
export OPENBLAS_NUM_THREADS=1 OMP_NUM_THREADS=1 MKL_NUM_THREADS=1 NUMEXPR_NUM_THREADS=1 VECLIB_MAXIMUM_THREADS=1
/usr/lib/R/bin/exec/R --no-echo --no-restore --file=MyGSEMAnalysesRscript.R --args argument1 arument2

Best performance is achieved by leaving the values for these backends at 1 and maximizing the number of cores in GenomicSEM (only bound by CPU or RAM constraints).

Detailed test results can be found in patchnotes.

License

This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.

This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with this program. If not, see https://www.gnu.org/licenses/.