The susieR package implements a simple new way to
perform variable selection in multiple regression (\(y=Xb+e\)). The methods implemented here are
particularly well-suited to settings where some of the X variables are
highly correlated, and the true effects are highly sparse (e.g. <20
non-zero effects in the vector \(b\)).
One example of this is genetic fine-mapping applications, and this
application was a major motivation for developing these methods.
However, the methods should also be useful more generally.
The methods are based on a new model for sparse multiple regression, which we call the “Sum of Single Effects” (SuSiE) model. This model, which will be described in a manuscript in preparation (Wang et al), lends itself to a particularly simple and intuitive fitting procedure – effectively a Bayesian modification of simple forward selection, which we call “Iterative Bayesian Step-wise Selection”.
The output of the fitting procedure is a number of “Credible Sets” (CSs), which are each designed to have high probability to contain a variable with non-zero effect, while at the same time being as small as possible. You can think of the CSs as being a set of “highly correlated” variables that are each associated with the response: you can be confident that one of the variables has a non-zero coefficient, but they are too correlated to be sure which one.
The package is developed by Gao Wang, Peter Carbonetto, Yuxin Zou, Kaiqian Zhang, and Matthew Stephens from the Stephens Lab at the University of Chicago.
Please post issues to ask questions, get our support or provide us feedback; please send pull requests if you have helped fixing bugs or making improvements to the source code.
Install susieR from CRAN:
install.packages("susieR")Alternatively, install the latest development version of
susieR from GitHub:
# install.packages("remotes")
remotes::install_github("stephenslab/susieR")See here
for a brief illustration of susieR. For more documentation
and examples please visit https://stephenslab.github.io/susieR
If you find the susieR package or any of the source code
in this repository useful for your work, please cite:
Wang, G., Sarkar, A., Carbonetto, P. & Stephens, M. (2020). A simple new approach to variable selection in regression, with application to genetic fine mapping. Journal of the Royal Statistical Society, Series B 82, 1273–1300. https://doi.org/10.1111/rssb.12388
If you use any of the summary data
methods—susie_suff_stat or susie_rss—please
also cite:
Zou, Y., Carbonetto, P., Wang, G. & Stephens, M. (2021). Fine-mapping from summary data with the “Sum of Single Effects” model. bioRxiv https://doi.org/10.1101/2021.11.03.467167
When any changes are made to roxygen2 markup, run
devtools::document() to update package
NAMESPACE and documentation files.
Run pkgdown::build_site() to build the website.
Getting pkgdown to work properly can be frustrating due to
numerous & fragile dependencies. If pkgdown does not
work for you out of the box you can use this docker command
to run all vignettes and build the site:
docker run --rm --security-opt label:disable -t -P -w $PWD -v $PWD:$PWD \
-u $UID:${GROUPS[0]} -e HOME=/home/$USER -e USER=$USER gaow/susie \
R --slave -e "pkgdown::build_site(lazy=TRUE, examples=FALSE)"