mashr: Multivariate Adaptive Shrinkage in R

This package implements methods to estimate and test many effects in many conditions (or many effects on many outcomes).

Empirical Bayes methods are used to estimate patterns of similarity among conditions, then exploit these patterns of similarity among conditions to improve accuracy of effect estimates. See Urbut et al for details.

Note: the R package used to generate the results for the manuscript is here.

Quick Start

1. Follow the setup instructions below.

2. See the introductory vignette for an introduction to mashr.

3. Then walk through these vignettes to learn more about mashr: Introduction to mash: data-driven covariances and Simulation with non-canonical matrices.

Citing this work

If you find the masr package or any of the source code in this repository useful for your work, please cite:

Sarah Urbut, Gao Wang, Peter Carbonetto and Matthew Stephens (2019). Flexible statistical methods for estimating and testing effects in genomic studies with multiple conditions. Nature Genetics 51, 187-195.

Setup

Please follow these steps to install mashr.

1. Unlike most packages available on CRAN, mashr is not precompiled, and therefore to install mashr you will need to make sure that your R installation is properly set up to compile packages with C++ source; in particular, the C++ compiler programs supported by your version of R should be installed on your computer, and R should be correctly configured to call these compilers when installing packages from source. For more information, see the CRAN documentation.

2. Install the latest release of the mashr package using devtools:

   install.packages("devtools")
   devtools::install_github("stephenslab/mashr@v0.2-11")

   This command should automatically install any missing dependencies that are available from CRAN. This command should also automatically retrieve and install the latest version of the ashr package from Github. If it does not, you can install the ashr package separately using devtools:

   devtools::install_github("stephens999/ashr")

3. By default, the devtools::install_github function does not build the vignettes. If you would like to build the vignettes as well, you will need to several additional packages, including flashr, that are used only in the vignettes. This can also be done with devtools:

   devtools::install_github("stephenslab/mashr@v0.2-11",dependencies = TRUE,
                            build_vignettes = TRUE)

Developer notes

• When any changes are made to roxygen2 markup or the C++ code in the src directory, run devtools::document() to update the RcppExports.cpp, the package namespaces (see NAMESPACE), and the package documentation files (in the “man” subdirectory),

• These are the R commands to build the website (make sure you are connected to Internet while running these commands):

  pkgdown::build_site(lazy=TRUE, examples=FALSE)

• After editing C++ code in the src directory, please use uncrustify to format the code using configuration file inst/misc/uncrustify_default.cfg. For example:

  uncrustify -c uncrustify_default.cfg --replace --no-backup -l CPP mash.cpp

• To load the package into R without recompiling the Rcpp attributes, run pkgbuild::compile_dll(compile_attributes = FALSE), then run devtools::load_all().

• Prior to submitting the package to CRAN, the following modifications need to be made: (1) remove the Remotes: entry in DESCRIPTION; (2) remove the flash_mash.Rmd vignette; (3) remove “flashr” from Suggests: in DESCRIPTION; (4) Make sure version number is of the form x.y.z.

• For one of the Solaris computing environments, rhub::check(platform = "solaris-x86-patched-ods"), we encountered an RcppGSL linking issue, probably due to symbols that were inappropriately defined in one of the RcppGSL headers. A workaround for this linking issue is to remove #include <RcppGSL.h> from RcppExports.cpp, and move any RcppGSL-related function definitions to extreme_deconvolution.cpp. For an example of what this looks like, see commit 4a41f14.