SUCCOTASH: Surrogate and Confounder Correction Occuring Together with Adaptive Shrinkage
This package is no longer maintained.
Please see the mouthwash function in the R package vicar.
Description
Let
Y = XB + ZA + E,
for
- Y an n by p matrix of gene expression data with n samples and p genes,
- X an n by q matrix of of q covariates,
- B a q by p matrix of unobserved coefficients for the observed covariates,
- Z an n by k matrix of hidden confounders,
- A a k by p matrix of hidden coefficients for the hidden confounders, and
- E an n by p matrix of independent normal errors with column variances s1,...,sp.
Not accounting for the hidden covariates, Z, can reduce power and result in poor control of false discovery rate.
succotashr fits this model under a two-step empirical Bayesian approach. It places a non-parametric prior on B and jointly estimates B and ZA. The main function is succotash.
Installation
To install, run the following code in R:
install.packages(c("devtools", "SQUAREM"))
devtools::install_github("stephens999/ashr")
devtools::install_github("dcgerard/vicar")
devtools::install_github("dcgerard/succotashr")The following packages are suggested:
install.packages("cate")
source("https://bioconductor.org/biocLite.R")
biocLite(c("limma", "sva"))