Shrinkage Covariance Incorporating Prior Knowledge

Implements estimation methods for shrinkage covariance matrices using user-specified covariance targets. The covariance target is a structured matrix towards which the unbiased sample covariance is shrunk, optionally incorporating prior knowledge. Shrinkage intensity is computed analytically. The method is described and applied to microarray gene expression data in Jelizarow et al. (2010) .

SHrinkage covariance Incorporating Prior knowledge

The SHIP package implements the shrinkage estimator of a covariance matrix given any covariance target, such as described by Schaefer and Strimmer in 2005. In addition, it proposes several targets based on biological knowledge extracted from the public database KEGG.

To use the shrinkage estimator, one should just have at hand a data set in the form of a $n \times p$ matrix, and a covariance target.

If one wishes to use the proposed targets, the data set should be compatible with KEGG, i.e. it should be possible to extract for each gene the pathways it belongs to. This information, for example, can be found in libraries such as hgu133plus2.db.

Installation

Use the following command to install the CRANN version of ‘SHIP’:

install.packages("SHIP")

Alternatively, you can install the development version of SHIP from GitHub with:

# install.packages("devtools") # If necessary
devtools::install_github("vguillemot/SHIP")