Dimension Reduction of Non-Normally Distributed Data

Implements a generalized version of principal components analysis (GLM-PCA) for dimension reduction of non-normally distributed data such as counts or binary matrices. Townes FW, Hicks SC, Aryee MJ, Irizarry RA (2019) . Townes FW (2019) .

R package: glmpca

Generalized PCA for non-normally distributed data. If you find this useful please cite Feature Selection and Dimension Reduction based on a Multinomial Model. (doi:10.1186/s13059-019-1861-6)

A python implementation is also available.

Installation

The glmpca package is available from CRAN. To install the stable release (recommended):

install.packages("glmpca")

To install the development version:

remotes::install_github("willtownes/glmpca")

Usage

library(glmpca)

#create a simple dataset with two clusters
mu<-rep(c(.5,3),each=10)
mu<-matrix(exp(rnorm(100*20)),nrow=100)
mu[,1:10]<-mu[,1:10]*exp(rnorm(100))
clust<-rep(c("red","black"),each=10)
Y<-matrix(rpois(prod(dim(mu)),mu),nrow=nrow(mu))

#visualize the latent structure
res<-glmpca(Y, 2)
factors<-res$factors
plot(factors[,1],factors[,2],col=clust,pch=19)

For more details see the vignettes. For compatibility with Bioconductor, see scry. For compatibility with Seurat objects, see Seurat-wrappers.