This repository provides a conda environment definition to install the R version of 'PEER' (see here for details on PEER) alongside 'tidyverse' and a simple R PEER wrapper script.

UPDATE: We also provide a Dockerfile to create a docker/singularity/charliecloud/... image. We found that using the conda env for calculating PEER factors might result in very slow behaviour, so we recommend using the Dockerfile

Background

PEER is a factor analysis method which, despite being relatively old, is still being used throughout scientific literature. Installation of PEER can be tricky, which is why I provide this conda env definition including R-based PEER and tidyverse as well as the Dockerfile.

Installation

conda

As usual with conda, you can simply create a new environment using the provided 'rpeer.yml' or 'rpeer_explicit.yml' file (the 'explicit' file is a direct export of my working conda env):

~: git clone https://github.com/jhawe/rpeer.git
~: cd rpeer/
~: conda env create -f rpeer.yml

Please note that specifying r-cli in the env file is necessary in order for tidyverse to work properly at the time of writing this.

General note: I usually recommend using 'mamba' instead of the conda command to speed up environment solving. It seems that the original conda command can also run into errors when solving the environment where mamba doesn't.

You can then activate the environment and use PEER as follows:

~: conda activate rpeer

~: R
[...]
> library(peer)
> 
[...]

docker

This respository now provides a PEER dockerfile which can be used to build your own docker/singularity/charliecloud image. The dockerfile includes some basic packages for more advanced data manipulation in R (dplyr, magrittr, readr). The docker image is provided also on docker hub.

Application

Just a few notes on application of PEER (see also the original publication: I provide a simple script to 'correct for peer factors' in this repository (utils_peer.R).

1. PEER assumes a NxG matrix, with N=number of individuals and G=number of genes/molecular entities
2. to be feasible, number of factors to be estimated should not by set above 100 (default: N / 4)