marionilab / furthernorm2018

Code for further development of the summation-based scaling normalization method, as implemented in the scran package.

Further scaling normalization

Overview

This repository contains code for further development of the scaling normalization methods implemented in the scran package. It is based on the code at https://github.com/MarioniLab/Deconvolution2016, which accompanies the paper Pooling across cells to normalize single-cell RNA sequencing data with many zero counts by Lun et al. (2016).

Setting up

First, install the helper package in package/ with:

R CMD INSTALL package/

... or some variant thereof. It is also worth running:

BiocManager::install(ask=FALSE, version="devel")

... to ensure that the latest versions of all relevant packages are installed.

Simulations

To run the simulations, enter the simulations/ directory and run:

• sim_noDE.R, which simulates a variety of scenarios involving no DE between populations.
• sim_biDE.R, which simulates a variety of scenarios involving DE between two populations.
• sim_multiDE.R, which simulates a variety of simulations involving DE between three populations.

Note that the parallelization framework assumes a SLURM cluster with the Rdevel command (to run a version of R with BioC-devel packages). This can be changed by modifying createBatchParam.R in package/R and/or slurm.tmpl in package/inst/scripts.

To summarize the simulation results, run summarizer.R to cluster the simulation scenarios based on the pattern of errors across all methods.

Real data

To analyze real data, enter the real directory and run:

• zeisel.Rmd, to compute size factors for the Zeisel brain data set.
• pbmc4k.Rmd, to compute size factors for the PBMC 4K data set.