Out of interest, would it conceptually be possible to support a sparse matrix such as CsparseMatrix (as commonly used in Seurat and SingleCellExperiment) in estimateDisp rather than expanding it to an ordinary matrix during DGEList creation?
Hello-- I am a bit familiar with the preface of pseudobulk differentiall expression analysis, however cannot find succinct tutorials for newbies concerning edgeR for scRNA-seq. I was wondering if anyone was familiar with a tutorial taking a Seurat scRNA-seq object all the way to edgeR analysis-- volcano plots and generating the list of DEGs?
Thank you in advance!
3 answers
See Section 4.10 of the edgeR User's Guide. The Case Study is titled "Single cell RNA-seq differential expression with pseudo-bulking" and it gives complete code for conducting an edgeR pseudo-bulk analysis starting from a Seurat scRNA-seq object, so it seems exactly what you're after:
https://bioconductor.org/packages/3.17/bioc/vignettes/edgeR/inst/doc/edgeRUsersGuide.pdf
The code in the above edgeR case study already fully supports Seurat sparse matrices. The case study converts the Seurat object directly to pseudo-bulk counts without ever expanding the single-cell data to a full matrix. estimateDisp() is only run on the pseudo-bulk counts.
Note that the case study requires edgeR for Bioconductor 3.17, which will be the official Bioconductor version in a week or two.
This is a very good resource single-cell best practices. I find this tutorial provided by in the decoupler package (python) to be good as well.
Log in to answer this question.