Downsampling of cells in scRNAseq DE analysis

Question

[mytp]

Hello,

This is a general question with no specific example concerning differential gene expression analysis in scRNAseq data.

We sometimes want to compare two conditions in our DE** analysis but we see the the sample size of these groups greatly differ.

For example, we compare control (n=1500 cells) vs LPS (n=240 cells).

Someone suggested to me to downsample the cells, so at the end one would have a similar / equal number of cells in each group.

I didn't find any information whehter this is something that is acceptable to do, and I am a bit unsure about the approach, or how I can perform it well, or whether there's a nother way to perform the comparison.

I would greatly appreciate any ideas/comments/suggestions !

** We perform DE analysis using FindMarkers/FindAllMarkers functions in Seurat with Wilcoxon Rank Sum test.

[ruchikabhat31]

any guidance in this regards

Answer 1

[fracarb8]

I would avoid downsampling as you might remove some interesting data.

In your case, I would use a pseduobulk approach, using AggregateExpression. If you want to use FindMarkers, you cna set return.seurat =TRUE otherwise, you can use those aggregated counts in DESeq2 or edgeR