Regressing variables for finding Differentially expressed genes

Question

[AB]

Hi,

I have a single cell RNASeq dataset with 5 healthy and 5 diseased samples and I want to find genes that are differentially expressed in each of the celltypes. Now unfortunately, all of the diseases samples are male and and of the 5 healthy controls, 2 are female and 3 are male. Now when i do FindMarkers, I see a gender bias and genes like XIST pop up. I tried using Sex as latent.vars argument in FindMarkers as mentioned in here ,here and here, but some of the sex specific genes still pop up as differentially expressed. Should I scale the data using Sex as vars.to.regress and recluster ?

Appreciate any advice on how to approach this. Thanks!

Answer 1

[Bastien Hervé]

What species are you working on and which cell types are you interesting in ?

As you don't have female disease samples it is complicated to make a fair comparison.

I would suggest you to use pseudobulk approches to do your DGE with either DESeq2, EdgeR, limma-voom... Something like this

In your design formula you can add sex as covariable : design = ~ condition + sex

However, it is highly probable you will get sex related genes as differentially expressed again.

Another solution would be to do a DGE within your controls only (male vs female, design = ~ sex). Save the list of DEGs and substract them from your first list of DEGs (condition + sex)

Adding vars.to.regress to the scaling will not change anything as the DGE is done on raw counts. Reclustering will only potentially modify your cell type annotation, which, I guess, is not specific to sex.