Pseudobulk DE gene Analysis in scverse ecosystem

Hi Everyone,

I was wondering if there is a way to run a pseudobulk DE analysis within scanpy/scverse?


Hi @wbrett87

Recently we have added a vignette in decoupler showcasing how to do this in the scverse: Pseudo-bulk functional analysis — decoupler 1.4.0 documentation

Hope this is helpful!

Hi @PauBadiaM

Thanks for getting back to me! I tried using Decoupler, but I ran into issues because I dont have replicate samples. Are replicates always required for this kind of analysis.

Oh I see. But then, may I ask, between which groups of samples do you plan on applying DE analysis if no replicates are available?

So I have three “conditions” (single cell experiments from the same tissue from mice with different genotypes). I would like to do DE between cluster A from condition 1 with cluster A from condition 2 and 3. Sorry if my question is naive!

I think I might have been confused about the use of the term replicate. When we refer to replicates, are we talking about biological replicates? Or genetically different samples considered as “replicates”?

It’s okay, this forum is the right place to ask these questions :wink: . Then, I take that you have 3 different samples, one for each condition, and you want to compute DE analysis between them, is that right? Unfortunately, if no replicates per condition are present you cannot perform DE at the pseudobulk level. One strategy could be to build metacells, using SEACells for example, and then treat those as replicates, but they will still hold the same variability of the sample they come from.

What I meant before is, inside each condition, do you have more than one sample? Or is it n=1? If n >= 2 inside each condition, you can apply pseudobulk DE analysis.

So inside each condition we pooled tissue from three separate mice. Unfortunately, we did not hash the cells, so there is no way to separate them into replicates.

Unfortunately, pseudobulking for DE analysis is not possible in this situation. I would still try to generate metacells per condition and then perform the DE on those, this way at least you work with less sparse data than with pure single cell. Good luck!

Ahh I see! Thanks for the tips, I will definitely check out SEAcell.