scRNAseq analysis with a few samples (n=2 per group)

Question

[Song]

Hi all,

I was performing scRNA-seq analysis with only a few samples (n = 2 per group). This was due to financial limitations in our lab. However, I was wondering if I could still perform reliable differential gene expression (DEG) analysis and proportional comparison analysis with these data. It seems that at least three samples per group are required for reliable statistical results.

The total number of cells is sufficient (> 60,000).

Any suggestions would be appreciated.

Sincerely Song

Answer 1

[ATpoint]

Your n=2 is enables pseudobulk aggregation, and after that you can run typical bulk testings such as DESeq2 or limma. That is in my opinion and experience almost always better than treating individual cells as replicates. Basically, you would sum up the counts of all cells per sample and group/cluster/celltype/whatever, and then go down the normal bulk testing regime. Does that make sense to you?

[Song]

That makes perfect sense. Thank you for clarifying, your explanation was very helpful.

[carolofharvest]

can we also create fake pseudobulks (to mimic biological replicates) from one single cell data by doing random cell sampling and then make the comparison with deseq ? in that case how can i understand how many fake pseudobulks that i need for enough statistical power ?

[ATpoint]

In theory you can, but as you say, it's made up aka fake, so what's the point? You have two samples. Two vs two is not super powerful, but at least it's real. I mean, after all if you get no DEGs at all with that youn can (for hypothesis generation) lower the FDR cutoff, or fall back to something like geneset testing rather than strict per-gene analysis. It's all better than making up data.