Hi Biostars,
I'm currently working with a scRNA-seq dataset obtained from foetal ovarian tissues. We have 4 replicas : two ovaries at day 22 after fecondation, and 2 ovaries at day 24.
So far, I've merged the two 22 together and the two 24 together. I've clustered the cells and extracted germ-cells in two distinct fresh SingleCell object ( one for each time ).
My problem : I merged the two populations of germ-cells in a new SingleCell object. I want to look for DE genes between the two stages of development. To do so, I'm using Monocle3 linear regression algorithm, accounting for a "sample" variable ( i.e development stage ).
My question : because of the STRONG heterogeneity of single-cell chemical sequencing methods ( droplet here ), is it " correct " to look for DE genes between two cell populations originating from different experiments ? I don't think I can use methods to mitigate batch effect because I fear losing biological signal, so I'm kinda lost.
Thank you for reading me, Simon
If you have biological replicates it's generally recommended to follow a pseudobulk approach for differential expression. See chapter 4 of OSCA.
Thank you very much, I'll dig this part of OSCA
The question as old as statistics itself. Theoretically no, because of potential batch effects all DE genes could be technical rather than biological. In reality no PI would allow to throw data away because of that. Do your analysis, interpret with the necessary case and validate important findings. Batch correction does not apply here as day and batch is nested. Per-gene batch correction is anyway not the same as things like fastMNN/Harmony/Seurat-Anchoring does, that is a common misunderstanding.
Thank you very much !