Hopefully a quick question. I am using the R package called Seurat to analyze my single cell dataset. I know that when I am trying to identify cell types after clustering that you should normalize the whole dataset. Then perform differential expression analysis by comparing a single cluster to all other clusters and look at the top differentially expressed genes within each cluster. However, I am trying to run differential expression analysis between two conditions within each cell type cluster. My question is, should I normalize the whole dataset, or normalize the filtered dataset after filtering to a particular cell type cluster? Any help appreciated. Still relatively new to scRNA-seq data analysis.
I think the recommended strategy right now (by most people) is to use a pseudo-bulk approach where you classify cell types (seurat has a good vignette for that), calculate pseudobulk expression levels for each cell-type (there are several options out there for how to do this), then analyze using limma/voom. Pseudobulk values are calculated from the raw reads, then normalized as you would bulk RNA-seq libraries.