I am wondering whether there are best-practices to use edgeR on single-cell (10X) data that were quantified with the Alevin module from Salmon, given that we want to use the inferential replicates that it produces via bootstrapping. I know there exists the catchSalmon
function to calculate a per-transcript overdispersion value. Alevin first of all outputs gene level counts, and second I personally prefer to perform single-cell DE between clusters on the pseudobulk level (given one has biological replicates of course).
Therefore, are there best-practices to import the inferential replicates from Alevin into edgeR and can we "sum" this inferential replicate information per single cell to the pseudobulk level?
Thanks for the prompt reply. The reason I am asking is that 10X is a 3'-tagged rather than full-length protocol and according to the Swish paper taking into account mapping uncertainty might be beneficial for these kind of data, even on the gene level. So the main question would be whether one can robustly "sum" or "transfer" the inferential replicate information per single cell into the pseudobulk.