Can I run cellassign on samples independently if there is batch effect present?

Question

[gt]

I am running the R cellassign algorithm on a set of 9 samples. Each sample consists of cells extracted from lungs. When I look at a UMAP plot and color by sample origin, there is a notable batch effect. So, here's my question..

Can I run the cellassign algorithm on each sample independently for cell typing, or do I have to combine the samples and run the algorithm with correction for batch effect?

I have not been able to find any resources online that take the first approach. The reason I am asking this question is because I am unable to run the algorithm on all the samples combined as I get an out of memory error. Any suggestions/resources are appreciated!

Answer 1

[Francesco]

Hi! Whenever you do make independent analysis or aggregate data and analyze them, most cell type assignment tools start from raw counts to annotate cells, and also integration tools (cca, rpca, harmony, scVI) do not modify raw counts/data layer so in my opinion there are no differences between the two approaches. Also for me is a bit confusing the decision about the best approach to integrate different datasets and then perform the cell type assignment. It would be nice if more expert people could tell us something about this topic