1M cells object, can i subset by cell type and create metacells within?

[Gimenagomez2230]

Hi, I have a big object with 6 different cell types. I understand MetacellsByGroups may already considers cell type ADN sample, so there is no need, but it is crashing specially if I try bigger k (also I might want different k per cell type).

So I thought about:

1. subsetting per cell type and running MetacellsByGroups per sample. Would it be correct?

2. should i re run PCA and Harmony in my subsetted object before MetacellsByGroups or should i use the original (all cell types) embedding?

Thank you!!

[smorabit]

Hi, yes it does sound like a good plan to subset by cell type and to re-run your entire clustering pipeline (QC, variable features, dimensional reduction and batch correction etc) on the subsetted object.

Alternatively you could try to run the pseudobulk version of hdWGCNA, which can save a lot of computational time since you won't have to worry about metacell aggregation.