I think maybe skip it for small numbers of cells.
My interpretation is that we are running into trouble when the validation set is getting too small... I might expect maybe that the 10 would fail with 0.9 if 6 was failing with 0.6: Maybe it needs at least 3 in the validation set?
I would check with just 10 cells to see which of us is correct, and see if 0.9 works there. Or what the minimum # that does work is. (I'm going to guess 30.)
Okay, you're right. So less than 30 it doesn't work with the default and then anything 30 or above works. Based on your reply, I'll update this to just skip running CellAssign for anything < 30 cells.
We'll want to add this to our celltyping processing information, which should be implemented as a stacked PR on #223 (branch sjspielman/celltype-release).
sjspielman
commented
9 months ago
From https://github.com/AlexsLemonade/scpca-nf/pull/662#issuecomment-1901225922:
We'll want to add this to our celltyping processing information, which should be implemented as a stacked PR on #223 (branch
sjspielman/celltype-release).