What fraction of cells per celltype is recommended to use as subsample_size?

[kaizen89]

Hi!
Thanks a lot for this very useful tool.
I succesfully ran Augur on a large dataset ~70K cells using default parameters, I want to increase the number of celles in subsample_size to 1000-2000 cells to better capture the heterogeneity. I was wandering if you can give any recommendation about the fraction of cells per celltype and per condition that you think would be reasonnable to use.
Thank you!

[skinnider]

I essentially never change this parameter from its default values, but doing so should not impact your results very much. In Supplementary Fig. 6 of the Augur paper, we show that prioritizations are quite robust to the value of subsample_size - the main risk is removing cell types with too few cells from consideration.

[kaizen89]

I guess you are referring to Supp Fig7 (not 6), I saw this part but unfortunately it seems that the dataset tested is a bit small and it's difficult ton conclude whit subsample_size 50-100 .