SCT Normalization for Reanalysis of a subset of scRNA-seq Data

[xyang2uchicago]

Hi,

I'm reanalyzing a large public scRNA-seq dataset containing approximately 100 samples. After subsetting cells of interest from each sample, I'm working with the raw read counts for these specific cell populations.

Following subsetting, the number of cells per sample varies considerably, ranging from 5 to 300. I'm considering SCT normalization to account for potential sample bias.

Specifically, I'm interested in using the SCTransform function with the vars.to.regress=c('initial.ident') argument to regress out sample identity as a source of bias.

My question is:
Are there any limitations on the minimum number of cells per sample (initial.ident) for reliable application of SCT normalization in this context?

Thank you for your time and insights.

Holly