Setting a threshold on gene expression for subsetting (scRNA-seq)

Question

[carolofharvest]

I want to subset a specific cell type by using cell markers. Should I take cells with those markers above 0 after normalization? Considering the nature of single-cell data, could this method fail to capture all the cells I want?

[ATpoint]

That depends on the data and the marker. Generally I would plot violins per cluster for the marker(s) and see whether one robustly overexpresses the marker. Mariers are often enriched rather than specific so > 0 is often not robust.

[LauferVA]

Generally, I cannot advise this in the abstract. The thing is, depending on the treatment group and so forth, there are some transcripts that, while never really produced at high levels, may nevertheless be produced with a high degree of specificity (in particular in an activation state dependent context).

So it is going to be tenuous to do a blunt thresholding procedure - sure you may get rid of noise, but you also may get rid of a really strong signal too - you just don't know. I'd try to start with the biology of interest and what is known and work outward from that point.

Good luck! VAL