Is there any minimum number of counts in scRNAseq to consider that a gene is expressed?
0
1
Entering edit mode
26 days ago
ev97 ▴ 30

I am quite new in the scRNAseq field and I am wondering if there is any minimum number of counts of which we can consider that a gene is expressed too much or too little in a particular condition.

I am asking this particular question due that I am used to work with bulk RNAseq data and I usually consider a gene is worth to focus in if the expression is quite high (>=1000 counts for example if we have genes that they have < 100).

I am aware that I cannot make the same assumptions of expression in scRNAseq cause I am looking at particular cells and maybe a expression of 20 it is enough to infer that a X gene is expressed more in one cell than another. However, I don't know if there any guidelines or something that people usually do/consider in terms of expression (apart from that if you have cells with a expression = 0 you should remove them).

scRNAseq single-cell seurat scanpy UMI • 294 views
2
Entering edit mode

I am asking this particular question due that I am used to work with bulk RNAseq data and I usually consider a gene is worth to focus in if the expression is quite high (>=1000 counts for example if we have genes that they have < 100).

That is wildly too stringent to me, and also arbitrary. There is imho no reliable justification to make such a statement. Counts depend on total sequencing depth, and even after normalization longer genes in full length RNA-seq have more counts than shorter ones, so a simple (and large) cutoff like this is not getting you anywhere, other than enriching for genes with large counts.

In scRNA-seq one often says that "at least 1 count" qualifies as "expressed" for QC, e.g. total number of expressed (or rather detected) genes, but if you really want to infer biology this is as arbitrary as the above cutoff.

I try to avoid the term "expressed" if possible as you really cannot tell, and just do differential analysis between groups, and then call it "over", "under", "not differentially"-expressed.

Just do what you can justify and what serves your analysis.

0
Entering edit mode

Thanks a lot for the feedback! I will take it into account.