how to filter the scRNAseq data, by percent genes expression, counts and others

Question

[alwayshope]

Dear experts,

How to filter the scRNAseq data? When using the only keep the genes that have expression in above 10% of all cells, near 8,000 genes left, when only keep genes that have expression in above 50% of all cells, near 4000 genes left.

So which way is better? Use a relatively loose threshold that genes have expression in above 10% of all cells ?

Thank you very much!

Answer 1

[jared.andrews07]

These types of questions have been asked here probably a dozen times. In short, nobody can tell you specific numbers, you must let the data guide you.

Limiting to genes expressed in half of all cells is a very bad idea. I don't see a reason to filter genes beyond the absolute bare minimum. Even the 10% requirement is a bad idea, as what if you have a given population of cells that account for a small fraction of cells (say 5%) that all express it?

Don't throw out data with no reason.

[alwayshope]

Thanks a lot for your kind guidance!

Yes, the caveat for this filter is that it may miss some important genes that expressed at a small fraction of cells...

Seems it's better not to filter out for this extra filter, the tutorial only have filter for nFeature, nCount, Mt ratio which to do the basic filtering.

Best!