search

MarioniLab / DropletUtils

Clone of the Bioconductor repository for the DropletUtils package.
https://bioconductor.org/packages/devel/bioc/html/DropletUtils.html
56 stars 27 forks source link

emptyDrops not producing the same results as cellranger 3.+ #42

Open anoronh4 opened 4 years ago

anoronh4 commented 4 years ago

Hello,

I've recently tried to "replicate" cellranger analysis by running STAR solo and then empty drops on the unfiltered data. STAR solo gives me the correct (as in, the same as cellranger 3.0) number of unfiltered droplets, but emptyDrops is not so faithful, as i mentioned here: https://github.com/alexdobin/STAR/issues/879

I am getting nearly 6x the count that I got with cellranger 3. Do you have any insight as to what parameters I can tweak in emptyDrops to get the same count? It would save us some time in testing if this was already a solved problem.

I appreciate your time in advance. Have a great weekend!

LTLA commented 4 years ago

Hold on. emptyDrops is not being "unfaithful" to CellRanger. In fact, the opposite is true; CellRanger is unfaithful to emptyDrops. They took inspiration from the algorithm but implemented the general idea in their own way; you can examine the methodological differences on their website.

I guess they had their reasons for changing stuff, and it would have been nice if they made a PR back into DropletUtils to implement a "CellRanger mode" in emptyDrops. But they didn't, and I don't care to have to chase after their development team to get the same results, so here we are.

anoronh4 commented 4 years ago

Didn't mean any offense, but my goal is to replicate cellranger with emptyDrops, not the other way around, thus the choice of words. Anyways, I understand your position and appreciate the response.

LTLA commented 4 years ago

Without further details about your circumstances, the only general advice I can give is to check your knee point estimates and/or filter out mitochondrial and ribosomal genes; see #36.