scrnaseq
scrnaseq copied to clipboard
nf-core
Call empty droplets
Is your feature request related to a problem? Please describe
I think it could be nice to have a step to distinguish empty droplets from actual cells.
As far as I know, Alevin/Kallisto only perform cell calling based on "knee plots", while cellranger implements the emptyDrops algorithm. According to the emptyDrops paper the method clearly outperforms filtering based on knee plots.
Describe the solution you'd like
Implement a process downstream of the aligner subworkflows running the emptyDrops algorithm.
Describe alternatives you've considered
This kind of filtering could be left to downstream pipelines such as #scflow. However, IMO, it would still make sense to have this as a default even when not using scflow for downstream analysis.
Additional context
STARsolo implements the emptydrops algorithm as of version 2.7.8a which can be activated using the --soloCellFilter EmptyDrops_CR option: https://github.com/alexdobin/STAR/blob/master/docs/STARsolo.md#emptydrop-like-filtering
I don't know if emptyDrops is still state-of-the-art of if there's something more advanced by now.
See comments by Rob Patro on Slack: https://nfcore.slack.com/archives/CHN5BV5DW/p1643209151035000
Hi all- FYI there is an 'dropletutils-scripts' package in Bioconda that will provide a container to do this already.
The scripts themselves are here: https://github.com/ebi-gene-expression-group/dropletutils-scripts
Here's Nextflow process that uses it (admittedly DSL1). Just thought I'd point this out- might be a shortcut to a module.
I will re-start the PR that has been on work on that since it was too problematic. And I will use the suggested: 'dropletutils-scripts' from bioconductor.
Using this, we were able to have it in a private copy of the pipeline. I can then work in bringing the script and the module to here.
I have a PoC running for the main aligners. Once I figure it out for cellrangerarc I will open a PR with the whole description of the changes for discussion.