Check that indices can be specified as expected

[grst]

Description of the bug

Specifying indices should work as expected from a nf-core pipeline, i.e.

• igenomes should be respected
• it shall be possible to specify pre-computed indices and they shall be actually used
• it shall be possible to generate indices from fastq/gtf otherwise.

There are a couple of issues related to that and it's probably best to address them all in one go and add tests where it makes sense.

Related issues:
#390
#339
#345

Command used and terminal output

No response

Relevant files

No response

System information

No response

[kopichris]

Happy to look into this if the issue is still open 👍

[grst]

It would be great if you could take a look, but it's better to wait until #369 is merged because @fmalmeida already made some changes to this (but without the aim to address this entirely).

If you already want to get started, you could also try out the branch from #369 - It would anyway be good to get some additional feedback on this because it entails quite some changes.

[kopichris]

Happy to play around with the #369 branch until it gets merged, thanks for the heads up!

[grst]

#369 is merged now. I actually think that most/all points are adressed above. @kopichris, would you be up for testing the behavior and check if there are still some pain points?

[kopichris]

That's great news! I'll test the behavior of #369 and report back by Friday!

[kopichris]

Hi @grst, I reviewed and tested the following issues from the dev branch:

[#330]: I ran the test,conda profile and received the following error in the NFCORE_SCRNASEQ:SCRNASEQ:H5AD_CONVERSION: ANNDATAR_CONVERT process: Error in library(anndataR) : there is no package called ‘anndataR’. I suspect this is because there is no conda recipe in the ANNDATAR_CONVERT process.

[#339 and #345]: Pipeline completed successfully when specifying the --fasta and --gtf parameters.

[#366]: Pipeline completed successfully when using the test_full,docker profile with the --skip_emptydrops parameter. Pipeline failed on time when running with emptydrops. This was likely because the emptydrops workflow is optimized for GPUs, which I don't have on my system. It may be worth skipping the emptydrops workflow by default in the test_full profile.

Anything else you think we should look at?

[grst]

I ran the test,conda profile and received the following error in the NFCORE_SCRNASEQ:SCRNASEQ:H5AD_CONVERSION: ANNDATAR_CONVERT process: Error in library(anndataR) : there is no package called ‘anndataR’. I suspect this is because there is no conda recipe in the ANNDATAR_CONVERT process.

Yes, this is expected as anndataR uses a custom container for now. So running with the conda profile is currently not supported (same as for cellranger).

[https://github.com//issues/366]: Pipeline completed successfully when using the test_full,docker profile with the --skip_emptydrops parameter. Pipeline failed on time when running with emptydrops. This was likely because the emptydrops workflow is optimized for GPUs, which I don't have on my system. It may be worth skipping the emptydrops workflow by default in the test_full profile.

That's a good point! Cellbender should still finish on CPU if one throws enough cores and time at it, but I think you are right and we should disable it at least for the test profile.