That gets linked to the person who fills the form, so it's tricky for the deployment. I'd say that it doesn't fit the licensing scheme of nf-core

That's what I thought. When implementing MAJIQ, I had the Sentieon implementation in nf-core/rnaseq in mind, as Sentieon also requires users to obtain a separate license to run the software. However, I'm not sure how they handle deployment there. My supervisor is in lose contact with the MAJIQ developers, so I'll ask her whether they would be willing to offer a licensing solution that is compatible with nf-core.

Great! We need to discuss this before implementing though as it's a bit border-line.

I will notify you here when I get a reply 👍

Happy to help with the discussion here. Some tools in mhcquant (( I think ) also have this issue. @Jonas S might be able to share some relevant code snippets on how he worked around this

Hi all, so. .. @felix.haidle615

IRFinder-S:

• Not available on Conda, but the developers provide a working public Docker image.

• Has a hardcoded minimum RAM requirement of 32 GB, which prevents it from running on standard GitHub Actions runners.

Looks good, should work with singularity as well 👍 Make sure to throw an error if someone wants to run this with

-profile conda

. Also exclude this tool from running on CI but include it in

test_full

, which will be run on AWS

MAJIQ:

• Not available on Conda, and there is no official public container image.

• Can be installed from its GitHub repository. In my current implementation, I build a custom Docker image in CI before running the pipeline.

• Requires a license, which I currently handle using Nextflow secrets.

I just quickly checked it, looks good. Maybe a user-friendly alternative would be to let wave build the docker image upon pipeline execution. That should work during CI as well. The license should still be parsable via nextflow secrets. I used wave in this module for your reference: github.com/nf-core/epitopeprediction/…/mixmhcpred