Current nf-core and nf-test limitations

[AlexVCaron]

There is a few problems / hurdles / efficiency issues using nf-core right now. Let's use this issue to list them and discuss.

• git-remote and branch in commands to target nf-scil
• nf-test stacktrace is hard to understand, not human readable

[arnaudbore]

• lint fail when using params.scilpy_version

[arnaudbore]

generic tag for docker release

[AlexVCaron]

Talking points on NF-TEST

tests infrastructure (asserts and data managers) for medical imaging

• there is an assertion for gzipped files, need to test if it solves md5 problems
• if we want to add assertions on image data (msse, variance, ect.), everything in nf-test for us !
  • plugin framework
  • libs can be placed in test directories, they hotpatch nf-test

[AlexVCaron]

Talking points on multiqc

• it's an aggregator, can do some population-wise analysis, less powerful subject wise
• uses template defined for bio-informatics, which are default
  • we'd need to create templates for generic Neuroimaging QC
    • not sure, we're not there yet
    • maybe the template can be nothing more than an image viewer ?
      • anything else plotted or analyzed with stats integrates ?
• QC measures need to be generated per subjects priorhand
• there is a plugin/extension infrastructure to hotpatch multiqc with our stuff

[AlexVCaron]

Talking points on NF-CORE

how to make the CLI lighter for custom module/subworkflow repos

• dependency resolve is a no (Paolo DiTomasso veto), better have one thing used only, if it is updated somewhere, should be updated everywhere
  • modules and subworkflows can have the same name, in different org_path, if needed
• Lean down CLI for remote component repositories nf-core/tools#2993
  • right now, best way is through environment variables, all opts have a NFCORE_ equivalent
  • looking up what is standard in python, think it's using XDG_CONFIG_HOME

metadata and usage for task.ext options

• task.ext are bout to disappear, should be translated as inputs
  • REQUIRE optional inputs to be usable, else module calls get monstruous

integration of nf-scil in nf-core

• We now have a Special Interest Group (SIG) in nf-core (link to come), where we'll discuss how we integrate
  • the idea is to have our modules and subworkflows in a groups aside nf-core
    • located in the modules and subworkflows directories, aside of the nf-core directory
  • First thing is to PR a module and subworkflow to nf-core to create the directories and framework