medaka

[laraPPr]

• link to support ticket: #2023072860000271 + #2023082360000751
• website: https://github.com/nanoporetech/medaka
• installation docs: https://github.com/nanoporetech/medaka
• toolchain: ``
• easyblock to use: `PythonBundle'
• required dependencies:
  • Python with cffi == 1.15.0
    OR
  • cffi == 1.15.0 #195
  • Tensorflow ~= v2.8.0
• optional dependencies:
  *
• notes:
  • latest version of medaka (currently 1.8.1)
• effort: (TBD)
• other install methods
  • conda: yes (https://github.com/nanoporetech/medaka)
  • container image: yes (link?) / no
  • pre-built binaries (RHEL8 Linux x86_64): yes (link?) / no
  • easyconfig outside EasyBuild: yes / no

[RVerschoren]

We received a similar request, for a more recent medaka version (>=1.6, so 1.8.1 fits the bill). I am looking into an installation in foss-2022a. The intervaltree-python, ont-fast5-api, Racon packages will need an update as well in that case.

[RVerschoren]

Pushed current .eb files to a wip directory. Still having some trouble with the hard dependency on TensorFlow 2.8.

[boegel]

@RVerschoren It may be OK to just loosen up that strict version requirement for TensorFlow 2.8, see also my question to them on this a while ago: nanoporetech/medaka#124

[RVerschoren]

@boegel Thanks for the suggestion! It seems the error is in the sanity check part.

Both commands in the EasyConfig's sanity_check_commands work successfully, but a python -m pip check is also run. That blocks on the tensorflow~=2.8.0 line in requirements.txt.

I made a quick patch for that to accept other versions, and tested that with both TensorFlow 2.9 and 2.11 versions in foss-2022a.
The package now builds and installs a module fine for both. However, executing the make test mentioned in nanoporetech/medaka#124 only seems to run a coverage report in 3 collected tests. That coverage is about 18% of the code, so it fails to clear the 80% COVFAIL threshold they set in the Makefile.

Looking for the test files, I came across your other post nanoporetech/medaka#125. It seems they no longer ship a medaka/test directory and I did not find the CI pipeline they mentioned there (broken link to the walkthrough). I don't see much to test the installation here...

Edit: Requester linked me to the medaka/test/data in the GitHub repo itself. I will check it out.

[boegel]

@RVerschoren Any updates on this? We now also have 2 researchers who would like to see an updated version of medaka

[RVerschoren]

I re-installed with Pysam 0.19.1 (avoiding the version conflict we discussed during the last meeting).

Our requester is trying to test it, but keeps running into memory issues: with 3-day jobs that keep failing 2 days in, this is going slow. Unfortunately, it doesn't seem they have any other test input...

The medaka tests still report a coverage of 89% for me. Not ideal, but it looks to me like it installs and can be used as intended.

That said, I would appreciate additional testing on other sites, or other suggestions or advice on how to proceed.

[boegel]

@RVerschoren That sounds like it ready for tinkering with by researchers who actually need it...

Do you think it makes sense to open a pull request for your medaka easyconfig + whatever else is needed (patch files, easyconfigs for dependencies) to the central easyconfigs repo?

[RVerschoren]

@boegel There is nothing specific to UAntwerp or VSC in it, so yes, it can be shared, both in the VSC as well as the easybuilders-easyconfigs repository.

[RVerschoren]

Sent PR to central easyconfigs, easybuilders/easybuild-easyconfigs#18639

[boegel]

@RVerschoren PR merged, so the corresponding directory in wip branch can be cleaned up, and this issue can be closed

[boegel]

@RVerschoren I noticed there were several more recent releases of medaka, so I've done a trivial update in easybuilders/easybuild-easyconfigs#18651

[RVerschoren]

@boegel Thank you for the update, I missed the last releases.

I cleaned up the corresponding wip directory, so I'm closing this issue.