ENH: tpx 137 (GROMACS 2025.0) support

[tylerjereddy]

• Add support for GROMACS 2025.0 .tpr version 137, which mostly involves accounting for the new PyTorch neural network potential field, F_ENNPOT. Most of the other changes are similar to the approach I previously used for bumping .tpr version support in ENH: support tpx 134 (GMX 2024.4) #4866.

• Some other .tpr additions are noted at GROMACS 2025 TPR support #4888, but we don't appear to need to use those within the context of our current .tpr parsing tests. It may be sensible to occasionally generate .tpr test files that use those extra bells and whistles to see how we fare, though as a first pass here the full testsuite passes for us if we simply do the usual gmx convert-tpr at the latest stable GROMACS release tag v2025.0 and propagate our current TPR testing to additionally use the converted files.

PR Checklist

• Issue raised/referenced?
• Tests updated/added?
• Documentation updated/added?
• package/CHANGELOG file updated?
• Is your name in package/AUTHORS? (If it is not, add it!)

Developers Certificate of Origin

I certify that I can submit this code contribution as described in the Developer Certificate of Origin, under the MDAnalysis LICENSE.

---

📚 Documentation preview 📚: https://mdanalysis--4919.org.readthedocs.build/en/4919/

[tylerjereddy]

For this and other new v2025.0 bells and whistles, cc @al42and for possible generation of new test files that have the new bells and whistles.

I don't now if we really need a new test file with i.e., the new PyTorch potential handling + reference coordinate scaling with multiple centers of mass + init-histogram-counts to enable continuation of expanded ensemble runs.

Probably would be good to have eventually, if only to make sure we don't need to skip over any other bits of information when "seeking" through the file.

[IAlibay]

Do we think there are attributes we will be missing if we're not properly parsing the NN potential things? i.e. what I'm asking is if we need a warning here.

[al42and]

For this and other new v2025.0 bells and whistles, cc @al42and for possible generation of new test files that have the new bells and whistles.

I can make a TPR with a bunch of things turned on, but there are issues if we want to stick with existing test sets:

• 2lyz: they only have TPR, which is neither editable nor "decompilable" into parts; can't change settings there (and looks like the original receipt is lost in time, since the new are just converted from the old ones, per 336ff1d).
• bonded/dummy: currently, for NN, we need something with "physical" element names.

If that's ok, I can create a new TPR roughly based on GROMACS's own tests.
Reconstructing 2lyz and making a test based on it is doable, but I'd rather avoid that as long as you're ok with adding a new test system.

[codecov]

Codecov Report

Attention: Patch coverage is 0% with 2 lines in your changes missing coverage. Please review.

Project coverage is 93.61%. Comparing base (596cd7c) to head (cd912fd).
Report is 25 commits behind head on develop.

Files with missing lines	Patch %	Lines
package/MDAnalysis/topology/tpr/utils.py	0.00%	2 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop    #4919      +/-   ##
===========================================
- Coverage    93.64%   93.61%   -0.04%     
===========================================
  Files          177      189      +12     
  Lines        21857    22925    +1068     
  Branches      3077     3078       +1     
===========================================
+ Hits         20469    21461     +992     
- Misses         938     1013      +75     
- Partials       450      451       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[IAlibay]

I'm not very versed in the new 2025 changes, but this looks reasonable to me, thanks!

[tylerjereddy]

Reconstructing 2lyz and making a test based on it is doable, but I'd rather avoid that as long as you're ok with adding a new test system.

@al42and I think making a new .tpr unrelated to our current testing .tprs would still be useful if it has the new bells and whistles turned on, especially if you could keep the file size suitable for a testsuite (i.e., similar file size range to what we currently use). Presumably we could at least use gmx dump to get some ground truths for sanity checking our parser.

[al42and]

Hi!

Here's an example file + test. The NN potential is stored as a reference, so there's not really much to test except that it does not screw up anything else. Also enabled a few other random features in the MDP. The file size is 11KiB (that's just ALA dipeptide in vacuum)

e6e8bb5

[orbeckst]

Having the capability to read the basic features for the new format (as demonstrated in the PR on the converted TPR files) is "good enough" for getting this into the 2.9.0 release #4939 . Given the time constraints I'd suggest to follow up with further work in additional PRs.

If we're worried that other things really won't work well then we could add a note to docs/release notes. @tylerjereddy I'll leave this to your judgement.

[tylerjereddy]

I'm at the airport in Austin, but just pulled in Andrey's commit and the testsuite seems to be happy locally even with the fancy new TPR, so I went ahead and pushed that up alongside Andrey's addition to the CHANGELOG.

So, no known reservations on my end, may even be able to remove the residual TODO, though that doesn't hurt

[IAlibay]

Amazing, thanks @al42and and @tylerjereddy !

[IAlibay]

black linting (couldn't push from the CLI for some reason)

[IAlibay]

CI was passing earlier and now that linters are I'm going to go ahead with the merge.