Support OpenFF Toolkit v0.11+

[Yoshanuikabundi]

Description

Update BespokeFit to work with new versions of the OpenFF Toolkit

Todos

Notable points that this PR has either accomplished or will accomplish.

• Add pins for Toolkit v0.11+
• Transition to OpenFF Units
• Update for breaking changes in 0.11+
• Check optimize.err for errors in force balance handler
• Set minimum versions of toolkit, qcsubmit, and qcengine once they are released (see status)

Status

• Requires Toolkit 0.11 support openff-qcsubmit#204 in a release on conda forge
• Requires Metal ions with unpaired electrons are rejected as radicals with RDKit openff-toolkit#1420 to be fixed in a release on conda forge
• Requires OpenFF Toolkit v0.11.0+ compatibility leeping/forcebalance#259 in a release on conda forge (passes tests without this)
• New ForceBalance release
• Ready to go

[codecov]

Codecov Report

Merging #198 (78a3ea9) into main (3abdcca) will decrease coverage by 0.06%.
The diff coverage is 81.25%.

Additional details and impacted files

[j-wags]

The molecule in test_submit::test_submit_multi_molecule is failing because RDKitToolkitWrapper thinks Cu2+ is a radical. So the toolkit should be changed to accept Cu2+ so that tests here pass.

This also touches on the issue that QCSubmit sometimes puts multiple non-covalently-bound components in a single OFFMol, but while that's still something we need to get rid of some day, that's not the root cause here and so it shouldn't be a blocker to this PR.

So, accepted once the tests pass (which will probably involve merging a Toolkit PR)

[jthorton]

@Yoshanuikabundi I think this test should work with any multi-molecule system maybe something like Cc1ccc(cc1Nc2nccc(n2)c3cccnc3)NC(=O)c4ccc(cc4)CN5CCN(CC5)C.CS(=O)(=O)O imatinib mesylate would do the job to save you waiting on a toolkit PR?

[Yoshanuikabundi]

@jthorton Thanks but the Toolkit PR is already in :)

[Yoshanuikabundi]

Looks like we also need leeping/forcebalance#259, which is currently in limbo. Tests for force balance pass, but trying to actually run an optimization fails. The error message is extremely cryptic and the true cause is difficult to lock down owing to a lack of optimize.err handling in ForceBalanceOptimizer._read_output() and the aggressive clean up of temporary files throughout the codebase. BespokeFit believes FB to have completed successfully when it actually errored out at an early stage, leading to a failure to find result outputs. I'll add in some code to handle this once I've got something ready for the BespokeFit workshop.

[mattwthompson]

Tests are starting to come in with some green checks - if anybody wishes to once it over again I'll give a window of a few days before merging (and another window for checks prior to making/pushing for a release).