Partial charge methods fail in clean environment #1911
Comments
|
Thanks for the detailed writeup. I'm unable to reproduce on an ARM mac, but I'm charging my old intel mac to try it there. |
|
Some of our CI should be testing on intel macs, and I'm not seeing any issues there, though
|
|
Unable to reproduce with intel mac on OSX 13.2.1, updating to 14 and will report back. |
|
Thanks! And on my end, this is reproducible (not related to this specific molecule/solvent box). Hit the bug trying to use OpenFE stuff (on real ligands) with @mikemhenry and we initially thought it might be related to RDKit, but it does not seem to be. I'm happy to do additional debugging if you need. |
|
Darn, my intel macbook isn't compatible with OSX 14. If you're up to iterate a bit, could you let me know if this reproduces the issue? from openff.toolkit import Molecule
mol = Molecule.from_smiles('[H][O][C]([H])([H])[C]([H])([H])[C]([H])([H])[C]([H])([H])[H]')
mol.assign_partial_charges("am1bcc") |
|
And if that DOES reproduce the issue, could you run mol.generate_conformers()
mol.to_file('temp.sdf', file_format='sdf')and in the terminal and let me know what happens? |
|
The small snippet does not reproduce the error. |
|
Oh wait! I ran it once and it completed without error. I ran it a second time and it did reproduce the error. Fun. ~/tmp is 📦 v0.1.0 via 🐍 v3.12.4 via 🅒 openff-clean took 10s
❯ python tmp5.py
(openff-clean)
~/tmp is 📦 v0.1.0 via 🐍 v3.12.4 via 🅒 openff-clean took 27s
❯ python tmp5.py
/Users/slochowe/miniforge3/envs/openff-clean/bin/wrapped_progs/antechamber: Fatal Error!
Cannot properly run "/Users/slochowe/miniforge3/envs/openff-clean/bin/sqm -O -i sqm.in -o sqm.out".
Traceback (most recent call last):
File "/Users/slochowe/tmp/tmp5.py", line 3, in <module>
mol.assign_partial_charges("am1bcc")
File "/Users/slochowe/miniforge3/envs/openff-clean/lib/python3.12/site-packages/openff/toolkit/topology/molecule.py", line 2677, in assign_partial_charges
toolkit_registry.call(
File "/Users/slochowe/miniforge3/envs/openff-clean/lib/python3.12/site-packages/openff/toolkit/utils/toolkit_registry.py", line 280, in call
raise ValueError(msg)
ValueError: No registered toolkits can provide the capability "assign_partial_charges" for args "()" and kwargs "{'molecule': Molecule with name '' and SMILES '[H][O][C]([H])([H])[C]([H])([H])[C]([H])([H])[C]([H])([H])[H]', 'partial_charge_method': 'am1bcc', 'use_conformers': None, 'strict_n_conformers': False, 'normalize_partial_charges': True, '_cls': <class 'openff.toolkit.topology.molecule.Molecule'>}"
Available toolkits are: [ToolkitWrapper around The RDKit version 2024.03.5, ToolkitWrapper around AmberTools version 23.6, ToolkitWrapper around Built-in Toolkit version None]
ToolkitWrapper around The RDKit version 2024.03.5 <class 'openff.toolkit.utils.exceptions.ChargeMethodUnavailableError'> : partial_charge_method 'am1bcc' is not available from RDKitToolkitWrapper. Available charge methods are {'mmff94': {}, 'gasteiger': {}}
ToolkitWrapper around AmberTools version 23.6 <class 'subprocess.CalledProcessError'> : Command '['antechamber', '-i', 'molecule.sdf', '-fi', 'sdf', '-o', 'charged.mol2', '-fo', 'mol2', '-pf', 'yes', '-dr', 'n', '-c', 'bcc', '-nc', '0.0']' returned non-zero exit status 1.
ToolkitWrapper around Built-in Toolkit version None <class 'openff.toolkit.utils.exceptions.ChargeMethodUnavailableError'> : Partial charge method "am1bcc"" is not supported by the Built-in toolkit. Available charge methods are {'zeros': {'rec_confs': 0, 'min_confs': 0, 'max_confs': 0}, 'formal_charge': {'rec_confs': 0, 'min_confs': 0, 'max_confs': 0}} |
|
🫠 |
|
@slochower can you try this? |
|
Looks like 5/10. Perfect. I made a slight modification Here's another version: |
|
Okay, so it's Ambertools that is leading to the mess. WHen it fails, it's linked to And this is what happens when |
|
The input file should have the coordinates in it after the |
|
Can you try installing ambertools 22? Hopefully it doesn't cause too much package churn, on my system it looks like this: which for our testing purposes isn't too bad, another option would be to keep ambertools 23, but just install an older build, like before mpi support was added for example. |
|
Oof. Couldn't install with the existing 3.12 environment. Stepped back to 3.11, also dependency conflicts. Stepped back to 3.10, also dependency conflicts... |
|
sigh, how about |
|
Sadly not. Looks like |
|
Is that making a fresh env? Using some hacks (I am on linux) it looks like it should solve on osx-arm64 |
|
Good point. Although 23.3 is still flaky. I can step back to 22 in a new environment. Here's 😰 |
|
okay one more idea, see if adding the |
|
I think that needs something after it? |
|
You are correct! Another option to pass in is |
I don't think this is a problem inside |
|
IIUC, the toolkit doesn't actually write that file, so something before that step is failing? I can't reproduce this behavior this morning using |
Describe the bug
In a new, clean OpenFF environment on an x86-64 Mac (14.4.1), partial charge methods fail. This is either from
antechamberor possiblysqmand probably related to issues that I've seen @mattwthompson talking about with the Amber folks.For this test script:
I'm seeing:
The test cases also fail.
To Reproduce
conda create --name openff-cleanmamba install -c conda-forge openff-toolkittmp.pypython tmp.pyOutput
`conda list`
CC @mikemhenry
The text was updated successfully, but these errors were encountered: