Fit / Conformer Generation for Long Flexible Molecules

[MaTiHo]

Dear OpenFF-Bespokefit Team,
First and foremost I would like to say thank you for the great work you did with BespokeFit!

I would like to ask for you help regarding a bug/problem I am facing testing out openff-bespokefit:

For long alkyl chains the conformer generation within the fitting workflow throws an error. Most likely it stems from the same problem one faces when using rdkit directly (see for example: rdkit/rdkit#3323).
One solution that seems to work within rdkit is incresing the maxAttempts keyword:
AllChem.EmbedMolecule(molecule,maxAttempts=50000)

I attached the json output as well as the molecule file using the BespokeFit workflow below:

openff-bespoke executor run --file "tmpC11.sdf"
--workflow "default"
--force-field "sage-2.1.0rc.offxml"
--output "acetaminophen.json"
--output-force-field "acetaminophen.offxml"
--n-qc-compute-workers 2
--qc-compute-n-cores 3
--default-qc-spec xtb gfn1xtb none

Thank you very much in advance!

output.txt
tmpC11.txt

[mattwthompson]

I can reproduce your fragmentation error, but not with OpenEye Toolkits installed. Something is going wrong with fragmentation, but I haven't isolated exactly where.

The toolkit can generate conformers for this molecule using RDKit, albeit slower:

In [1]: from openff.toolkit.utils import *

In [2]: from openff.toolkit import Molecule

In [3]: molecule = Molecule('tmpC11.sdf')

In [4]: %%timeit
   ...: molecule.generate_conformers(n_conformers=800, toolkit_registry=OpenEyeToolkitWrapper())
   ...:
   ...:
6.35 s ± 85.4 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

In [5]: %%timeit
   ...: molecule.generate_conformers(n_conformers=800, toolkit_registry=RDKitToolkitWrapper())
   ...:
   ...:
2min 17s ± 3.45 s per loop (mean ± std. dev. of 7 runs, 1 loop each)

I hoped the toolkit would crash on this input since that would make the fix straightforward. Toolkit's call to AllChem.EmbedMultipleConfs might have room from improvements here but that's not obviously the issue.

@Yoshanuikabundi is there some magic happening in the fragmentation pathway? I can only find a tiny amount of magic that might cause something like this, nothing obviously making big changes to the molecule. Molecule.generate_conformers is only even called once from that package.

https://github.com/openforcefield/openff-fragmenter/blob/0.2.0/openff/fragmenter/chemi.py#L96-L100