About the way soft torsion are identified about openff-bespokefit

[j-wags]

Hi @xiki-tempula, I think the torsions are selected using a particular SMIRKS pattern. The idea is discussed in the docs here, and the SMIRKS itself is default value of the target_torsion_smirks setting.

This is a particularly complicated SMIRKS - To read it you'll probably need to refer to the Daylight SMILES and SMARTS guides

[j-wags]

Like it wouldn't recognise the H-C-C-H dihedral in cyclopropane or dihedrals in cyclo molecules in general (as these won't go well with torsion scan).

The current value of the target torsion SMIRKS is '[!#1]~[!$(*#*)&!D1:1]-,=;!@[!$(*#*)&!D1:2]~[!#1]'. The part that describes the central bond is -,=;!@, which (I'm pretty sure) means "the central bond is single or double (-,=), and must not be in a ring (;!@)". So it should exclude all ring bonds.

And you don't do any downstream heuristics to do more filtering.

I don't think we do additional filtering, but I have low confidence in this answer, so I might escalate that question to @jthorton (who is out right now, so he may not get back to us on this for a few days).

[jthorton]

Hi @xiki-tempula, @j-wags is correct the SMIRKS pattern he shows is the only thing we use to identify torsions which should be targeted for optimisation and this is a fully customisable option in the workflow, we accept a list of patterns which can describe the target torsions so if you have a case in mind which is not hit by BespokeFit you can add a pattern for it to the list but this may uncover some issues when trying to scan in-ring torsions.

[xiki-tempula]

@j-wags Thank you. Does this SMIRKS pattern captures everything? Like it wouldn't recognise the H-C-C-H dihedral in cyclopropane or dihedrals in cyclo molecules in general (as these won't go well with torsion scan).
And you don't do any downstream heuristics to do more filtering.