LW: did you look into the torsions affecting the ring pucker?
AMI: not yet
DLM: ring pucker is a complicated interaction between angle and dihedral parameters, so it’s not shocking that the pucker would change on re-fit. Not sure if we should be checking if pucker is part of the fit somehow. Probably part of why we see it move around so much is that we’re not fitting to it, and it may not affect the fitting targets so much?
AMI: I would imagine it would affect the RMSD between QM and MM structures
DLM: doesn’t look like, e.g. distant R groups, change much when geometry changes
DLM: how big a deal is incorrect ring pucker?
PB: we do have some issues with ring puckering. Spirocycles are an important class that would be affected by puckering
LM – LW and I weren’t sure about releasing since the 4-membered rings got so much worse. Interested in feedback/ideas.
PB: at one point cyclobutane was planar, that should also be checked again
LW – IIRC, one direction we thought about looking at is splitting the angles in a C4 ring from those in 4-membered heterocycles(?)
LM – Worth trying, we’d looked at something similar but with the info I have I’m not sure it’ll help
LW – One tricky thing is that the good performance we see on r4 in sage 2.1 is because we relaxed the contraints on r4 values. I’m against arbitrarily pinning it to the value that we want, but if there’s something we can do to make it data-motivated that could be good.
BW – How many targets are we looking at here?
LM – I’d need to check.
JW: do we get non-physical angles because sometimes these angles would be really obtuse or acute?
AMI: in my previous experiments this wasn’t really an issue, but possibly that’s because it was allowed to fluctuate more
AMI: PB, how come some of the sulfonamides have good angles but some of them optimize to the weird geometry? Did you notice any unifying feature between the separation?
PB: I think it’s the sulfamides, a subclass of sulfonamides, that are optimizing poorly
CC: it looks to me like the Os in Sage-2.1.0 look weird relative to the nitrogen, makes me wonder if the angle of the N-S-O angle is driving this
JW: I get the sense that when N-S-O angle starts becoming smaller, it may be stronger than the O-S-O, and everything gets pulled apart
TG (in chat): IIRC it's due to torsions, I think some had wrong phases and other wierdness. For example that first bad molecule looks like it has a torsion that wants to push to 0 deg
PB: hydrogens in the 2.2 are still in a weird angle, still pointing cis to the O vs trans in the QM
JW: I’d be interested in the energy/force numbers for this angle, my impression of angles is that they’re much more stiff than torsions, and that moving away from equilibrium should have a much higher energetic penalty than twisting the torsion
DLM:
LW – Not sure release should be made with r4 degradation. There are several options to solve this and the right one will depend on how much time we have.
DM – People are pretty vocal about wanting a sulfonamide fix, may be worth taking a hit on r4. But it’s ultimately your call.
LW – Doing a refit with the r4 parameter split would take about a week, so we’ll know soon.
DM – That makes sense. It might also be acceptable to just fix the r4 stuff in a future point release.
LM – I could do two fits in tandem: one without fitting new values for r4, one with splitting the r4 param. Then we can see how they look.
LW – Please go ahead and do so on HPC3.
BS – AGobbi spent a lot of time looking at relative energies of ring conformers and barrier heights. He found that basically all FFs were getting it wrong, though he didn’t have access to OpenFF . He diagnosed that the angle term was too stiff and it was messing with the energetics of ring puckers.