MT – Any direction on this? Enthusiasm around the team seems lukewarm but I’d like to confirm.
LW – The idea of fitting to interaction energies is very much in the air. This isn’t something that we should put into the standard workflowin the short term. The idea of benchmarking to dimer energies is a good idea - they’re likely to be a good way to evaluate the benefit of vsites. It doesn’t seem like it’d be a huge amount of work, and I’ve done a bit of the workflow myself, and I think it would fit well in the internal benchmarking code.
MT – Two qs:
Scientific: The motivation behind these energies at all is to bridge the gap between single-mol QM and phys prop calcs doing tons of molecules. So this would give info at an intermediate scale on QM. So I wonder what this would provide that phys prop wouldnt
LW – These can help split out LJ and electrostatics terms. There’s also some pretty good data out there (the DESRES sets are quite good).
Practical: In terms of implementation - We don’t have an agreed upon structure in IB, but assuming that the data is out there, is it as simple as doing an MM energy eval given the input coords?
LW – Yes.
MT – Ok, that’s good news. I’m not going to prioritize this in the short run.
JW – I’m also thinking about how to get multi-mol handling in QCSubmit, which would be needed for us making our own training data.
MT – If we start with DESRES sets this shouldn’t be blocking.
LW – DESRES sets are a little messy, it’d be good to have our own processing, but this isn’t crucial.
JW – Earliest I could get this into QCSubmit would be December, so decision isn’t urgent.
LW – InChI vs. SMILES pruning - I wonder if this check could be turned off.
MT - I’ve opened a new ticket on isomorphism pruning - I’ll basically change the default to keep the molecule in the set and emit a warning.
LW – One thing that may have caused the plots to look different is that some of PB’s scripts contained a bunch of 0s where MT’s scripts have NaNs.
MT – But you want to keep the behavior with NaNs?
LW – Yes.
(time permitting) Brent sage reproduction issues
…
BW will use PB’s exact env to try rerunning the openff-2.1.0 fit to isolate whether the differences in newly-tried reproduction attempts are due to changes in deps or run-to-run variation