2022-02-02 BespokeFit meeting notes

Update on TYK2 calculations with bespokefit

@Joshua Horton

• JH: Updated pmx to work with latest version of OpenFF toolkit, PLBenchmarks. Currently (something) is a zipped package folder on a GH page, but I’m not sure if we should broadcast the update.

  • DM – Would be good to broadcast an update about the pmx-workflows package.

  • DC – We’ll release a tutorial and scripts with the bespokefit paper.

• DC – In terms of accuracy, we don’t have a good idea?

  • JH – Yeah, only a few edges are finished, so I don’t have a good guess until more runs complete. I did a test edge and the results were close to an openff-parsley result.

  • DC – “No change” is a good result there

  •  

  •  

• DM – DC, are you guys involved in virtual site work?

  • DC – We’re using the functionality a bit as users, but not coordinating with developers.

  • DM – We’re doing a different project with a chlorine in a protein, and getting big errors.

  • DC – It’s looking like sigma hold/bondchargevsites aren’t giving that much of an improvement.

  • DM – My previous experience has shown that monochlorinated atoms aren’t that changes, but polychlorinated ones benefit more.

  • DC + JH – Kinda seen that in comparisons of CF3Br and CH3Br. You end up getting a more significant sigma hole in the former because there’s more elctron withdrawal.

  • JH – For small molecules, can run them through qubekit, and it’ll recommend vsite positions.

  • DM – Here’s what I’ve seen before:

Open

from DOI 10.1007/s10822-011-9528-8

• DC – We’re trying some vsite HFE stuff soon, I can include this in experiments.

•  

Boron and Silicon

@David Mobley

• DM – People are, in principle, interested in this. I’ll respond to Reaxys (who offered to open up some of their B and Si containing physical property dataset) saying that we’re interested. I’m not sure whether we get to SEE the data that we’ll get released with a permissive license, but if we’re allowed to browse it we could make some judgement calls on exactly what would be most beneficial.

  • DC – What I’d suggest is working with OMadin to select those. He was concerned about the small size of the dataset, but to us 100 mols would be great, since most of our datasets are smaller, and we aren’t training all that many parameters.

  • DM – We’ll hit a chicken and egg problem, where a SMILES → parameterized molecule workflow would need a way to generate conformers for B and Si-containing molecules (both for AM1BCC charge assignment, and at least to make the outputted geometry), and OE can’t generate conformers for those atoms. So WE’D need to provide the functionality for them to make these conformers, but we might need OE to support these to even fitting it.

  • JW – We should be able to handle assignment of valence parameters to this. Not sure about charges.

  •  

  • PB – We could get charges from XTB - That’ll handle B and Si.

  • DC – Our atoms-in-molecule charges should be sensible too.

  •  

• DC – If reaxys offers us a choice of chemistry, I’m not sure that I would know what to ask for.

  • DM – We can consult with OM and SB on that.
    
    

• PB – Cancel tomorrow’s FF release call for lack of agenda items and decisionmakers?

  • DM – Yes