2020-09-17 FF Release Meeting notes

Date

Sep 17, 2020

Participants

  • @Hyesu Jang

  • @Christopher Bayly

  • @David Mobley

  • @Lee-Ping Wang

  • @Jessica Maat

  • @Pavan Behara

  • @Simon Boothroyd

  • @Joshua Horton

  • @Owen Madin

  • @Matt Thompson

  • @Jeffrey Wagner

Discussion topics

Item

Presenter

Notes

Item

Presenter

Notes

 

 

DM - WBO, One or two months before being ready to fit. Need to time to go through each chemical series, find torsion term for each chemical series, check how diff SMIRKS shows different wbo dependencies, determine which torsions should include wbo. Will check on biaryl case too.

CIB - Starting point will be assigning SMIRKS for each legend of plot C.

CIB - Carboxylate vs carboxyl acid, will have different wbo dependenicies.

CIB will propose SMIRKS string for each line + other conjugated bond which may need wbo-dependent torsion term.

DB- need input on hierarchy between terms or replacement of existing term?

CIB - For example(generic case), double bond between two sp2 carbons in the center of the torsion (t43, t44, t45) need to be replaced to one term with single wbo dependent torsion term. need to check if these have any child torsions. (like atom 1 is specified to nitrogen or other specific atom) that may fall into the new category. this case will be a good starting point.

Aside from using chemical intuition, find what are the torsions describing a given torsion, and what torsions we might need convert into wbo dependent torsions.

CIB - need to check do we have torsion targets to train them?

DM- we should have enough training data.

 

(Jeff’s copy of notes)

CIB – Two action items: Come up with SMIRKS for vairants of…

  • 1: “Look under the light” Like figure 4 from Chaya, but include another one with carboxylate, since it’s different from carboxylilic acid. Include trivalent and divalent nitrogens.

  • 2: “Obvious areas of great need” like biaryl torsion.

CIB – Concened that my SMARTS patterns may be too promiscuous. JM and PB can review for this – Like, given a proposed SMARTS, pull down torsion profiles from QCArchive, and evaluate whether torsion profiles indicate that a new interpolated parameter will help.

 

+ checking MACCS key after grouping the molecules using the new proposed smarts.

 

 

new priors and amide re-fitting w/ planar amides

@Hyesu Jang

  • Slides: https://openforcefield.atlassian.net/wiki/spaces/~868974935/pages/670171278

  • LPW – Any thoughts about proposed new values of prior widths

  • Hyesu will add link to prior widths study

    • CIB – If you include all elements and bond orders, you wind up with many populations of parameters. I wonder if the same thing could be done for angles wrt the hybridization of the central atom

    • LPW – I think we need a combined approach with looking at populations of data and also using chemical intuition.

    • CIB – When would these changes be released?

    • LPW – Could have a 1.3 release

      • DM – Agree

      • CIB – Agree

      • JW – Agree

      • LPW – Concerned amides may now be excessively biased toward being planar. A longer term fix would be incorporation of JMs work on improper torsions. But I’m not sure about progress of that.

      • HJ – JM, could you send me cases of nonplanar amides to check against?

      • CIB – A FF that’s overly biased toward keeping amides planar is a better starting point for subsequent optimization than one that doesn’t keep them planar enough.

      • CIB – SB, I expect that the nonbonded fitting will

  • HJ – DM, how is dataset selection coming in your lab?

    • DM – Largely handled by JM and TG. I know a lot is being submitted to QCA.

       

openforcefields status

@Jeffrey Wagner

  • Made 1.2.1 release

  • Looked at automating HMR tests on coverage set – Would welcome feedback on whether this is a good/bad idea

Action items

Decisions