2022-04-14 Force Field Release Meeting notes

 Date

Apr 14, 2022

 Participants

  • @Pavan Behara

  • @David Mobley

  • @Christopher Bayly

  • @Daniel Cole

  • @Chapin Cavender

  • @Michael Shirts

  • @Joshua Horton

  • @Michael Gilson

  • @Trevor Gokey

  • @Jeffrey Wagner

 Discussion topics

LINK TO THE RECORDING: https://drive.google.com/file/d/1ZYxPGGY_Rsu5PU6wJ04CLtTjrmJ1LzpP/view?usp=sharing

Time

Item

Notes

Time

Item

Notes

10 mins

Brendan Duggan’s (UCSD) NOESY NMR data, utility in FF fitting or benchmark

  • DM – Offered to share NOESY NMR data with us. Offered to do additional measurements of organic molecules in solvent. My experience is that the tools to predict this data from simulation aren’t very good.

  • CC – For NOESYs there’s a simple model where you assume that signal is proprotational to 1/r^6, which is OK, but there’s a lot of discussion on how to do it better but no consensus on how to actually do it.

  • CBy – I did a bit of NOESY work back n the day. Used the data in conjunction with j-couplings, which paints a clearer pitures. If you also have transannular info you get a good diea of the 3D shape. NOESYs can basically tell you cis/trans/gauche info but it’s mostly just really local. Also, that 1/r^6 if an ENSEMBLE average, so you don’t know whether it’s a large population at medium distance, or a small population at a large distance. However, all the sources of liquid-state data are going to be complex like this. I’d recommend talking to David Case about this - He’s done a lot of work in the area

  • DM – (Shows some SMILES of the proposed datasets. Several macrocycles (like 15 membered rings, posted in Slack:

https://openforcefieldgroup.slack.com/archives/CK5MQG0TC/p1649948895403549
  • )

  •  

  • DM – Nobody is looking at this currently, may be a good rotation student project. Could still run the sims and see how the data actually does look compared to the experimental data.

  • CBy – Would be handy to look at this.

  • MG – If I had someone to look at this it would be cool to look into. The varied solvents make it especially interesting.

  • DM – It’s particularly exciting because we could get on-demand experiments done in chemical areas of interest

  • CBy – Reminds me of ionophore projects

  • MG – Matt Jacobs(?) had looked into this using macrocycles.

  •  

  •  

20 mins

Potential point release of sage

  • Slides - https://docs.google.com/presentation/d/1zvzXT49ocxIGmQu3N_spukCb4kelH0yRzzYxJCDrDvk/edit?usp=sharing

  • 8-9 - MSM appears to work and we can use them to reset values before each fit

    • DC - Sometimes bond k values can go negative for rings

    • JH - Got around it by using average over group, which hides the problem issues

    • CB - Using regular restraints? Yes

      • PB - sometimes the k value changes by 200 in the opt

    • CB - Using average is sensitive to outliers; possibly need to tune the restraint

    • DC - Could split parameters based on distribution

  • 10 fitting hessians appear to make geometries worse

    • MG - This is not surprising

    • JW - Is this a dataset issue? I would think geometries and hessians would compliment each other

    • CB - Want to get low range of the modes correct

      • MG - Agree

    • DM - If we are deciding to put Hessians aside, we should document the work we did so far

      • PB - Yes

  • 11 - CB - showing improvment; is it many tiny improvements or pathologies going away?

  • 12 - JW - Which would be the proposed point release?

    • PB - The green (MSM + optimize all + fit torsions)

    • TG - The RMSE plots are not weighted; might want to weight them to prefer minima

  • 16 -

    • MG - the bicyclo pentane issue doesn’t seem like an FF issue; maybe low priority

    • CB - The VEHICLE set has really strange chemistry

    • DM - We should focus on the biggest most common failures

    • CB - WBO could help in these cases

  • 17

    • JW - Simon needs to take care of pt 1 and 2, and 3 is what the infrastructure team can take care of

    • TG - It would be great if we could get pt 1 and 2 written down

    • DM - How to handle binding free energies for point releases. Is OpenEye willing to run these?

      • CB - Can’t commit but in our best interest to pursue. We can offer compute and infra but not man hours

      • DM - That would work for us

      • CB - Will be back June 13

      • CB - Overlap with OFE and OFF. FE are the goal of a good FF, but expensive

      • CB - Do we need to show improments in FE explicitly/do we make it a milestone for release?

      • DM - Want to make sure FE aren’t degrading

      • CB - Free energy or nice to have?

      • JW: Shouldn’t be a blocker for point release

    • PB: Should

      • DM + CB: Lets group things together and make fewer point releases than more

    • TG: How should order them?

      • CB: vdW/charge, bond, angle, torsions/impropers

      • DM - We have fixes that we need to get through

  •  



 Action items

 Decisions