2021-03-18 Force Field Release meeting notes

Date

Mar 18, 2021

Participants

  • @Hyesu Jang

  • @Simon Boothroyd

  • @Matt Thompson

  • @Jessica Maat (Deactivated)

  • @Pavan Behara

  • @David Mobley

  • @Christopher Bayly

  • @Joshua Horton

  • @Daniel Cole

  • @Chris Ringrose

Goals

  •  

Discussion topics

Time

Item

Presenter

Notes

Time

Item

Presenter

Notes

15 mins

ICH fitting study update

@Hyesu Jang


(will be updated soon)

15 mins

Virtual site and functional form (future plans)

@Daniel Cole

  • DC gives update on experience/progress using virtual sites in QUBEKit and related infra (OpenMM, TorsionDrive, GeomeTRIC, ForceBalance)

    • Getting densities and Hvap, free energies in progress

  • DM - where does virtual site become important?

  • CIB - Divalent aromatic nitrogen ( pyridine type nitrogen), sigma hole on bromine and iodine, aromatic sulfur. But not sure about carbonyl oxygen, aliphatic amine(trivalent pyramidal nitrogen), sigma hole on chlorine

  • CIB - How to treat virtual site in torsions? same treatment as OpenMM does.

  • DM- OFF has general virtual site implemented - will be nice to synchronize later

  • CIB - placement of VS? DC - calculates bisector, and move along the bisector, with symmetry enforced.

  • DC - non-LJ functional forms touch a lot of infrastructure, how to best avoid huge infra changes?

  • SB - In long term, update SMIRNOFF specification, but in short term, you can plugin force field to the toolkit with custom parameter handler, without needing to update spec.

  • CB - would be nice if toolkit made it easy to rapidly test functional forms, i.e. just by changing the header

  • SB - could have a “only works with OpenMM” implementation (perhaps not immediately part of the public API) that allows user to specify the functional form that would later be passed into OpenMM’s CustomNonbondedForce

  • SB - target property?

  • DC - condensed phase property to protein-ligand binding

  • CB - of the options for funcitonal forms, the 3 things that people care about are the well, replusion, and core, but we only care about the well and repulsion, and these are hard to get right with only 3 DOF (sig & eps in LJ). The 12 term isn’t great, would like something like gaussian + 6 + something else. Thinks we do need 3 DOF to get condensed phase properties + sterics right, skeptical that the common potentials (exp-6, exp-8-6, 12-8-6, 12-6-4, etc.) are the best possible functional forms. Practical idea: explore gaussian as the repulsive part. LPW & MRS might have something to add on this topic

  • SB + JH will sync up on putting together a plugin (probably standalone Python package) for using non-LJ functional forms with current infra

  • CB - Say we want to add a VS to a trivalent pyrimidal nitrogen. This group inverts itself quickly, and has several intermediate states that may be typed differently, other complications like stereochemistry are issues

    • TrivalentLonePair in SMIRNOFF spec is one idea for this problem, with the distance d going up and down during inversion

Action items

Decisions