2021_01_14 OpenMM/OpenFF check in

JW – OpenFF Toolkit 0.11.0 is scheduled for Feb/March. Will have breaking changes, JW will make sure a PR to OpenMMForcefields gets opened ahead of time.

JS – Suggestions for how to implement shrinking-ligand dynamics (90’s paper by Gilson)?

PE – What molecule size?

• JS – 30 - 100 atoms

PE – Advantages of this method over alchemical methods?

• Sampling phase space is easier if the ligand shrinks

PE – A few different approaches are possible:

• Every atom has two different positions, use one position when calculating one set of forces, another position when calculating another set of forces

• Every atom has one position, but at each scale-down you change the physics to compensate (this will probably be easier). But you want to do that in such a way that you’re only having taht happen for the interactions IN the molecule. So you’ll need to do bookkeeping to ensure that the intramolecular ineractions and intermolecular interactions are known. So you probably want a customnonbondedforce where you have a PerParticleParameter that says whether each atom is part of the group being scaled. That would be the most efficient, but it won’t work with PME.

  • JS – Not using PME is OK

• JS – Can customnonbondedforce also track the center of mass?

  • PE – No, it’ll only be able to affect forces/energies

• PE – Would the scaling change every step, or every thousand+ steps?

  • JS – Thousands of steps/100k picoseconds

• PE – So what I’d do each time you change the scale factor is go in and manually set the new positions and change the bond lengths and stuff. You could choose not to directly update the positions, but changing bond lengths without doing that would add energy to the system which may be unwanted.

• JS – I looked at the example plugin, still wrapping my head around how to implement this idea

  • PE – The plan above doesn’t necessarily require a plugin - I’d start just with custom python code.

  •  

JS – Does current OpenMM support amber19?

• PE – If the protein is parameterized by ambertools/tleap and loaded into OpenMM via parmed, then yes. But the typing rules are different and the built-in openmm forcefields doesn’t currently implement that typing logic.