2021-09-22 Industry benchmarks meeting notes

Manuscript updates

• Sage optimizations

  • LD – We got Sage rerults from Ian Craig (BASF?) – 3/10 complete.

  • LD – Heard from Bill Swope that he’s generating Sage results as well.

• OPLS

  • LD - TFox talked to Robert Abel at Schrodinger about allowing release of results. RAbel said that OPLS could be benchmarked, but that all molecules must be public, so the OPLS comparison will have to be done on public dataset. So we started this at Janssen and it looks like the OPLS optimizations will take a long time. If we could spread it over three partners that would be ideal.

  • LD – I’m a little concerned that we don’t have anything official backing up RAbel’s statement. I offered to do two things:

    • 1) Write an announcement so the pharma partners will be aware that we may get blanket approval for public dataset, and not do internal dataset

    • 2) Reach out to Robert Abel to get formal confirmation of the above policy.

  • LD – I’m pursuing 2) now, but I’d like to get confirmation/support from Gary about this path.

  • LD – In the meantime, I’m working on generating OPLS custom parameters at Janssen.

    • DD – How much CPU time is required per molecule?

    • LD – 1000 molecules = 3 days. Public set is ~15k molecules (70k conformers)

    • DD – If the ligands overlap a lot, I could see the final set requiring less than 15x the work. If there are just a few common scaffolds it could be more like 2x or 3x?

    • LD – The whole internal dataset at Janssen took 4-5 days.

    • DD – It’s possible that the whole public dataset could complete faster than you can coordinate work with the other partners.

  • DD – Best course of action is probably to coordinate directly with GTresadern, TFox, and DMobley in a single email chain. It’d probably be a lot more work to talk with them separately.

Updates from team

• DH – Nothing from me

• JH – Nothing from me

• DD – Nothing from me

• JW – Working on forcebalance release that will be compatible with the rest of our current infrastructure (mainly OpenMM, evaluator breakage)

  • aiming on spearheading the refactor, but biopolymers take priority before that work