2023-04-25 Protein-ligand benchmarks meeting notes

New MSK test cluster

• JC : new cluster coming online; will connect via MSK email addresses

  • setup looking in June

• JC : also new preemptible queue on Lilac with GPUs, so that will help

  • Queue now on Lilac (connect DD with lohit to get this working)

  • Hopefully we will have everything preemptible on new system (IRIS)

Working group name changes

• No objection to name change to disambiguate from PLB repo working group
  

PLB repo working group

• In discussion with OpenFold folks; not in a state where they can contribute

• OE interested, but looking to find someone on team to contribute effort

• IA : understanding that there may be several OE folks joining in?

  • they were looking to substantially invest time

• IP : only IA, RG, and myself have shown interest

  • Jenke Scheen is interested as well

• DD : probably reach out to individuals known to have vested interest

  • Toni Mey

  • Vytas Gapsys

  • others

• IA : OpenFE has a discussion on day 2; benchmarking will be part of that

  • will put it in main topics

User questions, issues

• DD : will do working session with JC, JS on May 5 to get started with alchemiscale

• IA : OpenFF bug impacting MCL-1 benchmark

  • JW – I’ll fix this and get a release out today or tomorrow

• IA – Found that default mask used in HMR was too high, used to be 4 AMUs, was going to change it to 3 AMUs, any objection?

  • IP – No objection, I saw the same thing

  • IA – Thanks.

  • JW – This is really helpful, we have tests for HMR in our FF release process. Those run with a hydrogen mass of 4 (subtracting 3 from heavy atoms)

  • (Perses PR was merged)

Synergies with Cole Group

• IA : saw chatter on Cole Group using FE calculations along with bespoke-fit

• JW : can we use unreleased FFs at all?

  • DD : not currently, FF needs to be specifiable via settings and then accessible in environment

  • MH – Wold be a good feature request

  • IA – We’d talked about serializing FFs and allowing for user submission, want to make sure we coordinate on timelines and what we say to Cole group

  • IP – We pass in a FF name

  • DD – It’s up to the protocol to determine what to do with the FF specified in settings

  • JC – Could we add URL support? Then we could pull in outside stuff.

  • DD – Will …

  • …

  • DD – Steps could be:

    • In OpenMMForceFields, make an issue for URL support

      • MH – I’ll make an issue ✅

    • Eventually, in GUFE settings, support input in the format of a full serialized FF.

      • IA – I’ll make an issue

  • DD : will invite Josh Horton to this working group

  • JW : I’ll follow up with Cole group from this discussion

Sprint retrospective

• Done

  • …

  • IP – Perses 1121 – (Supporting OFF 0.11) – Done, still some rough edges. Expect a release canadidate this week, otherwise early next week. Current changes aer in main.

  • Perses 1177 – Merges into noneqcyuclingprotocol branch. This is the main branch until the pereses 0.11 release. Some changes to follow openfe repex protocol. This is what we’re running in rpoduction with for the noneq cycling protocol.

    • JC – We’re still waiting to get some data from this to do a sanity check. Do you have/are you generating data for this sanity check?

    • IP – I’ll be doing that this week, looking to compare to TYK2 results from IA.

    • JC – In vaccuum and solvent is good. Just want a few hundred measurements to sanity check.

    • IP – I’ll join DD/MO meeting

  • DD - Repex protocol isn’t quite a discrete issue, will be more of a long term effort.

    • IP – …

    • DD – I’ll assign you

• New sprint begins tomorrow

  • Overlaps with ASAP in person meeting next week.

  • 0.1.3 milestone - performance fixes, can spin bugfixes out on a weekly basis as needed

  • 0.2 milestone - Docs and UX focus, may take a month or two

  • Coordination board

    • IA – PLB 83 – Just need to start working on it now.

    • DD – I’ll be starting work on initial user guide. Will use confleunce page as a start for sphinx docs.

    • MH – Alchemiscale 30 – Still need to do this

    • DD – Alchemiscale 129 – Creating networks can be slow, especially if they have PDBs - may be dependent on internet connection. Could use messagepack?

      • MH – messagepack would be good, hard to do worse than PDB

      • RG – adding gzip to to_dict could be the quick and easy solution.

      • DD – Yeah, that’ll be the easiest.

    • HMO – Alchemiscale 126 – task status

      • IA – (see recording, 55 mins)

      • DD – Can slot into 0.1.3

    • IP is taking on testing noneq cycling protocol against repex protocol.(no issue/PR#)

  • IA – Need anything for your talk/poster?

    • DD – I meant to put up a draft talk. Need results, want to get these from MO, JS, IP.

    • Drop results here: https://github.com/orgs/openforcefield/discussions/24

    • DD – Really need this by OMSF symposium, also will present poster at alchemistry workshop. So 2 weeks at the most.

    • IP – Specific format?

      • DD – Screenshots of plots are fine, and description of what was tested and the protocol/method used.

      • IA – a cinnabar-style CSV could be better, then plots could be standardized.