2023-09-26 alchemiscale Working Group meeting notes

alchemiscale.org user group

• user questions / issues / feature requests

  • MO – Reran BACE amides, errors look really high. So I’m running with our other workflow on the UCI cluster to see if we get similarly high errors. Talking with HBaumann and IAlibay on Hif2a benchmarks to see if those also have big errors

    • IP – Re: Hif2a - Is this a dimer?

    • MO – IA says that MBoby worked on it, decided that the dimer was a more accurate representation than the monomer. But ACalderuse from our lab ran with it as a monomer and it looked fine, so DMobley’s wondering about this.

    • JC – Is it a dimer in the experiment? We should be able to run on a dimer if we need to but the calcs should try to replicate experiments.

  • JS – I’ve been running it but am hitting system-specific issues. Don’t have them ready now but can share some info later. It’s a problem with automated protein prep. So we’re getting OpenMM complaining about residue protonation. Will show this next week.

• compute resources status

  • DD – 230 workers on NRP and lilac. We have more capacity for calcs.

  • IP – Is pre-emptible queue on lilac working?

    • DD – Yes, all our current lilac workers are pre-emptible.

• current stack versions (note these have been updated since last week):

  • alchemiscale: 0.2.0

  • gufe: 0.9.4

  • openfe: 0.13.0

  • perses: protocol-neqcyc

DD : alchemiscale 0.2.0 released!

• Please upgrade your environments following instructions here: User Guide — alchemiscale documentation

• JW – Did this update require a data migration?

  • DD – In the end, it didn’t need any migration. It looks like OpenFE added stubs to keep things from breaking. But if any users hit trouble please let me know!

DD : gufe#184 - openmmforcefields next release?

• JC – I have one last thing to merge that I’ve been delayed on. Should be able to get that out today. Will also have new espaloma 0.3 capabilities as well. This needs to get in because our manuscript is going up.

• IP – JC – Other than the note about the docstring, is there anything else needed on that PR?

• JC – I need to do some checks/extra local testing. Will merge today.

IP : Protein-ligand benchmarks working group update

• IP – Haven’t had time to synchronize work with IA and HB. Last time I shared and updated the edges that were manually curated, and HB is running benchmarks on that. HB is also working on improving MST maps, but haven’t seen activity on that PR. So I’ll meet with HB and see what I can do to help.

IP – Protocols migration to

• IP – This will host the perses protocols that alchemiscale will be running. One from PErses and another from OpenFE. First task is to get the protocols from the two sources to use the same OpenFE base objects.

• DS – For having OpenFE protocols in there, are you in contact with IA? I’m worried that we’ll end up needing to maintain a fork.

• IP – For now I’m just making sure the perses protocol can work with OpenFE base objects. But after that I’ll sync with IA.

  • DS – Ok, do contact IA after that.

• DD – And decouple from OpenEye?

  • IP – Yes, that’s another thing we should be able to do.

• DD – Any other ways I can help?

  • IP – I’ll let you know, probably in the next few weeks.

• DD – What’s the objective with pulling in OpenFE protocols?

  • IP – OpenFE protocol uses some modified objects that were extracted from perses. First idea is to have both protocols use same base objects (and migrate those to openmmtools). But as DS said, they’re actively working on that, so we don’t want to divert that and make a mess.

  • DD – Is the thought that, instead of having these pieces duplicated, there would be an option to update the original objects?

  • IP – That’s a possibility, I need to sync up with IA to discuss plans.

• DD – In the bigger picture, I’m hoping that this gives us a chance to test extends support in protocols, which will be critical for F@H. We can use the noneqcycling protocol as the test for this. Let me know once you have a PR up.

• IP – Perfect, please open an issue for this.

  • DD –- Will do

• IP – Also, I cleaned up the project board for alchemiscale so we don’t have the “test noneq cycling” ticker permanently there.

• IP – In branch, I found that when we pull results, we now get quantities.

alchemiscale development : new sprint spans 9/27 - 10/9

• architecture overview : PL Benchmarks on FAH - Architecture v6.drawio

• coordination board : alchemiscale : Phase 3 - Folding@Home, new features, optimizations, targeted refactors

• alchemiscale 0.3.0 milestone:

  • DD – Focused on F@Hcomputeservice and refactors. Made new repo, alchemscale-fah. First goal there is to try a noneqcycling protocol.

    • IP – How different would a F@Hprotocol look from a basic protocol?

    • DD – Mostly different in the portion that runs the simulation. We’ll be creating a set of files that we drop onto F@H cores using their adaptive sampling api. So this should look similar to the fenchiridion implementation.

    • IP – I see, I’ll keep this in mind - basically that we may need to swap out the compute setup to wrok on F@H.

    • JC – Yeah, it’d be good to make an interface where we can provide the three files that F@H needs (serialized system, integrator, context?). Should be able to do the same thing with extendable replica exchange.

    • IP – Great, excited to get started on this, I’m open to iput on how to structure the interface.

  • • DD – alchemiscale-fah 1 - I’m working on this first, will try to sprint for it, it’s possible I’ll get it done in 2 weeks but somewhat unlikely.

    • DD – AF 3 – Noneq cycling - WIll also work on this one.

    • DD – AF 6 – self adjusted mixture sampling - Open to ideas for how to implement a fah version fothis

      • JC – We’ll want to debug locally first. Initialy impolementation in openmmtools. What ahpens is that there’s a single replicatthat does on the fly adjustments and there’s a whole chain that does a series of adjustmenets. There’s also a variant called times square sampling. IZhang had expressed some interest in testing this. But there are open science questions that someone needs to chase down.

      • DD – Great, I’ll tag IZ on this. Also AF5 is for times square sampling, which should I prioritize?

      • JC – You should merge them.

      • DD (Merged both into AF6)

      •  

• starting new sprint tomorrow; call for volunteers for available issues

  • DD – alchemiscale 173 – Anyone interested?

    • JS – I’ll volunteer HMOpeskin and JHorton. They’ll be back next week.

• updates on In Review, In Progress, and Available cards

  • (all covered above)

new discussion items from ASAP roadmap: ROADMAP: Computational Chemistry Core alchemiscale-related roadmap | Notion