2023-10-10 alchemiscale Working Group meeting notes

alchemiscale.org user group

• user questions / issues / feature requests

  • MO – Not a new issue - DM said that JW wanted us to make a minimum reproducing case for the AT charging issue, but we’re having trouble getting a reproducing case locally.

    • JW – If we have time at the end of this meeting, please stay on and we can look at this.

  • DD : hif2a issues resolved?

    • MO – This is separate from the AT issue, we’re able to run locally but we’re getting a 5 kcal/mol error. Still diagnosing.

    • DD – The thing that IP’s working on may be able to bring down this error by having more information contributing to the convergence.

    • IP – Can also improve RBFE protocol with more lambda windows and longer sims.

    • MO – I think we’re just using the default 11 windows and 10ns. But again, we’re unable to reproduce locally. We’re basically looking at this because this structure ran fine with septop but is having a lot of trouble with alchemiscale/OpenFE.

    • DD – Are you seeing the same problem locally as you see on alchemiscale?

    • MO – HB ran 5 replicates, one of the 5 for each edge would have a >3 kcal/mol error from MBAR. I ran 7 replicates on alchemiscale, and also pulled them down and ran another 7 replicates locally. I was expecting to see some with >3 kcal/mol, instead the largest I saw was 0.2 kcal/mol.

    • DD – It’d be helpful to have a small writeup of this, especially in deciding next steps.

    • MO – Will do.

  • JS – With ASAP-alchemy, we have a fairly basic workflow in place to run stuff. JH is working hard on everything that comes before that (prepping, docking). The other thing that we need to do is everything after alchemiscale. What is the current recommendation/best practice that we should be using to analyze all of the free energies? I remember vaguely that IP was working on this…

    • DD – There’s a demo in the docs: https://docs.alchemiscale.org/en/latest/tutorials/demo/Alchemiscale Demo.html - Has example for comparing to experimental values.

    • JS – Gotcha, this is what we’re currently using.

    • IP – There are two faces of this - 1) gettng the FE estimates and plots from cinnabar and 2) when we need to explain the plots, we still don’t have access to trajectories/timeseries. Sometimes with perses we do some additional analysis to explain/look at the data underlying the predictions.

    • DD – Is this needed for you, JS?

    • JS – Was more wondering whether the tutorial is the latest info. I think it’s broken with the latest cinnabar… I’ll double check whether it’s broken/what to do.

    • DD – There may be some funny business with dev versions/pip installs, but I think with stable pacakges everything should be working.

    • JS – Gotcha - I’m glad the notebook is in official docs, I had first been introduced to it as a notebook floating around.

    • DD – IP, is there an option in feflow to include pieces of this in the output data? Is there some subset of information short of the full trajectory that we could optionally output?

      • IP – There’s the option to include some outputs in shared or scratch. It’d be possible to change the exact info that gets outputted.

      • DD – It could be handy to have some useful outputs short of the full trajectory, but what would those be? IP, maybe you could open an issue on the feflow tracker for this?

  • New submissions?

    • IP – I’ll be putting in some comparison simulations soon for feflow.

      • DD – We won’t have feflow deployed to alchemiscale.org workers in the short run.

      • IP – Oh, right, I’ll run those locally.

      • DD – Great, looking forward to deploying it once it’s validated/released.

    • MO – I’ll probably submit additional hif2a stuff as well.

      • DD – Sounds good.

• compute resources status

• current stack versions:

  • alchemiscale: 0.2.1

  • gufe: 0.9.4

  • openfe: 0.13.0

  • perses: protocol-neqcyc

DD : alchemiscale 0.2.1 released last Friday!

• JS – Awesome, looking forward to the retries.

alchemiscale development : sprint ended 10/9; next sprint spans 10/11 - 10/23

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

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

  • call for volunteers for available issues

• alchemiscale 0.3.0 milestone:

• alchemiscale-fah: 0.1.0 milestone

  • DD – Times square sampling needs some input from JC/IZ and is kinda blocked until then. But the other two can move forward.

• review Completed cards

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

  • IP – PLB 93 – I have a meeting with HBaumann tomorrow to discuss MST results. The underlying blocker is that we’ve worked somewhat separately and need to check in to coordinate.

  • IP – feflow 1 – I’ll be working on this today.

  • DD – alchemiscale-fah 1 - I wanted to get some insights from JC about how to get this working best. Right now we’re discussing the simplest option. Some basic implementation started. I’m gonna sync up with IP on using this approach for the noneq cyc protocol, will need minor modifications but should mostly be portable.

  • DD – Alchemiscale-fah 3 - noneq cyc on F@H. This will be our first protocol.

  • DD – Calling for volunteers on any available tickets

    • JH – I’m particularly interested in speeding up bulk-requests.

    • DD – I think we already have that - get_network_results

    • JH – Awesome.

DD : gufe#184 - openmmforcefields next release?

• IP – JC said he was going to do this last week, but I think he didn’t finish. I’ll push him again.

• DD – Hm, this is a blocker for us. It’s possible to cut a 0.11.3 without these changes and then put the unfinished ones in a later release.

• IP – …

• DS – This is also a blocker for OpenFE

• JW – I’ll mention this from the OpenFF side

IP :

• Protein-ligand benchmarks working group update

• Protocols migration to