2023-01-24 Protein-ligand benchmarks meeting notes

DD : fah-alchemy name change - final 3

• the top three names are, in order of popularity:

  • alchemiscale

  • alchemscale

  • alchemicloud

• these will be placed in a second round of polling; the winner of this poll will be the new name of fah-alchemy; opinions on the form of vote we take?

DD – I’d suggest to immediately settle on alchemiscale

• JC + DS + JW – Approved

JC – Should we register the domain (alchemiscale.org)

• MH – I can do this, will register to chodera lab

• (DS + JC + JW) – Approved

DD : current sprint - ended today 1/24. Sprint retrospective

• Done:

  • DD – Hugo MacDermott-Opeskin (HMO) has been rea;ly productive, knocked out 3 PRs. Included scope enforcement, I’ll assign subsequent work on this to him later.

• In review:

  • MH – fah-alchemy 25/52 - I have a few sentences left to add here, worked with DS on this, should be almost ready.

  • DD – gufe 111 – Topological approach implementation. Asked DS or RG to review.

    • DS – I’ll get to this ASAP.

  • IP – PLB 83 should be ready for review. I’ve been running tests, they all run. I checked TYK2 netowrk and it’s a star map. Is that intentional?

    • IA – That’s strange. It should be auto generated but it shouldn’t make a star, it should have made a minimum spanning graph. It’s really unlikely that that would be a star.

    • JC – Is the goal to make one of each map?

    • IA – Eventually, right now it should just be minimum-spanning.

    • IP – I’ll add a note to the PR. Also, all the edges run, which is great news.

      • IA – To completion?

      • IP – At least one iteration

    • IP – Other checks to run? I just check that they run one iter in perses. Then i ran TYK2 and CDK1 to completion.

    • DD – Do we think that’s sufficient?

      • JC – Sure, it’s better than nothing

      • IA – At the end of the day it’s a network, and we won’t know how well it does/its deficiencies until we add more edges and compare results.

    • JW – IP, I know you’re just using OpenMM to load these files. But please let me know if you encounter any issues with loading the proteins using OpenFF toolkit

  • IA – PLB 87 could be ready for review

    • JC – I’ll review this today.

• In progress

  • GUFE 110 – MH – Working on this

  • perses 1066 – IP – I have protocol working with settings implementation that was just discussed (with MH’s changes). Also checked with Davi’ds implementation of bootstrapping, still need to test with JC’s changes. But overall the protocol is working and we need to test running full sims. JC, do you have suggestions for n_cycles/lambda windows to use there?

    • JC – Sure, I’ll comment on perses 1066.

    • DD – Helpful things would be input params, n cycles, lambda windows, bootstrapping number

    • JC – There aren’t lambda windows - take continuous steps.

    • IP – Ah, right, just number of cycles.

    • IA – FYI - You’ll have trouble converging outside TYK2.

    • JC – We can use simple test systems like methane->fluoromethane in vaccuum or solvent.

    • IA – We have a benchmark set of benzene substitutions we can share

    • IP – Cool, I’m working on collecting standard tests so this is helpful.

    • JC – TYK2 tests aren’t interded to converge, just to be a not-crashing test.

    • IP – Messaged IA and RG about “what do we expect from chemical systems about components” - If we expect input to noneq cycling protocol to have exact same type of components - like input a protein-ligand system and the other is just protein… RG said he’d answer this

      • IA – For RBFE’s you’ll have the same keys in the dictionary. But for ABFE’s you’ll have disappearing components

      • JC – Are we given the transformation - Like a PL complex then L in solution as separate things - or does this set up the L in solvent as a separate thing?

      • IP – IT autodetects what the chemical system has, like if it’s protein+ligand+solvent, it’ll do the ligand … (lost the thread, see recording around 24 minutes)

      • JC – Is goal to look at components in states A and B…(see recording)

      • IP – A lot of this will be an implementation detail in protocols

      • JC – …

      • IA – …

      • DD – Does it make more sense for the mapping to bridge this gap? Like for chemical systems A and B they have a dict of components. Then instead of having the same keys, the mapping could have detailsof how these things are connected.

      • IA – That would help, but mappings aren’t required.

      • JC – Are you just considering ABFEs….

      • IA – Probably need to discuss this a bit, consider all options

      • JC – yeah, we’ll want to discuss what it means to eg. have a component in one input but not another. Like non chemical species could have dummy atoms.

      • IA – That would work, let’s discuss this among the options.

      • DD – I’m in favor of putting this into the mapping.

      • JC – In an absolute transformation, you need some things to map to dummy atoms, but in such a way that physical parameters are preserved…

      • DS – In terms of network-scale analysis,there’s a big difference between having a key pointing to None and not having a key… some approached would be possible but painful.

      • IA (chat) – ATM and Separated Topologies are other edge cases where you won't have a mapping

      • JC -- …

      • DD – Note that there are atom mappings and component mappings. The component mapping might include an atom mapping…

      • JC – Discuss next Tuesday? Then we could think about it and present arguments…

      • …

      • IA – Until a decision is made, we do need keys to be the same for ABFEs. If there’s a missing component then that’s the one that’s disappearing. So that should apply for reviews of the current PR.

      • IP – Got it, that doesn’t change things for me but I’ll expect possible change as a result of next week’s discussion.

      • DD – Sounds good.

      • JC – And I’ll review

• DD – fah-alchemy 56, 44, 40 – I’ve been short on time, need to respond to HMO’s comments. I have tests in place that do end-to-end results retrieval. I also spoke to DS last week to align the way that we store objects in the data store. And I think we’re there. I did work on how to have workers set task status but I’m going to separate that into a separate PR.

• DD – fah-alchemy 39 - Adding tutorial notebook to OFE/ExampleNotebooks - Still need to do this.

New items

• fah-alchemy 34 –

• fah-alchemy 69/70 - HMO is working on this

DD : new sprint kick-off - ends 2/6

• architecture overview : https://drive.google.com/file/d/1Elw5vWYXuGKSuO-E3jNMxkYnSQaioVF8/view?usp=share_link

• coordination board status : fah-alchemy : Phase 1 - MVP

• updates on In Progress cards

• seeking volunteers for unassigned cards in Available

• DD : fah-alchemy 0.1.0 milestone

• DD – I’ll update the sprint board for next time. Any suggestions for improvement?

  • (no comments)

IZ: Cinnabar bug

• JC – IA asked for additional tests. Is this blocking? IZ is a grad student and doesn’t have a ton of time.

• IA – Sure, I’ll go by and give it a final review, expect approval.

• DD – I spoke with IZ, willing to try pytest.parametrize

• IA – It’s not blocking, it wouldn’t block my approval.

JC : object models for representing free energies + uncertainties

• JC – Still working on this

• DD – RG mentioned that this could nucleate an object model for cinnabar.

• JC – That would be great.