2023-03-21 Protein-ligand benchmarks meeting notes

JC – RG made a new PDB reader that should interoperate with OpenFF. Have we tested this?

• JW – I think I tested it, but I’ve been meaning to make a new push for better PDB support and have integration with pdbinf be a part of that. So I’ll make sure to get tests in for this.

• JC – Great. Thanks.

• DD : sprint retrospective

  • Review Done cards

    • what went well?

    • what didn’t?

    • what do we need to improve our approach?

    • DD – (see recording) – HMO did some good work on NEQ cycling. We’ll talk more later.

• DD : next sprint begins tomorrow, spans 3/22 - 4/3

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

    • JW – So basically synccompute became yellow, once that’s green we’ll test and work to make asynccomputeservice, and then once that’s done we’ll begin on F@H compute?

    • DD – Right.

  • alchemiscale 0.1.0 milestone

    • DD - Alchemiscale ??? – HMO is working on this, should update how task statuses work

    • DD –? (see recording)

  • coordination board : alchemiscale : Phase 1 - MVP

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

    • IP – Perses 1066 – Noneq cycling - Yesterday I worked with DD to make this work with alchemiscale infrastructure. Found an issue with mapping,opened an issue to discuss. But largely the protocol is working and I need to find the time to add the mdtraj storage fix. Then I need to do a bigger tests onm protien-ligand indstead of just small molecules in vaccuum. Then we need to ensure it gives consistent results for tyk2 compared to previous.

      • DD – Can this group support you in some way?

      • IP – Could use advice on noneq cycling/the theory and how to set up tests.

      • JC – One test could be as simple as doing a “null” free energy calc, where we expect the cycle to be 0. I can work with you on others.

      • JC – I think the OPenFE folks have been working on a repex version - We could also use that for comparison

      • RG – Yeah, we could use that for back-to-back testing.

      • DD – IP, could you schedule a meeting with me and JC to set up noneq stuff? Then once that’s done we can plug in the comparison with OFE.

      • IP – Sure, I think that will need to be next week.

      • IP – Also, how should we handle the mapping issue? (gufe 155) - Basically when you create ligand mapping object, you make a mapping from component a to component b. The keys are indices in A and the values are indices in b. Two questions.

        • Do we want a parameter to instead to component B to component A when you create the object?

          • DS – We should put this on the user - They can reverse the dict manually using a one-line comprehension. I think OpenFE can provide reversed mappings from some methods as well.

          • IP – Maybe we could have the mapping exposed after construction as well?

          • DS – Nah, I’m against this,it should be immutable after construction.

          • JW – Mapping is always int-to-int and never int-to-none?

          • DS – Right.

        • What do we mean by “A to B” in this context - Like above, we should explicitly say that keys are atom indices in component A and values are atom indices in component B.

          • DS – We should absolutely make sure the docstring reflects this.

    • JC – Perses 1128 – Making decent progress, should be quick but I’m wresting with grant proposals. Should be done in the next day or two. Though I noticed that doctests were broken so I’m fixing that as well. I could use a time machine.

      • DD – That’ll be great, this will let us test on the new version of PLB.

    • DD – Alchemiscale 34 – Synchronouscompute - Finalizing a few things around storage and lifecycle. Basically two more features to add.

    • RG – ExampleNotebooks 36 – We’d agreed that I’d switch it from noneq protoocol to repex protocol based on timing/ software availability. But I’m falling behind on this.

      • DD – I think I have what I need at the moment (RG showed me how to use lomap), so this isn’t blocking me or other developers.

  • Available

    • DD – Gufe 143 – Some things that we could be doing better - Could be good in GUFE 1.0, will work on this once alchemiscale is out.

      • RG – I’ve come around to largely agreeing with you. We can drop many (but not all) labels.

      • DS – If there are two different small molecule components with different ffs, this may require a lot of labels.

      • DD – Yeah, there’s at least one solution to that, we could discuss alternatives.

  • create/nominate new cards for inclusion in this sprint

    •  

  • will post on #free-energy-benchmarking when next sprint is finalized

  •  

  •  

• DD : demonstration of compute service running on vulkan, performing NEQ Cycling via perses; alchemiscale client workflow

  • DD – Would love feedback.

  • (see recording)
    
    

• DD : protein-ligand-benchmark : establish new working group, or use this one?

  • potential new participants