2022-12-15 Protein FF meeting note

Chapin Update

• CC will post slides here

• Slide 1

  • MS – Where do library charges come from?

    • CC – I generated AM1BCC ELF10 charges for mainchain/capped/uncapped termini.

  • CC – Preprint for LiveCOMS paper in about a month?

    • MG – I’d really like to hear from coauthors by then

    • MS – Would be good to have that in time for the grant submission (beginning of March)

    • MG – For timeliness, it may be good to push harder on contributors to approve - Like, instead of requesting changes, we could get things into pretty much a final form and just ask them for approval. Or we could ask them to return things on a specific timeline.

    • CC – I like the latter - It’ll be good to take advantage of paralellization

    • MG – CC, could you and I sit down and review this?

  • JW – Confirming that we have plans for benchmark simulation compute - Plan A is UCSD + UCI clusters, and plan B is to ship benchmark sims off to LW and have them run on Lilac.

    • CC – Yes, that’s the plan. And benchmarking will be done in tiers, so we can rule out bad FF candidates before we invest too many resources into benchmarking them.

    • JW – Perfect.

  • MS – In terms of setting up and monitoring sims, is any additional help needed?

    • CC – I’ll enlist LW’s help for running these.

• Slide 3

  • JW – This is awesome

  • MS – Could see whether the protein-specific model is OVERspecified.

    • JW – Agree

GNC & Library Charges

• JW - Graph charges might be ready to allow it to be present on Rosemary. Should we choose it instead of Library charges?

  • LW - I think it would make sense having a ff with Library charges.

  • mattthompson (chat)
    +1 to “it doesn’t matter what it’s called"

  • Pavan Behara (chat)
    I love the idea of 3.0.0_with_graph_charges and a simulataneous release

• (General) – We should release the initial version of protein FF with library charges (not graph charges).

• (General) – We should make sure that the graph charges don’t end up adding sequence dependences (or do, if it’s intentional)

• …

• LW – Library charges are somewhat slow. On a big dataset of 10,000 molecules with 10,000 library charges, it took ~8 hours (but code was not optimized for efficiency – real ff will have ~135 librarycharges)