2024-07-23 BW/LW Check-in

General updates and discussion on projects

• Torsion fitting

  • Global benchmarks on Industry BM looking good

  • Haven’t run new TM benchmark dataset

    • Can work on this in parallel

  • Presentation: target Wed Aug 7

    • Global benchmarks on industry/additional benchmark script (+IC RMSDs)

    • TD minimizations

      • Training set – small

      • A bigger option: XFF 20% TD set

    • Look at opt benchmarks for each torsion separately

      • Pick one out that either improves or gets worse across training set, look at single point energies

      • Some relevant code in Sage 2.2.1 repo

  • Get to Friday, talk after leave

• benchmarking infrastructure repo?

  • BW – one way to make CI is to have a cron job on hpc3 that checks for new FFs

    • LW – one way is to have cron job/daemon on own computer that runs own benchmark

    • LW – what about using NRP

      • BW – interested on a technical level

      • LW – consult with JW, or DD. LW to set up session with JW/DD when BW is back

• smee

  • Results with Sage 2.1 are a lot better without TD

  • Good to plug in different stuff

    • LW – need to figure out cost of a smee job

    • LW – if that’s acceptable, XFF datasets?

• lipids – most important

  • BW – asked to follow up on F Pickard and D LeBard

  • LW – sounds good to drop the alkene t45, t46 parameters – let’s keep the t50, t95 parameters.

  • LW – probably a good idea to summarize in words

  • LW – if they don’t respond in 2-3 days let us know and DM can follow up

  • LIPID MAPS database

    • Fragmented the database, clustering still ongoing

    • Less important, validation set

    • LW – consider RDKit’s Lazy??Picker

•  

• infrastructure work --

  • add benchmarking repo to roadmap

  •