2024-10-07 Science team meeting

Project updates

LW

• Past week

  • Half out sick

  • Evaluator – fixing bugs, trying to get this running on NRP

  • Project plan templates

    • Preview here: Science project plan template

    • Will fill out with virtual sites or another project as demo by next week

• Next week

  • More work on virtual sites, re-fitting vdW, NAGL

Project updates

BW

• Past week

  • lipidmaps clustering

  • benchmarking visualization tool and CI

  • optimizations

• Next week

  • lipidmaps dataset

    • Open a PR with fragments

    • Work on full molecules, work on clustering threshold

  • benchmarking CI

    • Open PR to update yammbs with filters

    • Open new PRs for new benchmark sets

  • torsion multiplicity

    • Make a new TorsionProfileTarget, use a restraint instead of a constraint, run a test re-fit with that + explicit torsion multiplicities

Project updates

AMI

• Past week

  • DDX dataset guess update

    • ~50/700 had bad charges in successful computations

    • Restarting crashed calculations with vacuum wavefunction, didn’t help

    • Started other guesses, didn’t help

    • What to do?

      • Use the MLPepper dataset as a reference but add I and generate more conformers.

        • The 50k ESP dataset currently has ~1 conformer per molecule--may be due to ELF10 approach, typically generate 1000-10,000 conformers and pick bottom 2%, if we’re generating 5 and picking bottom 2%, that’s about 1 conformer per molecule.

      • Look into AIMNet/ANI/the method Josh used for optimization, and decide whether we want to use that or a more traditional optimization protocol

  • Hessian issues/new fit with S data

• Next week

  • Benchmark and analyze S FF

  • Work on NAGL2 ds

  • Maybe train FF based on MSM splits?