2024-10- Science team meeting

Project updates

LW

• Past week

  • General MSM

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

  • OpenFE collab on SFEs

• Next week

  • Project plans – trying to fix some of the organization structure in Confluence

    • Meeting with JE on this, then will check in maybe next week’s meeting

  • Lipid force field fitting

Project updates

BW

• Past week

  • lipidmaps clustering

    • With threshold of 0.65, clustered fragments down to ~1800

    • Some fragments can be very large (~300 atoms)

    • Cap fragments with H for now, can’t remember what previous issue with S was

    • Don’t bother with >70 heavy atoms. Might need to filter out full molecules from the fragment set. This is for training data. Aiming for ~4k molecules to start with

    • ~100 heavy atoms for molecules for benchmark data

  • benchmarking visualization tool

    • BW: plan to get AWS runner working, scale up to full size benchmark

    • BW: will port code into repo

  • optimizations

    • (see Slack thread)

• Next week

  • lipidmaps dataset?

  • benchmarking CI

    • ping LW or AMI if want feedback

  • yammbs PR reviews

  • follow-up on the minimization issue and re-do torsion minimizations in new environment

Project updates

AMI

• Past week

  • DDX dataset

    • 1/3 calcs with PBE0 crashed with numerical issues, e.g. SCF convergence issues. Many of those that didn’t crash may have also converged to the wrong state. JW suggested increasing memory

    • Using vacuum wfns as SCF guess may fix it. Technically challenging

    • Can also try huckel guess, which is less technically challenging

  • MSM splits

• Next week

  • Troubleshoot DDX dataset

  • Maybe train FF based on MSM splits?