2024-06-11 AMI/LW Check-in

General updates and discussions

• Ab Initio fit

  • Lots of numerical issues

  • Weighting confs by distance from minimum caused issues with dividing by 0

  • Getting huge changes from MSM, trying to modify penalties

  • ddE, RMSD worse; TFD better. Dihedral ICRMSDs worse?

  • PBehara tried combining ab initio with opt geo targets

  • Loss function goes down by ~10k, starts at 70k and goes to 56k

  • Parameters restricted from changing too much

  • ddE MAEs

    • Sage: 1.95 kcal/mol

    • Most penalized: 2.05 (pretty much just MSM)

    • 2.09

    • Least penalized: 2.13

  • Simon has fit ff using smee to ab initio targets – SPICE

  • Next steps:

    • Reach out to Lee-Ping for advice

    • Easy things to do in background

    • Benchmark Simon’s FF

    • Re-fit using smee, using ab initio targets

      • Lexie will send LW targets

• Looking at 4-coordinate sulfur data

  • Looked into different ways to fragment and build the dataset. Tabled as it overlapped with BW’s work

• Roadmap: https://docs.google.com/presentation/d/1CYhur-N-M31WKa41VyGZa-3vlV6sinWepod2V-tumNI/edit#slide=id.g2c8f740d84f_2_22

• Overall budget: 40 weeks

  • NAGL

    • Scoping out new level of theory, looking for or generating data, looking at new NN structures, etc

    • 30 weeks

  • Virtual sites

    • 8 weeks

  • Valence work?

    • Training dataset work

    • Relative energy fitting / conformer energy ordering problem

      • 2 weeks FT

• Protein + nucleic acids

  • vdW re-fit + valence re-fit

  • fitting water model to it

  • benchmarking NCAAs/PTMS

• Plans for next week:

  • Try to wrap up ab initio project before going on leave