2024-08-27 BW/LW Check-in

General updates and discussion on projects

• Torsion fitting

  • Try 1 kJ/mol

  • Status: Updated torsiondrive minimization script to include 1 kJ/mol restraint on non-torsion atoms, re-run and see if anything changes?

  • On minimization

    • 10-1000 kJ/mol dihedral restraint didn’t change anything

    • Add a different restraint, a harmonic one, force the angles to eq values, and see what other force has blown up. This would explain why these angles are so high in energy. <=2 hours, would provide information and Shirts/Swope would find interesting. Maybe angle at 0

    • Torsion shapes – let’s resolve angle stuff first

      • Filter out drives where other bonded forces make significant contributions to energy?

      •  

• benchmarking infrastructure repo?

  • Planning update tomorrow in IB meeting

• smee

  • Running!! Epoch 174, after 6 days.

• datasets

  • BW: need to get back to FPickard, MShirts about updates

    • Julienne H might be involved with this

    • Updates on the curated dataset – ping MS on ownership

    • LIPID MAPS fragment database

      • have database of all fragments

      • 474k+ fragments. <80 atoms = 128k fragments

        • 1-5k dataset

          • RDKit fingerprints

          • checkmol grouping

          • parameter typing

      • lipid maps paper for gauge of groups to hit.

  • LW: another sulfur dataset: Increasing dataset coverage

    • Most progress: updating curato.

    • Huge dataset, even with 80 atom filter

    • Manually picking molecules out may be quicker than clustering

  • Alkane dataset has completely finished – pass to MS if interested

• chemical perception work – re-fitting force fields

  • Died, restarted

  • Maybe if not moving to /tmp, will run temporarily

  • But should finish in <=2 weeks as 50 iterations

• infrastructure work

• Genentech

  • https://docs.google.com/document/d/1ZJmV9C1IwfRf0UVz6b8Mr_6DvKsQIgfvLhJAScY8K9E/edit#heading=h.97uz1di14ngk

  • https://pubs.acs.org/doi/10.1021/acs.jcim.0c01041

  • https://pubs.acs.org/doi/10.1021/acs.jctc.9b00239

• Performance reviews

  • Might be coming

• Meeting scheduling

  • These past 1-on-1 meetings have been very science- and project-based

    • Alternative: move to science team meetings for project work, keep 1-on-1s to less project-work

Roadmap progress

1. Infrastructure: 8 weeks

   1. Still 10% of time. Happy to increase

2. Torsion multiplicity: 6 weeks

   1. Been working on that constantly-ish the past two months. Some of that in background

3. PDB fraction: 8 weeks

4. besmarts: 4 weeks

   1. Trevor’s still updating code + examples

5. smee/alternative functional forms: 4 weeks

   1. Spent ~week on it so far

6. dataset curation: 7 weeks

   1. Progressing along

7. standardized benchmarking: 3 weeks

   1. Progressing along