2026-01-29 Meeting notes

Non-QCArchive data?

MT

• FC: Not important now, I take data like this and make a “fake QCArchive data” out of it. Could be useful in the future

More QCArchive datasets?

MT

• FC: Current benchmarking dataset is good. Worth thinking about
  
  

Other optional features

1. geomeTRIC-based minimization

    

   1. Simon has a geomeTRIC re-implementation that might be worth considering?

   2. FC exploring these two as non-default engines, keeping OpenMM as default

   3. FC still working on geomeTRIC branch

   4. FC will keep working on it

2. Updating torsion metrics

   3. What’s more to do?

      1. Re-run YDS PR on larger dataset. Add TorsionNet500 to YDS (separate PR), run it on that dataset

      2. Will ask Lily if TN500 should be re-computed at OpenFF level of theory

      3. FC uses GPL code in his plots, to add to YAMMBS/YDS or not?

3. Hydrogens in RMSDs

4. Outlier detection

5. Run-to-run variance:

   FC: Using 2.3.0 for now
   MT: Sweeping it under the rug
   MT: Need lots of older FF releases for science?
   FC: Generally only look back to the most recent mainline great

Internal vs public-facing

FC: Would be nice to be more public-facing if zero effort/free. Have a student working on it (benchmarking torsions generated from Presto/BespokeFit 2). Only “requirement” is that it would be nice if docs/installation instructions were up to date.