2021-06-09 Industry benchmarks meeting notes

Kaushik testing of Schrodinger pathway

Jeff+David H

• Propose sending Kaushik the whole repo zipped; provide instruction on

Needs for partner workshop

Jeff

• No requirements
  

Updates

• LD: last week working session with DD; addressing outliers in benchark analysis

  • tried some initial statistical methods

  • Problem is that, when there’s an outlier that has energies that are totally wrong (1000s of kcal/mol), this screws up aggregate statistics. I haven’t found easy ways to remove these outliers. But setting another max deviation cutoff could make these sensible again.

  • Worked one making plots for analysis, this is nearly done. There’s an error with them currently, but it may be good to discuss here.

  • JW – Two suggestions:

    • Cumulative histograms for each ridge plot

    • Maybe a matrix of heatmaps? Eg below

• DH – Not much to report.

  • Kaushik’s email about offline installation of schrodinger, which I just replied to with a tarball

  • Nothing from Christina Schindler

  • Two Qs about season 1

    • Any explanation for why 1.3.0 is worse than 1.2.1?

      • JW – Not sure why. It may be because we targeted improvements on specific functional groups, and degraded performance of the FF in other areas.

      • DH - 1.3.0 definitely does better on some areas, but overall does worse than 1.2.1

      • JH – The parameters could have “wandered” more and more in each version from the original GAFF/S99F values. These could indicate problems in our fitting method that just took several generations to appear, but weren’t specific to the way we trained 1.3.0

    • Should we do Sage in the benchmark?

      • JW – I’ll be working full-time on biopolymer infrastructure, which is behind schedule. And DD will be working on F@H for most of his time after June. So we’d be able to provide limited infrastructure support.

      • DH – We can look at running this just on the Janssen set, or in a more limited capacity.

      • DH – I’ll talk to GT about this

    • When is sage release expected?

      • JW – 1 month-ish? Not sure about exact timing.

      • DH – The sage release canadiate 1 is available at

      • (General) – We probably don’t want to do the release candidate in season 1, so we should look for more clarity on Sage release timing.

• JH: nothing from my side at this time

• DD:

  • Nothing new. Had some to-do items but was sick yesterday.

  • Have several followups from previous meetings, still need to do these.

  • Some plans for openff-geopt and executor, but haven’t done that yet.

• JW: Nothing to report.

  • focused on workshop materials