2020-10-14 OpenFF Benchmarking : Tools for Partners Meeting notes

FF choices

David H.

• the following forcefields are desired:

  • smirnoff99Frosst

  • openff 1.0.0 1.1.(0,1) 1.2.1 (latest)

  • gaff2.1

  • OPLS3 (probably not possible, but of high interest to industry partners)

    • is it possible to get an older OPLS running

    • would have to be a separate software path; we can establish a set of parameters everyone should use, David H. has experience here he can provide

Input format

• 3D SDF likely as the canonical input format

Software approach

Josh

• JH: Should have its own repo

• DH: Agree; should have its own repo

• TG: Wanted to do single points for TorsionDrive quite often; would tend to group this tooling with analysis approach

• DD: Agree on its own repo, will allow us to move quickly, have its own issue, PR space; will be duct-taping together components imported from QCSubmit, benchmarkff, etc.; generalizable things can make their way elsewhere if desired when we make things prettier

Missing components

David D.

• DH: if we wanted to compare torsiondrives, then we would need new functions

• DD: Optimizations easy by comparison; hoping this remains out of scope; will get their opinions on 10/23

OpenEye warnings

David D.

• How can we make the loud OpenEye license complaints go away? Technically not a problem, but optically can be confusing or a turn-off for users

  • TG: comes from the C executables; hard to make it disappear in the OpenFF toolkit Python layer

  • JH: if we don’t install OpenEye, won’t get the warnings

  • DD: can we remove OpenEye as a hard dependency?

  • JH: Fragmenter pulls needs it and pulls it in

  • DD: Can we remove fragmenter as a hard dependency, since we only use it in a couple places?

  • TG: I think it’s already imported where needed; could remove as hard dependency already

  • JH: will remove Fragmenter, OpenEye from hard dependencies next conda release

“Readily usable”

David D.

• What would count as readily usable data structures/output?

• JH: Have you ever used yank? produces a bunch of Jupyter notebooks, featuring rendered plots

• DH: Figures are good; data formats should be simple

  • CSV files work, with column definitions

  • JSON is readable, but a bit more complex