2021-10-20 Industry benchmarks meeting notes

Retrospective Survey

David

• DD: any additional thoughts on the retrospective survey?

  • LD: no additional thoughts; happy with it

• DD: should it be sent immediately, in a week?

  • LD: would be ideal to send it after the last Sage results, but this could delay the survey substantially

  • DD: can use it as another opportunity to ask for Sage results

  • LD: also have additional results from partners, so would like to announce this as well by end of the week

  • DD: we can draft the announcement today

OPLS computation

Lorenzo

• LD: hit some issues recently with SLURM infrastructure interaction with Schrodinger tools

  • got a workaround in place, not ideal but works enough

  • currently seeing that I have less jobs running than before; suspect changes from IT have impacted throughput

  • for a couple of weeks had ~350 jobs running; now down to about 1/3 that.

  • was originally expecting 2-3 weeks; might now take months

  • At about 50% parameterized right now with ffbuilder; once we’re at 70% will begin using the results for manuscript writing

Fragmenter questions

Lorenzo

• LD: aim is to use fragmenter to reduce calculation cost of torsiondrives for molecules with problematic torsions identified previously on Sage

  • LD: appears slow, should we reduce max_conformers from 800?

  • JW: slowest part of this is probably the Wiberg bond order calculation; won’t actually produce 800 conformers for this. So ways to make this faster could be

    • Change method to am1-wiberg (instead of am1-wiberg-elf10) (this may not be a real option)

    • Reduce the number of conformers in am1-wiberg-elf10 to 1

    • Use the PfizerFragmenter method instead of am1-wiberg-elf10

•