2021-10-01 QC meeting notes

Updates from MolSSI

Ben

• Will be taking production QCArchive down briefly for update; will include database migration for query indices fix, includes fix for duplicate tasks

  • DD: if you can, please notify in #qcfractal; helps avoid surprises from users

Compute

David

• We have QM workers on PRP, Lilac, MM workers on PRP, Lilac, and specific QM workers for Pavan’s Single Points dataset on PRP (pending)

• JH: was expecting dedup on fragments dataset; didn’t really see much

  • BP: if you’re using specs with null in method or basis, these won’t deduplicate

New submissions

• CC: created openff-qcsubmit issue articulating need:

• JH: can implement what’s needed; recommend we make a small test submission once we’ve got it in

• PB: working on submission PR for openmm/qmdataset draft submission:

  • JH: will require unmerged work on openff-qcsubmit HDF5 branch

  • PB: using that one here, running into issues

  • JH: have some uncommitted changes to push yet; there are issues that need fixing in toolkit or QCElemental

  • PB: can I use RDKit or OpenEye?

    • JH: probably use RDKit for reproducability

  • JH: getting validation issue with some of these systems with QCElemental mol

    • toolkit passes empty list for connectivity, gives validation error

  • JH: will put draft PRs up on either toolkit, QCElemental, or both; DD will shepherd them forward

User questions/issues

• PB: how do I start up a server with previous state?

  • BP: need to use restore, probably without init, on backup export file

• PB: having issues with HPC use in terms of scratch space usage; disk filling up, getting slapped by admins

  • DD: happy to do a working session to troubleshoot, find a better approach

Science support needs

• PB: is there a plan to submit all old datasets with STANDARDSv3 compliance?

  • an on-demand approach, in which a STANDARDSv3 compliant version is created when someone asks for it, would satisfy

• PB: question for Josh: can we include a try-except for retrieval to get past bad records?

  • JH: is that using latest?

  • PB: 0.2.3, yes

  • JH: are these validation failures?

  • PB: one is an inchikey issue, no inchikey on an entry

    • when I tried to download industry benchmark dataset, get some failures with OpneEye; have to use RDKit

    • PB: expect that both toolkits should be able to read the molecules; get stereochemistry errors with OE