2021-01-25 Core Devs meeting

Owned by Jeffrey Wagner
Jan 25, 2021
3 min read

Roundtable updates

  • DH – Worked on benchmarking.

    • Push backt o benchmarking discussion afterewards)

    • DD – Something tricky with analysis commands?

    • JW – Match-minima?

  • SB –

    • Worked with MT on conda-forge migration. Realized that we have to move openmmforcefields as well. Updated to have correct namespace and fix CI. Turns out I also had to fix QCEngine.

    • Made new library called constructure –

      • Can be used for molecule generation, eg automated chemical space exploration.

      • DD – Would be cool to automatically find gaps in current sets/parameter coverage and send them into QCArchive.

  • MT –

    • Been working on conda-forge migration. SB and JRG helped a lot. OFF Toolkit, forcefields repos are on conda-forge.

    • Last phase of migration is tidying up downsteams (eg openmmforcefields, qcengine, forcebalance).

    • Updated lots of packages from Travis to GHA.

    • Hunting down loose ends in documentation – Things that would fly under the radar of a dedicated documentation writer.

    • Worked on System, worked out interactions with Vanderbilt. VU has been slow on deliverables/noncommital to project, so I’m tkaing more of a “take it or leave it approach”.

    • Made GMX topology writer

    • Will be working on a neutral Topology representation, that can either handle OFF-style “cheminformatics” topologies, or other ecosystems' atom-type-based topologies.

      • JW – This will be helpful for me to know how to do a OFF Toolkit Topology refactor

      • SB – Will be helpful for biopolymers, meeting scheduled for mid-Feb.

  • CD –

    • Continuing on toolkit equivalence testing. I’ve been loading the same molecules using the cheminformatics backends. Found issues regarding bond orders around aromatic rings. Differences in stereochemistry detection/assignment. OE promiscuously adding protons to sulfates and phosphates. Some differences about SMILES outputs.

  • DD –

    • Working on benchmarking. Did lots of iteration on the software, gathered documentation and instructions for kickoff meeting last Friday.

    • Will be having partners deploy QC* infrastructure on their computing resources, will send them ~600 molecule dataset to set stability/performance expectations.

    • Will be supporting deployment/science issues on the #benchmark-support channel.

    • Announcement and scientific discussion will take place on #benchmarks-partners channels.

    • Will be deploying new conda environments for our QCArchive workers.

    • Putting together new submission set for pepconf.

    • Worked on changing QC jobs to DLC instead of TRIC, and reset=True option. appears to result in much faster optimizations.

  • JW –

    • Worked on benchmarking.

    • Made single file installers, worked on hardening/refactoring validation step.

  • PB –

    • Worked on WBO fitting.

    • Tweaked forcebalance setup on UCI clusters. Now running it doesn’t make 40GB output.

    • Looked at how priors+other settings affect outputs. Worked on testing out some chagnes to QCEngine. (ran an errored pepconf job and reran with new settings – Still running, can report later)

    • This week I’ll be continuing WBO work and continuing preparing Genentech data.

      • JW – Is Genentech dataset the same as Swope’s benchmarking dataset?

      • (General) – No

    • SB – Are changes to QC input files, is it changes to code, or input files?

      • PB – input files

      • SB – Great. I’d like to make sure we’re not doing “production science” with locally-modified code.

    • PB – Are QCFractal task objects immutable? Can I change parameter values for coordsys from tric to dlc? I was unable to change it so I had to make a copy of the task.

      • DD – This is where pydantic can make thigs difficult. I’d call task.dict and then modify the dict and pull in back in with something that constructs from dict.

      • PB – Is there a friendlier way to get the output?

      • DD – I’d use the package “JQ” – It lets you treat json like a dict, and allows grep/awk-style modification to JSON.

    • Break for 5 minutes – Return for sprint planning at 25 past the hour