| |
|---|
Roundtable updates | DH – Last week, did lots of benchmarking work. Worked on CLI for benchmarking analysis and integrating whole pipeline. Worked on benchmarking best practices manuscript. Will begin first run of pharma partner benchmark. Worked on PLBenchmarks repo. Decided that git LFS will let me keep API and data in same repo, which is nice. Will work with MT on release process. Still considering whether to change PLBenchmarks repo name. We had talked about it but there wasn’t much feedback. I’ll take the current possibilities and open a poll on Slack.
SB – Mostly scientific studies. Lots of BCC refitting, some work with ML models to predict AM1 partial charges and BCC. Some changes to recharge API to enable better integration. Some studying about best practices in RESTful APIs. Conclusion seems to be that there aren’t best practices.
MT – Pushing on many tasks. Switched openff-system over to use implcit units while unit discussion is underway. Working on getting system to export to openmm. Can now get an entire parameterized system exported to dictionary. Need OpenFF Toolkit Topology to have working to_dict and am working on a PR for that. Helped test c-f OpenMM package. Tested out unit models, tried idea discussed in previous unit meetings. It seems like the wiring-up-different-public-and-private-attributes/setters may be too difficult to meaningfully implement. MT – Not taking much time off for holidays. Will be working before Christmas and between christmas and new years.
DD – Lots of work on benchmarking. Worked closely with JW, DH, and JH. Goal was to get full-scale testing running this week. Worked with Bill Swope on protocol document, testing, environment setup. Developed pathways for persistent vs. nonpersistent server for benchmarking QCF server. Expect to work with Thomas Fox, Bill Swope, David Hahn this week on benchmarking trial runs, documentation refinement. This week, will meet with AWS spot team. Need to prepare written proposal in advance so they can prepare for technical discussion. Need to help them justify having them accept our project into the pool of groups that can use AWS resources for “the public good”. JW – Need support on this? You seem to be the primary driver on this interaction DD – May loop you in on review for documents/messages we send to them about our project. SB – Is Spot intended to just be for QCF? Or would other OpenFF workflows be run on this?
JW – Worked with MT on cutting 0.8.1 release. Went pretty smoothly except that we hadn’t previously had tests involving mdtraj, which is needed for Topology.from_mdtraj Worked a lot on benchmarking workflow. Found three major categories of RDKit stereochemistry issues:
PB – Worked on WBO interpolated parameters. Looking back on results based on SB’s comments during ff-release call. Doing another fit using correct FF file. Objective function is now below 1.3.0 but above the fully interpolated FF. Also checking performance on Lim+Mobley benchmarking set, may need help from DH or others on getting it running again. Working on OFFTK #720 (on roundtripping to/from qcschema). Still looking into problem statement and different options. Could use try/except statements to see whether the CMILES is present in various fields. SB – How far along is automated benchmarking? Could that be used for this case? JW – This would look like the last few steps of the benchmarking workflow, but with a custom OFFXML. The current workflow is a bit rigid, but we’ll need to expand the MM FFs to include s99F, so this may be a good opportunity to generalize the entry point DD – This will need to work with non-released FFs, which require some finessing with QCEngine. PB – It looks like benchmarkff can take a custom field for FF files DD – For new benchmarking workflow (openff-benchmark), this field isn’t exposed. But we could do some work and figure out how to inject it. PB – I’m able to pull data from QCA, but the MM minimization script is generating some errors. JW – Let’s give this a try together – If benchmarkff can work with WBO with some small changes, then it should be possible to move forward with that. SB – Would like to make openff-benchmark a good general-purpose benchmarking tool that we can use moving forward on subsequent releases. DD – Agree. Though there’s some friction about the stability of the expected behavior and the customization desired by internal users. JW – For the next few weeks, I’m opposed to adding complexity to the master branch of this repo, since this software absolutely needs to deploy and run on industry machines. Happy to have more complexity added on a branch, and eventually master in the future (once industry users verify that this can run) DD – Agree SB – Agree. Would like to ensure that the broader goals remain on the table in the long run though. JW will work with PB to get becnhmarkff running again with more current OpenFF toolkit/OpenMM versions
|
Benchmarking needs | |
0.8.2 release? | JW – Purpose would be to get double bond stereochemistry fix out to industry benchmarking testers MT – Don’t think there would be a problem with unusual release cadence. DD – Agree Decision: We’ll merge double bond stereo fix and cut 0.8.2 ASAP
|
C-F migration before holidays? | JW – Con: Adds complexity and potential fires (10%ish chance) before people leave for holidays Pro: JC wants to release openmm-forcefields on C-F, which requires OFFTK on conda-forge MT – I’m up to wait. Very small chance of emergencies, but not worth the risk. SB – Partially agree, but don’t see that much risk from this step. JW – Not so concerned about new environments being broken, but rather deployment issues like having openff-toolkit and openforcefield in the same env SB – Is there run-constrained defined in any recipes, to prevent both from being present in the same env? MT – It doesn’t seem to be there. Though it is there for ambermini
SB – We should make a note to add run-constrained. DD – Also concerned about deployment-related complexity Decision: There’s not a pressing need for this before the holidays, so let’s not risk it.
|
C-F openmm size | |
Sprint planning | |