Will be changing jobs. Joining QuanSight, largely replacing Scopatz.
Trying to schedule tasks before I leave. Do you need GMX on conda-forge?
MT – This is something I want, not clear to me why they’re remaining on bioconda. I assume it was previously because of CUDA deps, but maybe also organizational stuff with GMX core devs.
JRG – We’ve moved bioconda packages to c-f before, bioconda maintainers weren’t sure why we did this. Leads to duplication of packages which can confuse users. I helped with autodock vina because the maintainers were happy to have help. Not sure whether gromacs maintainers want the migration. Would recommend maybe moving CPU-only GMX to conda-forge initially, since people can use it primarily for system setup. Since there will be a combinatorial explosion of MPIs/CUDAs, don’t try to encompass that complexity initially.
MT – Could I take a stab at this?
JRG – Probably. At quansight I’ll be contributing from the Quansight side to the documentation.
Un-vendoring subpackages from AmberTools. ParmEd first, then packmol.
SB – AT clobbers dedicated packmol package, makes it hard to know which is installed.
SB
Fragmenter port is very close to done. Some remaining issues with OE and RDK. This will be closer in behavior with AT than the old OE.
About 25 PRs, reviews would be welcome, but not if they’ll take more than 10 minutes a piece.
JW – I’ll give these a skim, and will say when I’m done/ if I won’t get to anymore
SB – Main thing remaining is regression tests.
Need to get possible-stereocenter logic into OFFTK eventually. Also subsetmol functionality/having sensible molecules with unsatisfied valences.
DD - Where are these PRs?
SB – openforcefield/fragmenter, been putting PRs into the branch rather than master
IP
I’m one week in. Working on getting up to speed on the software/concepts behind it.
Did a PR for TopologyAtom.element getter.
Working on Topology refactor to support biopolymers. One part of this is to move classes to pydantic. Replaced TopologyAtom to a pydantic implementation, looking at which elements of structure can be maintained .
Will be turning aspirational biopolymer functionality doc into API examples soon, collecting more use cases.
MT
Merged first feature PR with Vanderbilt/MoSDef. Lets us use their typing/parameter assignment machinery and make an OpenFF System. Guiding them toward doing energy tests, thinking about which novel experiments/workflows we can support. Basically filling out the goals of the interoperability grant.
Porting over functionality from ParmEd. Different packages' energy tests.
Frustrated that GMX removed some functionality, not sure why, this wasn’t documented. I only learned about this from word of mouth, somewhat frustrating. This emphasizes how important it is that we continue having thorough releasenotes and a stable API. The change in GROMACS affected nonbonded energies/forces and were known to be a behavior change.
Made openff-utilities, trying to reduce code duplication.
Running into some problems with nonbonded stuff in the toolkit. SB found that the switch-width value in the force fields weren’t actually being used in the resulting System. Also some problems with the interreliance between vdW and ES handlers, and finding ways in which they misbehave
JW – Will there be cases where there’s ES but no vdW or vice versa?
SB – Resulting system could have just one CustomNonbondedHandler, or one NonbondedHandler and one CustomNonbondedHandler
MT and JW will talk to ensure that there’s a path forward for ES+-vdWHandler FFs
IP – Has anyone tried out Gromacs’s functionality (NBLIB) to do system setup using C++. Could that help us here? This is new to the GROMACS 2021 release.
MT – Haven’t looked into this before.
IP – Could this be the new home of the missing functionality from the latest release?
MT – I’ll look into this.
DD
Got SMIRKS based filtering into the benchmarking workflow, and made benchmarking release.
Sent instructions for how to submit results to pharma partners. 6/10 participating companies are done with computation. Expecting to get the first dataset tarballs this week.
Making torsiondrive executor prototype. Will pass this on to Lorenzo D’Amore.
Will be participating in a test session for Schrodinger benchmarking step with David Hahn on Wednesday.
SB – Strategy to convey that they’re comparing apples and oranges? Schrodinger does bespoke torsion fitting in some cases.
DD – Yes, industry is aware of the difference.
SB – How are we indicating when they’re getting bespoke results?
(General) – Schrodinger data is generated two ways – One where it’s forbidden from using the bespoke FFBuilder, and another time where it’s allowed to use this stuff.
Will be shifting focus to QCArchive management and development this week. Long list of to-dos, such as hessian dataset generator.
Submitted benchmark season 1 datasets. Some problems with the scale of this submission. Will be debugging this in QCArchive/QCPortal.
Added some new datasets on Friday, aniline is submitted, torsiondrives are still running.
Beginning a new project – Protein ligand benchmark automation on F@H. Will be working with John to set up protein-ligand FE calculations using OpenMM and GROMACS. Could run this after each FF release. This could be handled similarly to QCArchive automation. JC wants components to be as reusable as possible. OMM will be called using perses, GMX will be called using either PMX or public orion tools.
One big goal is to make this be a public interface, where it doesn’t involve privileged access to a particular machine.
Working with Josh Mitchell on example-deployer script, will copy examples from conda env to local directory and offer to install additional dependencies.
LW
Mostly working on manuscripts with ANU lab.
Been working in biopolymer sessions. Current working on reverse-engineering a residue-based FF using an existing polymer. Last session SB asked why we were pursuing this, which sparked a big discussion. Last weekend, I wrote down what I’d like to get out of this, which should overlap with OpenFF’s goals as well.
JW – Can anyone help LW craft QM dataset submissions?
SB – QCA queue will probably be quite busy for next month+, for Sage release datasets.
SB – Interested to know more about the needs of this data – Functional form, some description of molecules to be submitted, what kinds of geometric/chemical coverage.
LW – Haven’t made one of these datasets before, would look forward to talking about this.
PB
Worked with constructure to build a dataset for biaryl molecules without orthosteric clashes. Initial reports of WBO values are almost all around 1.0-1.1 (with a handful of outliers), but we were hoping that this would be 1.0-1.4. Will come back to this once there’s capacity in QCA queue.
JW – Would it help to have “range cap” functionality for WBOs? (eg, “interpolate betweel 0.9 and 1.2, but be flat otherwise)
PB – Waiting for more data to make this call.
SB – Hoping to get a stronger signal from this set – The initial group of molecules was somewhat small and the correlation could have been spurious. So more data will show this nicely.
PB – Could have an experimental line of force fields so that we can work on this without all-or-nothing decisions on whether to include WBOs in main FFs.
Worked on theory benchmarking datasets, doing a lot of reading on options.
Joined biopolymer working session.
JW – Going to open up Infrastructure space to the public. Also will move core-developers coffee to main calendar.