July 2026 Infrastructure Update

Hi everyone. I'm Jeff Wagner, infrastructure lead at Open Force Field, and this is our Q2 2026 infrastructure roadmap update. We have a grab bag of minor updates for you this quarter. Uh the first is that we've had two minor bug fix releases, uh one of tool kit, one of interchange. Uh these should come through undetected, no major um breaks or new features in these. We also are aware that 2 weeks to 1 month ago we had a brief issue where environment solvers would fail on systems with limited resources. Uh so this would be something like a Linux machine with like 4 GB of RAM or something. Um it would it would try to solve the versions of the dependencies to install and and the solver would run out of resources. Um we relaxed a number of our packaging pins to resolve this, so things like down pinning from Python 3.14 and and down pinning other small things make the solver's life very difficult. Um we removed a few of those and the problem seemed to go away, but if you were encountering this, please get in contact with us. We're happy to announce that we have Python Pydantic 2.13 and higher support coming. And uh for many of you this doesn't matter, but what this does translate to, which does matter, is compatibility with Python 3.14. Similarly, and this is a big thanks to Mike Henry, uh initially starting this work under the banner of Open Free Energy and now continuing it under uh the OMSF Eco Infra Team uh ecosystem infrastructure team, is um fresh builds of Amber Tools 24 and 25, and actually Mike's working on 26 right now. But importantly, that updates them to work with the common dependencies uh that most of conda-forge keeps up with. And also, we now have builds for Python 3.14. So I I that might make a lot of people in this call's lives easier if uh you've been needing to make environments with that. Past that, we have been applying for some grant funding from private philanthropy. This is through the Open Source for Life Sciences Fund. Um and OMSF is submitting a number of these. A couple of them involve OpenForceField, but the infrastructure team in particular uh is submitting one for uh improved workflows for modified protein handling. Uh I'm thinking of being the PI on this one, and Ashley Mitchell, um one of the key personnel. And so, you know, we're we're coming out with this force field that's going to have parameters for proteins. And now that it seems like the big bottleneck is how are we going to ingest these PDB or mmCIF files? Um it's difficult. We've been showing it in workshops, but the tooling could be improved and maybe uh given a more natural interface with AI to solve sort of these tedious um atom naming problems. And one thing we're realizing in the process of writing this is that something like um a light protein prep, something like protonation and maybe some heavy atom replacement, could become in scope for this grant. So, that'll be submitted soon. Cross your fingers for us. Uh finally, I presented at the first OMSF stack tour. This was in San Francisco uh two or three weeks ago. And I presented the introduction to OpenForceField and PTM workshop material uh originally made by and presented by Ashley Mitchell. I'll go ahead and drop the uh link to the repo and the recording for that in the chat cuz that was a really great one. We do have two more of these coming up. One in August. That's going to be in Chicago adjacent to the ACS meeting, I think August 28th. And another one in September that will be in Boston. So, we'd love to see you there. Uh following up on an item from last month uh or last quarter. The OpenMMForceFields 0.16 release has come out. Uh importantly, this includes support for Smirnoff proteins and virtual sites. Uh this is something that didn't used to work or if it did work, it was very hacky. Uh now we've we've got quite a bit of testing put in for these, so we're confident that these work. Uh I know that some of you access OpenFF Force Fields through the OpenMM Force Fields package, and this means that you can try out the Rosemary alpha now. Uh we don't have any virtual site force fields in production. So, when we do, boy will you be ready with this new release. Um but other than that, there there may be some uh behavior changes that annoy some of you, but they were bad behaviors to begin with. And I'll fight you about that. Uh and additionally, faster performance on large molecules. So, if you were hamstrung on large small molecules or if you were looking forward to trying proteins, uh now that's possible. Finally, this was a lighter month, uh lighter quarter. And this is because we've been rebalancing effort from the infrastructure team towards science goals. Uh so, Matt Thompson has been working on uh our new property calculation engine called DimSim, and Ashley's been working on uh benchmarks of Rosemary on cyclic peptides. So, you'll likely hear from Lily about those in more detail in her presentation. Thank you very much.