General updates | JW – I’ll be offline Friday Trello update Pint update JW — Re: Water models - I wonder if there’s some better way to state level of endorsement, but I’ve been vacillating for weeks with no conclusion so I think perfect is the enemy of good here. Looks like there may be issues with authoritativeness/provenance, so maybe we fill in the author field with a DOI or some other identifier. MT – Decided to pause on this. I think there are two possible ways you can help here, both are complicated. JW and I go back and forth and we decide on a canonical source. Most defensible seems to be using table from old paper, but I could see us getting disputes about using this sort of source. Another option is I just make a choice solo and see what happens Final choice is that we reach out to people and poll to identify the authoritative source.
JW – I’m thinking we could also just copy from someone else’s data - like tip3p_openmm_1.0.0 and/or tip3p_gromacs_1.0.0, then we have less surface area to defend MT – I’ll pause on this for now, may add tests against openmm assignment
JW — Re: Exscientia meeting: Thanks for taking great notes. Releasing “something like rosemary” - Not a bad for exscientia and others is a good idea. Could also include example FFs with more recent SMIRNOFF spec changes. Agree that there’s a tough call on whether to put in vsites, also a tough call with librarycharges vs. graph charges. JW – Mind if I put your writeup on confluence? JW – Big questions are how we handle the uncertainty in partial charge methods/vsites, and where we’d put the FFs. We could have different versions of the candidate FF for the different outcomes. MT – I guess there are more questions, like “What if chapin’s fit goes really well next month and we release without graph charges?” and “will there be vsites?”. If we add vsites, we’ll need to update recharge, evaluator, etc. I don’t want to stress-test all my infrastructure wtih vsites that we probably won’t have in rosemary. So I think it’d be on the science team to provide a FF with vsites. JW – Let’s assume that we won’t have vsites, and that there will be a big delay in the infrastructure if we add them . JW – I don’t think it matters whether we ship the test FF with proteins-specific torsions. JW – Big question may be graph charges vs library charges. I’d say this is currently unknown, so we could make two variants for this.
MT – Other question is graph charges. I think most of the questions are whether we want it to be a parameterhandler plugin, or a toolkitwrapper, or something else. MT – All they care about is vsites. MT – I’ll tell them that Rosemary almost certianly won’t have vsites, but that we can provide test force fields that look like Rosemary if they want them.
JW — I tried running tests locally on the ForceBalance PR for OFFTK 0.11, since some were skipped on CI. Two failures, I’m going to dig in to find the root of these. JW — Sounds like you’re also seeing OpenMM 8 rc problems? Ivan just reported one as well. https://github.com/openmm/openmm/issues/3832
|