Michael Shirts (chat): One question: what about loading PDB's from AlphaFold2/3 or other protein folding codes? Many people are now getting their initial protein structure to start simulations.
JW – PDBs are a giant mess and no one follows residue or atom naming. We’re creating a database of how things “should” work (based on Chemical Components Dictionary). That’s gen1.
Gen2 will actually be useful, because we know no one follows these conventions. We will support databases with synonyms.
I haven’t looked closely at AlphaFold outputs, but we will be able to publish a synonyms database to support these
MS – I have a student who might be able to help compile this kind of information
JW – see slide 4 for an explanation of how the workflow will work
MS – Less big question (can give long answer offline later): how difficult would it be to run Evaluator for optimizations on the national computing resource GPUs?
JW – short answer is “it’s in progress, it’s just really hard.” Let’s talk more later.
DC – Haven’t progressed smee work since JH left in Oct, but can present his final work.
MS – Should build in flexibility in case things overrun
(DM + JE will figure out how to get an openFE benchmarking update on the schedule)
Q&A
MT – External folks have been hearing for a long time that we’ve been making a protein FF, graph charges, etc. What is the channel where they should be looking for updates on this?
MS – I should be working on a paper on graph charges, currently in brainstorming phase.
DM – Graph charges are available, more info coming in preprint. I summarize protein FF as “more difficult than expected, we’re not going to release something until it’s better. We are working to make the amber port more functional”
CC – We’ve been working on the livecoms article on benchmarking protein FFs.
JE – We probably want to have these answers posted somewhere publicly available. This is why I want us to spend some time focusing on our communications strategy at in-person meeting. Would a blog post in the meantime be a good place for this info?
MT – That’d be better than the current situation, but would still leave room for improvement. But would prefer that we have a single-stop answer for outside folks to understand our progress and timelines. Also, would love something that is updated somewhat frequently, since our communication artifacts are frequently just point-in-time. So maybe something tied in with iteration planning or ad board updates.
MS – This is a good discussion. Preprints are kinda an infrequent way to give throrough updaes, but would be good to have something more frequent and actionable.
JW – Maybe we could get the zenhub roadmap publicly visible once we get it to be correct+updated regularly.
DD – Any use of alchemiscale that OpenFF is needing in the next few months? We have an agreement to support OpenFF usage.
JW – No update at this time, lots of science team turnover.
CC – I’ll reach out once we have a protein FF candidate to figure out next steps.