CC – Tea time on Tuesday afternoon at 4 PM US Pacific
Individual updates
DN
Last Thursday I had my check in with Karmen - She pointed out that there’s more that I could be doing. So I wanted to check with you all about whether I could do anything to make your life easier/if there’s anything that Karmen used to do that I could pick up doing. I’m open to input at any time.
With JW’s vacation, I’d like to make sure that we keep things running for 2 weeks. So we should keep meeting notes getting taken and at a high quality. In the future I want to take more of JW’s work and help run meetings.
JW – I really appreciate your stepping in in the ff-release call to end the discussion about water models and external ffs. It was a small action that had a big impact. I’ll keep thinking about more detailed requests in the future.
MT
(bad connection)
Interchange workshop
Went well, preparation was a timesink but most objectives were met
Low attendance, some engagement from live users, no offline engagement yet
JW – I think that engagement is a long-term effort. We’re building up userbase that will cross-pollinate in the field and grow massively
MT – I don’t think this is an ok situation - We don’t know what people really need, and we’re developing speculatively. I also keep hearing that adoption is slow-growing but inevitable, and I’m not sure at what point it will really “happen”.
DD – Do you think that having direct customer engagement is important for the health of the project?
MT – Yes - I’d call it “adoption” but we probably mean the same thing.
DD – I think it’s probably worth chasing these people down and developing a personal relationship. This will help you gain information about their specific needs. Gaining adoption in one partner will lead to bleed-over into other partners.
MT – Understood.
JW – I see it as a not great outcome if we have engineers doing external communication since this makes it hard to do focused work. But if this is a skillset you want to develop then go for it. But I’d rather have DN leading external communication.
MT – I’m not sure that the org can reach adoption under the current operational model.
DD – I strongly believe that unfiltered customer feedback is the best way to get adoption. Even if that means screensharing on their projects. This builds a relationship that brings benefits and candid communication in the future.
MT – I think I have a hazy but somewhat useful picture of why our user adoption is lagging - I think people are waiting for Rosemary so they can do proteins and small molecules. Or there’s something that’s not directly a tooling issue (installation, documentation, etc) that’s buried here. I think that the barrier to entry is pretty low, so I’m curious as to why adoption isn’t proceeding as fast as we’d like.
JW – Regarding gathering needs/reasons for slow adoptions- We’re going to be restarting the annual survey this year, and I’ll make sure to add questions about “what’s keeping you from having greater adoption?” and “would you like to work with an OpenFF dev?”
Water models
Another attempt at gaining consensus on prior proposal of adding water models to openff-forcefields
Failed to obtain consensus or stay on topic of decision points, bailing out
JW – I don’t think the scope creep was your fault. I am going to meet with DN and LW and discuss path forward.
MT – I’m not going to proceed with this regardless of your decision.
PB – I thought we agreed to include water models?
JW – I don’t think we agreed to do this.
MT – My understanding is that we didn’t reach consensus on some major issues, and we didn’t even get to discuss several others.
PB – If we’re using the same values as in a paper, where do the differences come from?
Scheduled meeting with Chodera + CCG, reschedule for this week due to another attendee cancelling
Was traveling for a conference since 9/15 through 9/27.
Protein-ligand benchmarks
demonstrated pulling alchemical networks through FastAPI server over HTTP, roundtripping them out of Neo4j - now also have submission through API server working - now working on `fah_alchemy` `Client`, which is the user interface for the deployed system - we did not make the 10/1 deadline; I will be proposing a 1 month extension at tomorrow's protein-ligand benchmark, deadline 11/1 - worked with Ivan on nonequilibrium cycling protocol in perses
also have ProteinComponents roundtripping
JW – I talked a bit with OpenFE, we’ll be working together on the next generation of protein loading
QCArchive
Dropped PRP workers, they were idling; still have Lilac pre-emptible running
DD – PB, are we done with SPICE?
PB – Yes, the remaining ones seem to be mbiis errors
PB
Worked on revisions to Sage manuscript. Answering some Qs like from OM - Like starting point FF was 1.3.0 instead of 1.3.1, so we’re not sure if it includes sulfonamide fix. So I used MT’s ligand-in-water example to test whether the geometry is fixed (thanks MT for the example!).
A schrodinger person raised an issue on the SPICE set repo about downloading issues (the QM specification was randomly set incorrectly during download). So I set the QM specification to be hard-coded instead of “default”
Started looking into BSwope’s recommendation to use energy levels as fitting target, so we could use information from optimized confs instead of just geometries for the fit. This is still a WIP.
DD – More details?
PB – In a FF release meeting, BSwope showed examples where ligand conformers optimize to almost the same structure in QM and MM, but their relative conformer energies are different.
DD – Thanks
PB – Did anyone here use the “ab initio” target in forcebalance? I know TG uses it. This sounds similar to what BSwope was asking for.
DD – I might recommend talking to LPW.
CC – I haven’t used it, but I think the point is to use a single-point energy, right?
PB – I think TG modified this to use the optimization trajectory instead of just the optimized structure.
JW – If we’re comparing rank orderings, does that give a non-smooth scoring function?
PB – BSwope’s method doesn’t just compare list orderings, it’s ddEs, so it is smooth.
CC
Submitted draft of abstract on protein FF effort for Biophysical Society Annual Meeting in February. Abstract is here:
Please submit revisions to me before end-of-day Tuesday
Followed up on additional chemistries (suggested by CBayly at FF Release meeting) that are assigned different torsions for chemically equivalent atoms with delocalized charge
TL;DR phosphates and sulfates are assigned correctly, amidiniums ([*:1]-[#6:2]-[#6X3:3](=[#7+1H2])-[#7+0H2]) are not
Still running protein parameter fits on TSCC, slow progress. Seems to be stuck on diagonalizing Hessian after first step for >3 days. Is this normal?
PB – I recall this taking >10 steps to find… 3 days is a long time. How many workers are you using?
CC – 10 workers at a time. Any more than that I get an idle timeout, since additioanl workers aren’t assigned any work. I think there’s a small number of jobs that are taking a long time and holding everything up.
PB – Maybe we should do a call with TG and hear if he has any tips for speeding it up.
JW – May be useful to try Simon’s inputs (without peptides) to see if the timing gets a lot faster. That would help narrow down possible causes
CC – That’s a good idea. I’m pretty sure that the peptides are getting covered by librarycharges
CC – Also it’s possible that ELF10 is taking a long time
JW – Didn’t SB use ELF10 for Sage fitting?
PB – No, he didn’t.
CC – Doesn’t FB cache charges?
PB – I’m not sure - I noticed that things take a long time with toolkit 0.11.X
JW – Which OFFTK versionare you using CC?
CC – 0.10.6
JW – Ok, I think FB CAN work with new toolkit, but it may break caching logic, so that could take a long time.
JW –
Some fixes for fragmenter and OFFTK following toolkit 0.11.1 release (all just broken tests)
DN – Should we advertise 0.11.0 on linkedin?
JW – I thought we did? If we didn’t advertise 0.11.0 on linkedin, we totally should!
DN – We didn’t
JW – Let’s talk at lead team meeting - We can probably reuse the slack postings or ad board emails.
Starting work again on AM1 proton transfer prevention. May also include canonical atom ordering to stabilize conf gen
