2022-11-08 Protein-ligand benchmarks meeting notes

DD : fah-alchemy - current board status

• 1 week out from 11/15 deadline

• fah-alchemy : Phase 1 - MVP

• @David Dotson development effort now focused on FahAlchemyAPIServer, FahAlchemyClient, and FahAlchemyComputeServer

  • Test suite coverage at 65%. Tests now feature integration tests for both ComputeAPI and ComputeClient; ComputeClient is used by ComputeService(s) to interact with ComputeAPI for Neo4j task handling

  • Focused on building out required functionality for synchronous ComputeService, which can be run on typical compute resources such as HPC to execute ProtocolDAGs

  • Aiming for end-to-end deployment by 11/15; heads-down this week on this

  • JC – A heads up - There was a recent F@H work server update. Check in with Sukrit if you need more info

    • DD – Thanks, I’m following this.

  • DS – Do you foresee needing my time? Would like advance notice.

    • DD – Not yet. If you want to start learning I could use your feedback through.

    • DS – My goal is to help you meet the deadline.

    • DD – It’d be hard to coordinate at this stage since I’m directly committing. It could be good for you to start studying the current implementation.

  • JW – Please deflect other support requests that reach you toward me in a public channel.

    • DD – Will do, I might not even see those since I’m turning off slack.

• JC – Perses atommapper broke in OFFTK 0.11.2 release, since it no longer supports molecule fragments.

  • JW – I think we never intended to be supporting this, sanity-checking is way more complex if we also need to support sane behavior for molecule fragments and full molecules

  • JC – How is openFE handling this?

  • IA – We’re not using the perses atommapper, have a different stack here

  • JC – Ok, I think I’m going to rip out much of the previous molecule class and rewrite this with a hard dep on openeye. I don’t have time to do this with RDKit.

  • IP – We don’t need a rewrite of the atommapper if we can get it from elsewhere.

  • DD – IP, did you have time to look at OpenFE atom mapper?

  • IP – I was going to bring that up here but RG isn’t in the meeting today.

IA : protein-ligand-benchmark : blockers and priorities

• • IA – Adding missing thrombin entries and then good to go. Fixing using #82. Had some missing ligands from other ligands.yaml files and that will hopefully be fixed tonight.

    • https://github.com/openforcefield/protein-ligand-benchmark/pull/82

    • IP + JC – I’ll review this.

  • IA – #83 has perses-derived minimum spanning graphs, but hitting the aforementioned OFFTK issue with atom mapping. IP, could you take a look?

    • IP – Sure. I’ll review.

    • IA – Thanks. This will depend on #82.

  • JC – What format is this atom mapping?

    • IA – List of tuples. Happy to store this in another format.

    • IP – This may not be yaml-native.

    • IA – I’ll look into this.

• • Waiting on #81

• • Waiting on one of the networks to return (PR is up)

• DD – Can we close #76?

  • IA – Yes, this is superseded by #82

  • DD – (Closed 76)

• IP – We were talking a few weeks ago about historical results and opening a repo for that. I don’t know if that’s a priority, just wanted to check in.

  • IA – BRies is working on historical data for protein-ligand benchmark. Currently private, but we’ll make it public soon.

  • IA – Also there’s question of validation dataset. But we’ve discsussed the action points for that, just need to work on it.

    • JC – This is a small subset of these targets?

    • IA – It’s more like a few solvation free energies, since those will converge well and have short runtimes.

  • 
    
    

IP : Nonequilibrium Cycling Protocol (perses#1066) update:

• IP – I just pushed latest changes to PR. I have to talk with Richard about mapping, since we may need information about what’s not being mapped/transformed. So that’s the last blocker for me.

  • IA – Could I ask for more detail - Is “not being mapped” referring to ligand component or entire system?

  • IP – I may need to clarify, but if you have information about the ligand bond transforming from constrained to unconstrained, you need to know where theligand is in the system

  • JC – In a FE calc, much of the system is the env and doesn’t change. But you still need to know how the atoms are mapped. But for the alchemical part of the system you need to know which atoms are being mapped, including with their identities. So it depends on whether we expect mappings for every atom in the system or just a subset of the system, and whether the mapping contains info on which parts of the system remain constant.

  • IP – I’ll follow up on this.

  • DD – RBFE protocol in the code live s in https://github.com/OpenFreeEnergy/openfe/blob/main/openfe/protocols/openmm_rbfe/equil_rbfe_methods.py

  • JC – These could be good things to add to the object model as convenience funcitons.

  • ….

  • IA – We can discuss outside of this meeting.

  •  

•  

MH : ProtocolSettings taxonomy (gufe#37) update:

• MH – This is merged and we have the docs building as well! Docs could look better.

• DD – Is this ready to use?

• MH – Please do, I’m hoping to iterate further on this.

• IP – MH, could you join DD and my working session tomorrow?

  • MH – Will do

• MH – Another thing: I had to make a PR against Matt’s models repo. Some question about deployment/conda-forgeness.

• DD – I think…

• JW – I’ll have MT move this to OpenFF org, and start a feedstock for this and apply for preliminary approval.

  • MH – That should be good.

• MH – I’ll open a PR soon with some changes I need and tests.

•  

Transcripts with no edits

F@H interface meeting - November 08
VIEW RECORDING - 46 mins (No highlights)

@00:00 - Mike Henry (he/him)

Now this bot. Hey, guys.

@00:04 - Diego Nolasco (Open Force Field Initiative)

Hey, there you go.

@00:06 - Mike Henry (he/him)

I'm very lucky I'm on my bot.

@00:08 - Levi Naden

Well, we'll find out.

@00:10 - Mike Henry (he/him)

I left the bot a special message. We'll see if it records before you're on it.

@00:15 - Diego Nolasco (Open Force Field Initiative)

Nice.

@00:16 - David Dotson

Oh, how's that?

@00:18 - Diego Nolasco (Open Force Field Initiative)

I don't know.

@00:21 - David Dotson

Hey, Levi. Is that your daughter? Oh, I'm cute. I realize I screwed up the, uh, hold on, guys. I messed up here.

@00:41 - Diego Nolasco (Open Force Field Initiative)

Hey, Levi. How old is she? She's three.

@00:47 - Levi Naden

Oh. She's here for a couple minutes because she wanted to see what daddy was doing today.

@00:53 - Iván Pulido

Oh, yeah.

@00:54 - Diego Nolasco (Open Force Field Initiative)

Today you'll be. I'm still one. I'm not looking with that. Would you like to. some science girl.

@01:04 - Levi Naden

Yeah. All right, then I could always back again.

@01:17 - John Chodera (he/him/his)

But how is there a Python?

@01:19 - Diego Nolasco (Open Force Field Initiative)

Yeah. Yeah. And if I can take note and then you can add this meeting to her resume.

@01:33 - Levi Naden

Yeah, her first one of her first experiences. Meaning.

SCREEN SHARING: David started screen sharing - WATCH

@01:40 - David Dotson

This is what the rest of your life looks like. I'll take that effect. Okay. Oh, that's the wrong screen.

SCREEN SHARING: David started screen sharing - WATCH

Can you guys see that? Okay.

@02:03 - Iván Pulido

Yep. Oh, well, there's a, okay. Yeah, that's fine.

@02:09 - David Dotson

Yeah, zoom, zoom makes these little windows that say what it's doing, that most annoying thing. Anyway. Okay. I think we can go ahead and jump in.

Are we waiting for, I think, Richard had said that he won't be able to make it today. I think it's what it's in your standing in for him.

@02:30 - David Swenson

I don't know if I am or if your fun is. I don't think we actually got. Was find that.

@02:39 - Mike Henry (he/him)

We forgot to do the ceremonial rock paper.

@02:42 - David Dotson

She said it should have been either either one of you. So whoever wants to take their responsibility.

@02:49 - Irfan Alibay

I can take it if you want.

@02:53 - David Swenson

Okay.

@02:59 - David Dotson

All right. Thank you. We'll go ahead. jump in. Then thanks again, Jeff for taking notes. Thanks, Diego, for the Phantom note-taker.

So we'll start at the top. So thank you again for the extension. We're one week out from the 1115 deadline.

So I'll talk a bit about the board status for now. Some things have cleared. So thank you, Mike, Irfan and David, as well as Richard, finally got settings through.

So we should celebrate that. I know I've got at the end of the agenda, just in case you wanted to comment on anything about that later.

I just wanted to celebrate that. Thanks for pushing it through. Otherwise, working with Ivan on non-equilibrium cycling, Ivan, is there a time later this week he'd like to move our meeting to our working call?

@03:50 - Iván Pulido

Yeah, let's do that tomorrow. So, in the afternoon.

@03:57 - David Dotson

Yeah. Okay. All right. I'll work with you on that timing. We'll just move it permanently because I think it conflicts for you in the current place.

@04:03 - Iván Pulido

Yeah, yeah.

@04:05 - David Dotson

Thank you.

@04:06 - John Chodera (he/him/his)

Okay. As we reviewed yesterday, we ran into a bit of a snag in that we have to upgrade to open Forcefield O11, and that's going to take some time to figure out since there was a API breaking change that we had not expected.

We have to rewrite the entire atom mapper. So I'll, this is a good time for me to ping your phone offline about how you folks coped with the upgrade in using the Percy's atom mapper.

So I'll just write you a little comment.

@04:40 - Jeffrey Wagner

Oh, is this the thing about, sir, I had seen this in my email inbox, but I haven't checked my I haven't read it yet.

10 seconds. So fragments not being molecules.

@04:48 - John Chodera (he/him/his)

Yes, so the original issue I raised was that a flag was incorrectly set by default in converting from RD kit to molecule that meant that.

inner conversion of fully or molecule subsets was sometimes not preserving all the information. And instead of correcting it, the decision was made to exclude any molecule fragment support, even though the toolkits all support it as does molecule and then push the validation of the molecule inside of or whenever you convert a molecule rather than when you need it at the time of actually applying parameters to the molecule.

So I tried to be descriptive about like maybe we don't have to do it there because it doesn't make any sense but it's up to you folks about how you want to do this but this is a I've labeled it as a regression, I've created a new label for that because it was a functionality that used to be there that now for reasons that I don't understand at least is no longer present.

So our strategy is to just rewrite the atom mapper so it doesn't use open force field to molecule anymore.

But that's not an ideal solution for us.

@06:04 - Jeffrey Wagner

Yeah, I do consider it kind of squarely out of scope to support molecule fragments. They come with a lot of complexity and basically prevent us from doing a lot of sanity checking.

I'll go ahead and I'll read this over when I have some time and give you a response here. But yeah, for me, it seems like supporting, not surprising behavior on molecule fragments and molecules at the same time will require a lot more work than I think we can afford to do.

@06:38 - John Chodera (he/him/his)

It does work. It's just like instead of change the flag so that it works by default, the solution was very surprising.

But it was working fine for the previous versions of the toolkit. So that I think you had already supported it.

It's just that maybe it's an assumption that you don't want to have to carry forward for all of the other behaviors of the toolkit.

But you can... create a partial molecule at the moment. That's totally fine.

@07:04 - Jeffrey Wagner

I wouldn't say it was working fine. I would say that you were getting undefined behavior that didn't happen to be surprising at the time.

@07:15 - David Dotson

John, wouldn't make sense as an alternative. So I know that within Percys, it's the idea as it's been up until now, use OpenFF toolkit molecule as like the canonical representation and then switch over to OpenEye whenever you need OpenEye tools.

You're getting a jump back, right? Does it make sense to use our dkit perhaps as opposed to OpenFF as like the canonical representation?

Cause our dkit's happy to handle fragments, right?

@07:41 - John Chodera (he/him/his)

I mean, it could be, but it would take us another three weeks for me to learn how to do that.

So, yes, possibly if someone else did it. And so this is probably a question for the OpenFE folks. Like are you rewriting the atom mapper to use already kitten turdily?

@07:58 - Irfan Alibay

If so, could we just use that? No? So we currently are not re-writing the process as a model. If it's something that requires our attention, I'm sure we could, I could raise it to the rest of the focus and see whether or not we can do it, at least if it is just a case of switching over until it depends on how it begins.

But it's something we could discuss.

@08:22 - John Chodera (he/him/his)

So what I'm able to offer is that I could rip out molecule and use the open eye molecule internally, but that doesn't get us closer to the idea of being open eye independent eventually.

But that's what I can offer in three days. What I can offer in three to five weeks is I could probably learn enough already to do it or we could have somebody else do it.

@08:47 - Irfan Alibay

Well, if you've got an issue that we could link to, I could probably raise it one of our meetings late in the week and then we can see if we can spend that time towards it.

@08:58 - John Chodera (he/him/his)

Sure thing, I'll document that in more detail. detail in that issue and then bounce it to your phone.

@09:03 - Irfan Alibay

Cheers.

@09:10 - David Dotson

Is this the issue you want you need from something on percys?

@09:13 - John Chodera (he/him/his)

Yeah, I need to add more detail about specifically what we traced the failure to be and what the section needs to be rewritten then.

Okay.

@09:31 - David Dotson

Any other questions or comments on this topic? John, do you feel like you have what you need or fun?

Do you feel like you have what you need?

@09:37 - John Chodera (he/him/his)

My plan is just to rip out open force field molecule and rewrite everything based upon open eye and if somebody wants a solution that's different from that, we can talk about it offline.

@09:51 - Iván Pulido

I think I mean, I think this was later in the agenda, but we're talking about it now. So. I just wanted to reiterate on the fact that we can write the protocol because we will be receiving the mapping so we don't need to generate the mappings.

So we don't need this rewrite of the atom mapper for the non-equilibrium recycling protocol strictly.

@10:20 - John Chodera (he/him/his)

If we can get atom mappings from elsewhere from the stream.

@10:23 - Iván Pulido

Right.

@10:29 - David Dotson

Well, I don't know. I pointed you to Richard for basically having a look at his replicate exchange protocol to see how he's using the atom mappings on their end.

And he had a chance to do that.

@10:48 - Iván Pulido

Or no, I was hoping to consider that meeting today in this space, but I'll reach out to him in separately for a meeting.

Yeah, because he's not here to give me.

@11:01 - David Dotson

much here. But yeah, I think you'll have to just chase them down separately. Yep. Okay, thank you. Otherwise, quick update from me on where I'm at.

Just building out the test suite for follow up. We as I go, test coverage is now at, I'm just going to find it 65%.

We have integration tests for these, the, the state store that's Neo4j. We also have integration tests for the compute API that's based on fast API.

I've referred a lot to Simon and Josh's work on this book fit for understanding enough of how fast past APIs use the world to make this work.

So we have integration tests working for that. We also have, there's an associated client with this that is going, is used by all our compute service variants that I also have integration.

test working for. And so I'm just building out the end to end functionality from along these lines right here, so that we can get up and running with protocols.

So that's where my focus is. I'm probably going to stay, keep my slack notifications off the remainder of the week.

So if you need to reach me, email may be the best way. I check occasionally, but just FYI, I'm trying to stay heads down because I want to aim for this end-to-end deployment of at least the synchronous version of the compute service.

And all these components to MSKCC to buy the next, buy next week. Any questions for me?

@12:48 - John Chodera (he/him/his)

Just a heads up for you. There was a recent work server update that you may or may not have been copied on.

So check in with Sucrate if not. Because it's not a good one to mention.

@13:00 - David Dotson

That doesn't impact the work server API.

@13:02 - John Chodera (he/him/his)

I'm also in contact with the Joseph.

@13:06 - David Dotson

So yeah, we're email back and forth. I asked him if the API docs are current and current. So yeah, he apparently hasn't had any feedback on the use of the work server, adaptive sampling API, but we'll now put it to you.

So he's excited to get some feedback on that. Any questions for me?

@13:30 - David Swenson

Do you foresee needing any of my time? I'm still available to you, but if we're gonna do that, which probably we should probably start soon.

So there's time to actually make use of me.

@13:40 - David Dotson

Understood. I haven't needed a reach for you yet. So, sorry. Oh, don't worry. Yeah, the bottleneck is just, it's kind of a, I, you know, building up the fast API things, building out all these layers.

And so I haven't hit a problem where I think. you be able to jump right in and provide some media help.

@14:05 - David Swenson

Yeah, but just if you do.

@14:08 - David Dotson

If you'd like to, like, I guess, because I've been operating in the assumption that I don't want to burden you too much with having to learn this.

All of Falc and these internals to really help me. Would you like to get familiar with Falc and these internals?

Is this, would this be of interest to you as in. If you go, if OpenFeat over time adopts some of these pieces as in like.

I know you're working the HPC solution serverless. The long long the Damonless solution. But if you guys end up needing to reach for a Damon solution that uses goofy objects, having familiarity with this might help.

@14:45 - David Swenson

Yeah, that is true. My priority here is helping you hit the deadline though. If I don't want to do anything that's going to slow you down and if me being involved can make you go faster.

I'm there for it. If I'm going to slow you down, then I'm not.

@14:59 - David Dotson

Yeah, because I'm. Not even operating with PRs at the moment. I'm just darting around and building all the pieces I need as I go.

And so it's really hard to coordinate with another person at the moment. And yeah. If you're interested in having, if you'd be willing to have a look at the code base, like maybe just walk through it yourself just to see kind of how it's structured.

That may be useful probably by the time I do reach out to you on something, you'll at least have at least have your geography in place.

Yeah. I'm going to. Yeah. That would be helpful.

@15:36 - David Swenson

Yeah. I'll do that at least.

@15:38 - David Dotson

Yeah. OK. That may give you some ideas as well for extra system things like that.

@15:47 - Jeffrey Wagner

Yeah. And I know David, you said that you're getting things from Bill Swope and stuff like that. Please redirect those to me in a public channel.

I think the request for further support for. that we're no longer supporting have to be answered with no, but I'm happy to say no.

So feel free to just put those in a public chat and attack me.

@16:07 - David Dotson

Okay, I can do that. Yeah, I did redirect that one to Pavan after about a day or so yesterday or last week.

So, but I'm just again, I'm going to shut off my notifications on Slack because I just want to stay heads down into space because I have a lot of forward progress.

If you need to reach me, email is the best way.

@16:29 - Jeffrey Wagner

Okay, perfect.

@16:34 - David Dotson

Any additional questions for me? Okay, other major item is pretty much in benchmark. You're fond. Do you want to take it away on this?

@16:51 - Irfan Alibay

I saw you open some PRs. Yeah, so just a slight update from last week. I think we've got a free

things that we've got left to do. So the first was number 8T1, oh, it's sorry, I'm gonna get a full request number up.

So it was the PR82, which adds fixes all the other logon confirmations and add some from the ligands and fix the same cell one.

That's nearly done. All that's left there is I need to add in the remaining ligands from from bin. So we had some, actually had some missing ligands from some of the other ligands that yammer files but that's all being fixed.

So I just need to mine all the performance information. So that hopefully should be done tonight. Then I opened 83, which has the minimum spanning graphs, using the process of a macro, which thankfully I was using.

for an FFTK 10, 0 to 10.6. So that's that I didn't encounter the previously mentioned issue. I'm still waiting on one of the systems because it will take a little bit of time to calculate.

But I would appreciate Yvonne if you could have a look at that at some point and see if the format works for you.

That'd be great.

@18:22 - David Dotson

Yep, is that directed at me?

@18:24 - Irfan Alibay

So that was a Yvonne.

@18:26 - David Dotson

Oh, thank you.

@18:27 - Irfan Alibay

I mean, if you also titled some, but I know you're quite busy the moment.

@18:32 - David Dotson

Yeah, I have my head in sort of in orchestration lands. So I prefer if somebody got their head in.

@18:38 - Irfan Alibay

I'm assuming if this works for you, Varn, then that should work. I mean, the main thing is if it's ingestible by Percy's, then that works for us.

@18:47 - Iván Pulido

Right.

@18:48 - Irfan Alibay

That's.

@18:49 - Iván Pulido

Oh, sorry. That's what I would test. Yes. If they run on Percy's like a one iteration and.

@18:55 - Irfan Alibay

Yeah. I mean, if you can just even load the mappings and like if the mapping is worked for you.

That's always a good start. So the mapping say are, so we have each edge, we have each ligand, and then we have the dictionary broken down as a list of sets so that you've got the sort of mapping between like an A and ligand B.

If that makes sense. So hopefully that should work. And then after that, the remaining issues are sort of dependent on these two being merged.

So when that's good, then I'll just then fix the BDB files and we should be good to go.

@19:36 - David Dotson

Were you intending to, I think, was this for provenance? These logs that you committed?

@19:41 - Irfan Alibay

Oh, sorry. No, that needs removing. That's from the one system that's still going. Oh, got you. OK. Yeah, sorry.

I'll squash my everything so it's not an issue. Well, thank you.

@20:00 - David Dotson

And so this is the, these give our mapping. These are, yes, this is the network plus mappings.

@20:08 - John Chodera (he/him/his)

So what format is the atom mapping? What format is in the atom, is the atom mapping in here?

@20:13 - Irfan Alibay

Is this a string?

@20:16 - John Chodera (he/him/his)

So what is, is that a list or a list of tuples?

@20:22 - Irfan Alibay

It's a list of tuples, sorry, yes. I can start some other way if you want.

@20:30 - Iván Pulido

But is this a format that Jamil supports? Like if I load the Jamil, would it be recognized as such?

@20:41 - Irfan Alibay

Because this doesn't look like a list of integer tuples. That's the list string. As far as the analyst concerns.

@20:48 - John Chodera (he/him/his)

I'm sorry. I'm pretty sure it's going to be a string as far as the analyst concerns.

@20:52 - Irfan Alibay

Yeah, okay, then that's not ideal. Okay, let me fix that. That's probably just my code going wrong.

@20:58 - David Dotson

So I don't think Yamil has a concept of tuples. It was Jason definitely doesn't only lists.

@21:03 - John Chodera (he/him/his)

But if you added if you added brackets and commas outside of these things, then it would be a list of Yeah, you might need to convert the the parens to brackets as well.

@21:16 - Irfan Alibay

Yeah, I'll fix that. Thanks.

@21:19 - David Dotson

Thank you. Thank you for putting this together, Airfawn. So like I said, last week, I'd like to doesn't have to be immediate, but I'd like to switch over to using protein ligand benchmark as the source for tests for balcony.

In particular, some pre-baked networks with pre-baked atom mappings as well. So, so this is putting us on that path.

So I appreciate that. Any see your currently working progress, you said you wanted at least a bond to have a look.

@21:55 - Irfan Alibay

Yes, I mean, you know, if everyone's good with it. with it then I'll take that as a proof and we can just measure.

@22:03 - Iván Pulido

Yep.

@22:04 - Irfan Alibay

I'll do it.

@22:05 - David Dotson

Yeah.

@22:06 - Irfan Alibay

Yeah. This will depend on 82 I think because I'll need to, or at least we'll need to recreate a network for Frombin just in the Aligin style.

@22:23 - David Dotson

We are we pretty close to being able to merge between you two then?

@22:27 - Irfan Alibay

Yeah. But I should be able to do Aligin today.

@22:29 - David Dotson

Do you need a review for me? Where would you like reviews from someone else?

@22:34 - Irfan Alibay

I don't necessarily think I, I guess if someone has a time but it should ideally be straightforward. I don't think there was anything too blunt on that.

@22:48 - David Dotson

Okay. I'm just asking who would you like to review this?

@22:58 - Irfan Alibay

I don't necessarily. I don't necessarily. So I don't necessarily think we need an extra review, but if someone wants to do it, then that's great.

Otherwise I'm happy just reviewing and merge myself.

@23:11 - Iván Pulido

I can review it. I mean, I will try.

@23:14 - John Chodera (he/him/his)

Yeah, I can take a quick look too if you just paste the URL.

@23:33 - David Dotson

Hi, I'm Agnes John.

@23:33 - John Chodera (he/him/his)

Thank you. I think that took.

@23:42 - David Dotson

I'm actually removing from the reviewers of our earphones since you're doing a lot of work there. I don't think I'll have time to take a look here in the in this week.

Is it vital that.

@24:03 - Irfan Alibay

I do. I mean, it depends more on what you want to do you want to be a software blocker on or as a no block it was a new if you want to have a sort of decide where we're going to imagine it's right.

Ideally, this dissolves blocks all the other issues.

@24:21 - David Dotson

So we have to. Yeah, I'd rather not be the bottleneck here.

@24:24 - Irfan Alibay

Okay.

@24:24 - David Dotson

I'm happy to defer to John and and if on this. Okay, thank you. Yeah, so this one's blocking pretty much everything else.

Are there any other PRs that we're able to move forward with or these? This is mine, the restructuring of the repo so that I can continue to take care of that.

Otherwise, number 76 is that one.

@25:01 - Irfan Alibay

I think that's a proceeded by two.

@25:04 - David Dotson

Okay. Can I just put that down and close it.

@25:07 - Irfan Alibay

I, yes, I should. Thank you for that.

@25:21 - David Dotson

Any other updates you want to give on any other. I think that's a good question. Hi, priority.

@25:34 - Irfan Alibay

Issues for zero, not three. I think that's that land. Again, everything else will. So once 82 and 84 managed.

Anything else should then fall in place. Okay.

@25:52 - David Dotson

Thank you. I had one more. I had one more. Question number 68 is being addressed by.

@26:04 - Irfan Alibay

Yeah, so once 82 is much I can do 68 that will just. I need all the BDB files fixed before all the.

Present before I fix that.

@26:15 - David Dotson

Okay, cool. All right, thank you. Any other questions for your phone.

@26:22 - Iván Pulido

So. Maybe it's not exactly ready to the release, but we were also talking about a couple of weeks ago.

Like historical results and we were opening like a repo for that. I don't know if that's a priority, but yeah, just wanted to ask about it.

@26:43 - Irfan Alibay

Yeah, so then it's currently working on taking on the extremely. He just has a working on a set of historical data for the protein link in benchmark.

So that's ongoing somewhere we've needed private for now because we were taking stuff. I filed David Hansberg wrong within one.

I saw Mick it seem like we were stealing his revert repository data too much. But we'll make that public as soon as we can.

Then the other thing we discussed was a validation set, which Ben and I need to sit down on top of that.

I mean, if you want to think we we already discussed all the action points for that, right? I just need to raise all the issues and get those titles.

@27:26 - John Chodera (he/him/his)

The validation set is just a small subset of things to make sure that this hasn't gone off the rails.

@27:35 - Irfan Alibay

So the validation set as we want to be, it will be more salvation-free and using smaller systems where the likelihood is that the starting corporation isn't going to change much.

Okay.

@27:49 - John Chodera (he/him/his)

So it's in the like in the live comm spirit of the validation set, just to make sure the free energy calculation tools are performing correctly.

@27:56 - Irfan Alibay

Exactly. So then we can do that release. It should be fast enough and. We don't know. We're not veering off too much.

@28:03 - John Chodera (he/him/his)

You might also ping the siremol folks too on something like that because they presumably also have some accumulated tests.

@28:11 - Irfan Alibay

It will be thanks.

@28:21 - David Dotson

Great. Are fun. Is there anything else you could use from this group?

@28:26 - Irfan Alibay

I'm trying to say anything.

@28:27 - David Dotson

No. OK, thank you. OK, we'll move on then. If you want to give us an update on non-equilibrium cycling.

@28:40 - Iván Pulido

Yeah, we basically did that already, but I just push the latest changes to the PR. This is still a work in progress.

So we have to I have to talk with Richard about the mapping and some concerns we have about the mapping because

As John mentioned yesterday, we might need information about what's not being mapped, like what's being transformed and also what's not being transformed.

Maybe, so we have to dig into this and I want to ask Richard House, he's dealing with that. Other than that, yeah, basically that would be what I need.

That's what blocking me for continuing and finishing the non-equilibrium recycling protocol.

@29:34 - Irfan Alibay

May I ask what you mean by not being mapped to this with regards to the ligand components or the entire system?

@29:46 - Iván Pulido

So I have to check in detail, but we do need information on if you have the transformation of the ligands and some of them change, for example, from...

constrained from across strained bond to an unconstrained bond. You need to know where they are in the whole system.

So at least you need to know where the ligands start in the whole system. So you need some kind of reference.

And I don't know if we get those.

@30:15 - John Chodera (he/him/his)

Let me just explain on this a little bit more. So we, you know, in a free energy calculation, much of the system is considered the environment and doesn't change, but you still need to know what in the old and new sets of coordinates, how is the, how are the atoms mapped, right, for those, even if they're not alchemically modified.

For the rest of the system, or for the alchemical part of the system, like the small molecule that you're modifying or the residue that you're changing, you need to know which atoms are being mapped across that may change their identities, and which ones outside of those are going to be created and destroyed, right?

So we just have to make sure we can get all of that information from the objects that you folks have.

Now it's maybe possible to figure it out, but we were worried that if, for example, if you only have the ligand component mapped in the old and new systems, we don't know.

know how, like, if the ligand is in the middle of that set of atoms, we don't know how to map all the rest of the atoms in the system that should be part of the environment.

Or do you expect to have every particle in the system that is constant also mapped, in which case we also have to figure out which part is the alchemically modified part?

Does that make any sense?

@31:19 - Irfan Alibay

Yes. I've just sent you a link to where we do that. You know, but if you currently, it's not the most well-tested piece to code.

Actually, it's one I think we're currently investigating for a bug. But we can probably align on how to do that.

It's kind of derived of how to pull your proposal used to do it. So it's kind of a data-stoology proposal.

If it's something else, but we can sync up on that if you want some point.

@31:47 - Iván Pulido

Yeah, thanks. I'll look at it and ask questions. Yeah.

@32:00 - David Dotson

Yeah. The RBFE protocol in OpenFE lives, I'm gonna drop the link here in the chat. I would use this as a reference.

@32:14 - Iván Pulido

Yeah, I've been using it, but it's so uncoded.

@32:22 - John Chodera (he/him/his)

If there are utility methods that could go into the object model themselves to retrieve bits of things that we need that might also be another thing to consider.

Because if everybody has to implement this piece of code, then it seems silly not to just have that be a common part of the library.

@32:40 - David Dotson

I mean, it looks like just from when looking at it here, then taking the mapping that's coming in, pulling out the ligand specific component and then slotting that into various places that.

Let's see. Yeah, Ivan, you just have to kind of follow. follow like, or does Richard use the mapping here and how does it, how does he hook it into various things because he's fundamentally using open to them at some level to.

So that, that may give you some insight as to how to go about using these mappings.

@33:14 - Irfan Alibay

And we can discuss this. It's a little bit complicated, but yeah, I think I agree with John, if we can at least have some utility somewhere that we can move somewhere else that be great.

@33:35 - Iván Pulido

Yeah, that's, that sounds good.

@33:41 - David Dotson

So, I have questions for Ivan. Ivan, do you feel like you have what you need. Yep. Thank you. We'll move on to a protocol setting so I know this is merge but Mike just wanted to give you an opportunity to.

Celebrate. run some victory laps and tell us what the next steps are.

@34:12 - Mike Henry (he/him)

So it's merged and also got the as part of that PR got the docs building for it as well.

So now it's just going to be a matter of kind of getting a iteration on it as we work it into protocol and figure out which pieces need to go back into this and which settings will pull out and the protocol specific.

So, the docs don't look the best because the theme we're using limits the horizontal space so it kind of wraps around in a funny way, but this is generated from the JSON schema of the object models.

So it doesn't this this particular documentation is generated from the JSON schema. So it supports like what the JSON schema has for its typing, but.

It's linked in there so that way you can see like the Python bits as well.

@35:10 - Jeffrey Wagner

Well congratulations Mike.

@35:12 - Mike Henry (he/him)

Yeah thank you.

@35:15 - David Dotson

Excellent. This is awesome. Are you getting um so I think the next steps I know we want to iterate on this but do you have any concrete next steps for where you're like working with either Richard or Yvonne on getting them to use the settings objects and their protocols?

@35:35 - Mike Henry (he/him)

Yeah I'm gonna see about Richard. I think Yvonne should just keep doing what he's doing because I think he's pretty close to the finish line now so rather than like also in the spirit iteration I rather get that Percy's protocol stuff done and then we can figure out maybe the best way I can suit this.

@35:58 - David Dotson

Okay because once we are... actually trying to use this protocol in a production setting, we're going to have to inject settings into it.

And so I think we're going to need it sooner than later, unless we...

@36:11 - Mike Henry (he/him)

Yeah, I mean, it's one of those things that if it's not to go to a lift to start using it now, I'm sure I'll expose some deficiencies or ways it will break.

@36:19 - David Dotson

That's what I'd like to push for, because if possible, I'd rather start sooner than later.

@36:25 - Iván Pulido

So we have... So we are made in tomorrow, right? Or we're planning on meeting tomorrow. Maybe you can help Mike and devote a few minutes of the meeting for this.

@36:38 - Mike Henry (he/him)

Yeah, you should be a invite.

@36:41 - Iván Pulido

Yeah, I'll send the invite.

@36:45 - David Dotson

OK. Yeah, can you give on... And you go ahead and just set up a meeting with us. Yep. And I'll cancel our current series that was running today.

@37:00 - Iván Pulido

I think I deleted that one. Well, I deleted it locally, but I'll actually delete it. Oh, I see. Okay.

Got it.

@37:08 - David Dotson

Yeah, that's done. Thank you. Okay, cool. Any other questions from Mike. Thank you.

@37:24 - Mike Henry (he/him)

Is there anything you'd like from this group? No, I think that's it. I guess one more little thing of like next steps is I had to make a PR against Matt's model repo for this to work.

So I, then one of the other concrete steps now to work with him to get that upstreamed properly and done correctly.

Not to mostly tested. It's tested as in it. I needed it to work, but it needs some independent tests for the open if I have models repo.

@37:52 - David Dotson

Okay, so yeah, is open if five models. Is it do we have a kind of forge package for it?

@38:00 - Mike Henry (he/him)

No, not yet. No. Okay. Not to my knowledge. Right now I'm pip installing off of my fork anyway. But we probably I don't know it's up to Matt if he wants to get a kind of forged package going forward or not.

But I think that since it's kind of in its like very much developer mode, it's probably better to keep it off kind of forge because I hate for SEO to make it start getting a high hit.

And then people think it's part of the open F like the name of the repos already open F models like to so the name sees packaging and stuff work.

So the repo name but like the package name. But I'd like to maybe keep it under the radar for now.

I feel like kind of forge kind of puts a stamp on it. It shouldn't. It's just like pipi but it kind of does.

@38:50 - David Dotson

Okay. I mean, the alternative is that if this is now because this is now a hard dependency of goofy, right?

Yeah. So I think we really do need to move that direction anyway. I guess, yeah, I understand your point then maybe it's early days for it yet, but.

@39:13 - Mike Henry (he/him)

Um, I could, we could vendor it also inside goofy.

@39:19 - David Swenson

Maybe it's changing quickly. I don't know if that's a good idea. Yeah, if the argument, not putting it on kind of forages and still in rock and flux.

@39:30 - Jeffrey Wagner

What we could do is so the one thing we found with conduct forages, sometimes you get your recipe approved very quickly and sometimes it takes a really long time.

Um, but once the recipe is approved, you don't necessarily have to put a package up or you don't have to put a package up on the main label.

So we could, I could have Matt open up a feedstock for this now and try to get it approved.

And then, um, that doesn't mean we have to make a release, but it means a minute we want to we can and we don't have to maybe wait three weeks.

@40:00 - Mike Henry (he/him)

No, that's a good point. That's a good point because yeah, we've ran into that to the lead time being very variable depending on how the conduct forage teams.

Thisiness is, and then if it's on a, um, like a beta label or dev label or some other main label will never get accidentally pull in or something like that.

So yeah, I think that's the idea.

@40:19 - Jeffrey Wagner

Okay, cool. I'll tell Matt to start a few stock for this. Sorry, the original point here is, is there a pull request that you want to highlight for review.

@40:31 - Mike Henry (he/him)

Not yet. I need to add tests to it. So it's not all bug and basically I got it on there and.

So that way it's just there, but like it's lacking tests in this week but it's obviously tested with how I'm consuming it, but it could use some specific tests inside this and also performance too because I, in order to get the models to work to be recursive, you have to make them recursive.

Okay. Imagine that. So there could be some performance issues as well. So I just kind of want to play around with it a little bit too.

@41:09 - David Dotson

Cool. Can we also, similar to the Conda forage question, can we also migrate this to the open force field or.

@41:21 - Mike Henry (he/him)

That's above my pay, Greg.

@41:23 - Jeffrey Wagner

Oh, I see. Yeah, it's in Matt's name space. Yeah, I think that's a good idea.

@41:32 - David Dotson

So you're concerned about users trying to use it a lot. You could put a label on it just to say like this is also software.

Yeah, I mean, you could also consider renaming it so it doesn't, you know, pull.

@41:49 - Levi Naden

We were a few word about the SEOs and whatnot. You just give it a different name.

@41:52 - David Dotson

I mean, I mean, I'm not sure if that's true. You're a bunch of. like we have a bunch of condens in files with pip install paths, right?

From GitHub and then you have to go in and fix all of these later. I'd rather just let's get it in the place we want it so we can continue to develop it.

But especially since it's now a hard dependency of goofy.

@42:17 - Jeffrey Wagner

Yeah. Well, an open, big open FF name, GitHub organizations are full of misleadingly named and abandoned packages. So what's one more?

@42:25 - Mike Henry (he/him)

Yeah, one more to the pile. And I know that to the part of this package ripped out or like duplicates some stuff in interchange.

So depending on that's ready for that, we can kind of swap that piece of that. So that way we only have like one copath for some of this stuff.

@42:44 - David Dotson

Wait, so does that mean that some of this stuff is just going to migrate into interchange later or?

@42:50 - Mike Henry (he/him)

No, no, no, no, no, this there's stuff in interchange. OK. That this is ripped out of interchange is the right way open enough of models.

His his. then is stuff is pulled out of interchange.

@43:04 - David Dotson

Okay. And the idea is that this is a lighter set of dependencies than interchange.

@43:09 - Mike Henry (he/him)

Yeah, so the provenance of this is Matt wanted to use fancier units in interchange and also wanted to use some gigantic data models in interchange and only an interchange.

Like that was the only use of this. So it didn't really make sense to make a little micro package in a vacuum.

And I came along and said, hey, I kind of want to reuse some of these models in interchange. So that's then what made this turn into a new package.

@43:36 - David Dotson

Okay. I see it. I think I understand. Yeah, I'm just asking for clarification.

@43:49 - Mike Henry (he/him)

Yeah, I think Lily's using that too. So I think it's starting to get more adoption. So eventually, like if it's not because of us pushing for it.

get more mainline. I'm sure other OpenFF toolkit people will want it. Okay cool.

@44:08 - David Dotson

So yeah can we pursue just for action items? Move into the OpenFF, Open Force fields work and then start to pursue after that kind of forge packaging just because it might take a while.

@44:21 - John Chodera (he/him/his)

So openFF or OpenFE?

@44:25 - Jeffrey Wagner

OpenFF, OpenFF, okay. Yeah if you scroll down I've already made an action item for myself.

@44:32 - David Dotson

Oh sweet. Thank you. Thank you Jeff. Perfect. Any other questions for Mike? Okay thank you. That's it for our agenda.

Are there any other topics folks want to add? Okay if not... That will give everyone 15 minutes back. Appreciate it.

If there's anything that comes up again, if you need to hit reach me, you can still reach out to me on Slack, but I just checked very, very solemn this week.

Otherwise, email may be the best way to get ahold of me. All right. OK. Thank you.

@45:23 - Irfan Alibay

Take care. Thank you.

@45:25 - Jeffrey Wagner

Thanks. Thank you very much.