DevoWorm Lab Meetings, 10 Jan 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: DevoWorm (2022, Meeting 1): Looking forward, lab preprints, 3-D microscopy, and soft biomaterials.

Description

Looking forward into 2022. Lab preprint review on Game Theory in Development and the DevoLearn platform. Google Summer of Code 2022 projects. Papers on 3-D microscopy techniques (PIV and OCM) and the biophysics of soft material thermodynamics. Attendees: Susan Crawford-Young, Richard Gordon, Karan Lohaan, and Bradly Alicea.

A

B

C

Good to see you here.

C

So why don't I get the meeting started? um Other people come in we'll catch them up on what we've been talking about. So how have you been doing uh over the last few.

B

Weeks, uh yeah things have been okay, just finished with my.

A

B

Exams after that,.

B

C

B

C

Well, thanks for showing up uh yeah yeah, welcome to the meeting and uh we're gonna talk about something a couple things this week. um This is the first week of 2022 or first meeting with 2022 so uh welcome to the new year. um We'll talk.

B

About some new things.

C

Yeah, so let's see um first thing I want to talk about is I can launch right into g or google summer of code news or what we want to do this year.

B

C

So, uh as you as people know, here, uh we are affiliated with incf. That's our uh managing.

B

D

C

And uh that you know that's the they're, the ones who interface with uh the google summer code program, so they manage everything and and sponsor us. So this year it's gonna be the program is gonna, be quite a bit different. They're gonna have two different lengths of program, uh like I think, a 10-week in a 22-week program.

C

E

C

About the details of what they, I just got an email this morning about what they want exactly in the uh calls for involvement like when your your group puts forward a proposal to the sponsoring org and they advertise it and get people um so that that's that, and then uh we also thought about uh having our own organization hi. Dick.

C

And so our own organization is uh the idea there is that I've got some other people who are interested in participating, but they don't have an uh sponsoring organization. So this year, male 1 and incf suggested. You know that if people want more slots and if they want to do so, they can, they can fill out their own mentoring organization, app application and then incf would sponsor it, and it would be a better chance.

C

It was getting accepted that way, the benefit there would be we'd have more slots and we could do more things, but it's uh it's up in the air and we probably will do both and then we'll see which one works out. So I have a let's see so I'm going to share my screen.

C

So our let me see where I put this stuff, I think I put it in here yeah, so this.

D

These are the proposals.

C

That I've come up with where we've come up with and I put together um and so this is. We have two and I think, that's probably enough. In the last year we had three, but uh this one is. uh I don't need that one for this document.

C

You want to connect, so this is the first one is uh gnns as developmental networks. So this is a graph neural networks and.

E

C

Is that we have all these networks in in biology and developmental biology, neural connectives, gene regulatory networks, interactive networks and anatomical networks, and then we can do basic, what we're doing with the work in deep learning and with diva learn, and we can construct these graph neural networks, libraries and fit them into that toolkit.

C

So we have our deep learning toolkit, which is a number of different models that we can use on analyzing data. But then we could also use these graph neural networks, and the benefit here is that we can create graph embeddings that resemble actual biological networks. So we can sim.

C

I don't know exactly how this would be pulled off, but that's one of the things I like to do is like to get people a chance to uh sort of propose their own solution and then once they do, that you have say like 10 applicants, you can find the best one and then the people who don't get selected would can um you know, follow up on that on their own or depending on how many slots we get. We can accept more people, but this is um so.

C

This is just kind of using these kind of graph embeddings uh to sort of characterize biological rules. uh Look at the differentiation of nodes in a network. So once we have these graph embeddings, we can do a lot of things with them potentially, but we need to get that those that work done first. So that's why I propose that we start here.

C

So the goal is to create a library of gnns that can simulate developmental processes by analyzing time series microscopy data, so we.

E

C

Data that we've been using for our deep learning work. We can just simply use those data sets and maybe find a few more, and that would be that, and uh so you know we do this, we kind of tell them what they can do before the program starts.

C

uh It's the same as every year and I'm not sure about the skills and requirements. I may have to update that, but I think pie torch and tensorflow. You know if you want to use one or the other, then that's that's a good basis for it.

C

So that's that's the first project and I'm not sure about the gui. I don't think we need to do that. So I'll. Do that um and then the second one is digital microsphere, and that is a project that is the one we proposed last year. This is with susan's data, so this is the one where you have her ball microscope that can take pictures from all these different angles and you can also use this flipping microscope technique.

C

So you have all these data that basically show an embryo, the surface of an embryo with different perspectives and then we'd build a spherical atlas of this, so it would be, like you know, you'd, be able to tile the surface of a sphere and explore those. um You know that surface and you know so this is basically I know we had a couple people uh produce artifacts on this last year. I don't know I'll, probably send it out, hopefully they'll reapply, um but this is enough.

C

You know this is again one of these things where you have to come up with your own solution. You, you know um just basically say I'm going to use this technique. uh This will work good with this project.

C

You know, there's a lot of uh kind of like figuring out how to do these tilings of space, so you take like flat images and you're tiling them onto a sphere, so there's some mathematical transformation that has to happen, and then you have to have a base sphere that you can tile these things onto that you can navigate. So there's a you know a bit of work that goes into that part. So it's a little bit different than the first one, because you're not just throwing an algorithm set of algorithms at it.

C

There's a lot of stuff that you need to kind of figure out how you would you know.

E

They're going to be different tools.

C

It's going to be more heterogeneous, I guess so, and then the skills, I think, are uh higher handling higher dimensional microscopy data building, an intuitive, a gui uh feature extraction, so yeah these are base the basic skills, and so these are all the ideas. Are there any ideas about? You know if we want to add another project or if we want to.

C

The account differentiation tree and phylogenetic trees as networks yeah. Actually we do. We can count those as networks and um that's something you know when I guess we can. In parallel with that project, we can kind of characterize different things that are networks and kind of put together a short list of those things that you know what how we might work with them in in terms of not only as networks but like in these kind of graph neural networks. So that's a good question.

C

So any thoughts about the projects or or I guess other thoughts.

B

uh What what could be the potential applications for the graph neural.

E

Network project like could it be protein synthesis and all those things or maybe.

B

Something more uh basic because I remember like alpha fold, was using uh something similar to.

B

Do the protein like.

C

uh Well, I guess you could you know there are a couple things that we could do. We could characterize networks just in general, so you know we could build the embeddings and then the idea would be you'd, be able to train different types of data and get analyses. So you know with the deep learning stuff it was just like taking uh microscopy images of an embryo and segmenting those images and then using the segmentation. Information to you know build these classifications.

E

D

C

A

C

Like numeric data that we can analyze data yeah, of course it could be other types of data as well. It could be like.

B

E

But the idea would be that.

C

You know, depending on the type of graph you're interested in extracting, you would look at the data and then say this is how it's going to be extracted and analyzed. These are the features that are important.

B

Oh, it's very complicated right now: okay, okay,.

C

Let me see another message in the chat: see my papers on google embryo yeah. I think.

E

C

Cited one in the list here to read so yeah this one here, so this is the uh the google embryo for building quantitative understanding of an embryo as it builds itself and yeah. I usually give people these two papers to read if they're interested in the project. So it's a good starting point.

C

Yeah send send along the copy, because I'm not sure that people can access always access it. So.

C

C

Just making some notes here on this so yeah, I think that's good. um Yes, I think you know the benef only benefit of having another project here would be that if we don't get any applicants for these we'd still have another one, but I think these two will probably this one will probably get a lot of applicants.

C

This one will get a few, I'm sure so we'll see how these go.

C

um This all starts like we have to have these projects into uh incf by the end of the month and then, if we apply for or we will be applying as an organization and that'll be another month after that, so by march 1st, we it should be the beginning of the application period and that's when people come and start asking about the projects and then applying- and you know we work with- I work with them, mainly to develop the proposal, and then you know we get some pretty good, um pretty good solutions to some of these things.

C

So so we'll see how that goes. I think it'll be a good opportunity, if nothing else.

C

So that's good um any other comments or questions before we move on.

C

All right, let's see, oh, we got one here, yeah scent. Thank you. So speaking of this uh diva learn uh this. This is the print link to a preprint here, and I think I showed this uh last couple meetings ago last year, and uh this is now I've got this up on the research gate is a stub. So this is the diva learn paper, so I've had it on github and I'm not really sure it's long enough to really publish it um or even make it into something like an archive preprint.

C

But it's oh, we got 11 reads on it. I just uh yesterday, when I looked, we had three. So that's good, um but this is the pre-print for evil learn. So we were working on this all last year and uh I had the I've turned it into. What I did was I created.

C

We actually was. I think it was myself and my yook, but it were. You know uh we created this uh document in markdown, which is the language that we use on github you you know make commits to github in this language. It puts together this document and then I converted this to a pdf using atom.io, which is a nice tool for conversion, converting things from, like you know, mark down to pdf or uh html to markdown, or you can work with latak in it as well.

C

So that's a nice document tool and I created this pdf, um so my my knock bradley and usual are all the authors and then the abstract is pretty short.

C

We have a summary of the software here and then there's a little gap here and view the file like this there's a little gap here, because the image is actually this image is actually it's animated in the markdown version. But it's it's static here, that's the only disadvantage, but and then you know, headings about accessibility, statement of need, technical details and then walking through like segmenting a c elegans embryo again, these figures are animated in the markdown file. So it's it's with the pdf.

C

It's like limited utility, but it you know, that's just something that is the uh disadvantage of a pdf file, and you know we have our data sources uh generating synthetic images using gans, hyper parameter. Optimization lineage prediction, which is something so this basically goes through all of the different features of diva learn, and I was surprised when I went through this to put this together.

C

How many things there were- and this is from uh mayak and minox projects, as well as some other documentation that we wrote up on it so and then the diva learned platform, which is this uh github organization that has the different. So we have diva warm ai. We have evil, learn and then.

E

C

Devo zoo, which is something we haven't talked about in a while, but this is the collection of data sets that we have for different species, and it's something that I'd like to get back to at some point. But it's uh you know. Maybe it needs to be upgraded a little bit more, but I think there's a role, there's a solid interest. I think in people talking about different model organisms.

C

So you know we talk about the the name of the group. Here is devo worm, but we could be talking about devo fly or devo human or devo mouse.

C

Those are all relevant to develop to understanding development at a broader scale. So this devozu is actually a nice resource and it needs to be developed more. But it's all under this umbrella- and the idea here is to kind of in in this. The thing nice thing about a preprint is that we can add in. If we decide the devozu needs to be highlighted a little bit more in here and highlight it a little bit more and describe what it is. In fact I think that'll be.

C

The next version is going through evo, zoo and even diva warm ai, which wasn't really mentioned here too much. But uh so that's that's that and I think that's a nice way to like kind of cap off some of the work that we've done on diva learn and um I don't know what the next steps are for. Diva learn. I guess it's just you know when, as we get people coming in, um they can, you know, maybe add on their own functionality or you know.

C

I think the big thing, though, would be if we could get a user base for this and people actually using it, and but that's that's something that we haven't really seen. We have.

E

Had a lot of downloads.

C

Of the software, but it's you know it's difficult to really build any more of a community.

A

C

It or build you know any sort of more work if we're just okay, people are just kind of using it and uh thinking about next steps.

C

So my knock and my hook, I haven't really been involved with it much lately uh by lately I mean the past two or three months, so they might come back and work on it, some more or we might get some new people who might want to do some more work on this. So that's very good. That's I mean this is something I want to keep sort of sustained and I want to keep bringing it up and coming back to it.

C

Let's see, we have a comment: suggestion write people who write papers on model organisms yeah. That would be a good idea.

C

Okay and then yeah, I think, that's a good. I know there have been some interesting uh reviews of model organisms of late. I think we've reviewed a couple in the group and then they're others that I've seen that I need to go back and grab hold of, because I think that's an interesting kind of topic to build on, um and so you know, work on model organisms is two-fold.

C

One is like: if I'm working with one model organism, you know, I always ask the question: how does my work generalize across the tree of life or other model organisms, at least so people working on you know, drosophila or c elegans they're typically focused on their own world of that organism, and you know people will use celiac c elegans to look at things in humans, but it's much better if we could get like a broader characterization of some of these things, and so you know some.

C

Sometimes, when you look at model organisms, processes are very different across different animal species or even animals and plants or, however, you look at the tree of life, but sometimes they're, very similar processes, and so you know where you can find those similarities.

C

Something like vivo zoo is nice, because you have this, you know cross pollination of the different model, organisms, kind of put them in a broader context.

C

So look I'll think about the people who you know, maybe I'll contact some people and see um what they're thinking.

C

Okay. The next thing I want to cover is this game theory in of developmental processes. So this is a paper that I I just put this preprint out. This is something that I think it presented last year a couple months ago in the group this is uh this was something that was submitted to the dynamics days conference, which just happened this last weekend, so it was a pretty decent conference. I was hosted by georgia, tech and they were hit.

C

You know virtual because they, you know, couldn't meet in person, and um so they had, they had the whole thing set up in talks posters. We got in as a poster, so this is uh like a two minute thing and that video of that presentation is on youtube on the youtube channel. So if you want to see that poster it was a very short two minute uh summary of the poster. So then, and then this uh preprint is meant to be a supplement to that poster.

C

So you know it's like you want to find out more read the preprints. So that's why this is here. Actually I just wanted to write up a preprint for it to keep moving the idea moving forward, but um there's a longer 30-minute talk on also on the youtube channel for this uh basic idea that I presented at another dynamics days in europe last year so and it was also virtual. So this is the preprint and it actually goes into a lot more depth in the different presentations.

C

um So let me view the file here walk through this a bit.

C

So this is a paper that takes developmental processes and applies game theory models to it. So the idea is that you can use this class of model to model developmental processes in different ways, and you know there are different types of games uh and they're different, a lot of different types of developmental processes. So there's a lot of variation there that you, you know, but the general idea is, is what I just said, and that also requires looking at game theory a little bit differently.

C

So uh people have thought game. Theory started back in the 40s and 50s with john nash and von neumann and some other people, and they came up with these basic models of game theory. It was basically you know, games of strategy so like something like um like uh they call the tit for tat game, which is a game where you have one player, plays a strategy and another player tries to one-up that strategy and then the other player responds by one-upping that strategy, and so so on and so forth.

C

um They're they're strategic games like chess and poker and those are all you can model those as games. So you can basically use a computational model to model those things as games, and uh you know this is. This was something when you, if you take the chess example, this was something that was used in early ai to sort of uh serve as a model for intelligent behavior.

C

So an ai would uh play chess. It would be able to play like a game of chess or maybe like a game of uh tic-tac-toe or not. What's called in the paper here knots and crosses or something similar, and I would just implement basic strategy.

C

It was usually some combinatorial game, because computers are good at combinatorics and you know eventually, this became something that was able to beat humans, so in the 90s uh ibm's deep blue was able to beat a chess master chess grandmaster and so that so that wasn't a long time that was like 40 years but anyways. That was the idea.

C

So we can use these for developmental processes and they're different types of games, but before we get to that, we have to talk about different types of things that the agent or the player has to embody as a developmental agent versus like a strategic human player, or something like that, I mean the idea of uh rational strategic player is a little misleading because although it's been applied, a lot to humans, they've also used this to analyze, animal behavior, so you've had games that can model like birds and their visual behaviors and their different decisions that they make, but still even in birds.

C

You have like a cognitive agent, it's a cognitive agent and developmental developmental game theory. There is no there's, no cognition assumed because you could be dealing with an organism. You could be dealing with a cell. You could be doing with molecular interactions.

C

So there's this idea of the.

E

C

Agent, which is this type of agent that doesn't have this- uh it doesn't explicitly, have cognition and it doesn't behave in this rational manner. It behaves sort of in a natural manner, and so what that means is that you have these. It basically operates on the uh principle fitness.

C

So this is a lot like what they call evolutionary game theory where the agents are evolutionary, they have a fitness and they sort of maximize their fitness for each of them, and then that leads them to these different uh evolutionary trajectories that end.

E

Up you know either in some.

C

Equilibrium or some other result that is able to maximize fitness for each player in the game, so the other benefit about taking cues from evolutionary game theory is that you have this uh dynamical trajectory so in development. Of course, that's very important where you have this trajectory of what goes on in development, and there are actually two ways to view this. One is through an evolutionary game, three lens which is maximizing fitness, but there are also these iterative games which are forms of games that get played over and over.

C

So I think, there's one example in the paper of like.

C

In the prisoner's dilemma, where there's a form of the prisoner's dilemma, which is the iterative prisoner's dilemma, and so you have this single game that you can characterize and then you have the iterative game which can just be played over and over and the payoffs change based on um the you know, few past states, the past state informs the present state which informs the future state I'm trying to find that table.

C

um Maybe I don't even have it in here, but there's a table that kind of goes over the prisoner's dilemma, and it's payoffs. I don't have it in here. I don't think I put that one in anyways.

C

The prisoner's dilemma is a game where you have two prisoners and the idea is that they're going to you know you have them in different rooms and the idea is you want them to confess to a crime, but if they can coordinate their behavior they can both sort of not confess you you threaten them, maybe that the other person is talking right now or that you're going to punish them severely, and so the idea is that they it to reach a optimal equilibrium.

C

For the prisoners, they both cannot say anything that would incriminate each other, but oftentimes they'll, incriminate, one or the other, and so that's where there's a sub-optimality there. So that's the prisoner's dilemma without going into too much detail. But the idea, then, would be you know with the presenter's dilemma: the iterative version do both prisoners. Are they able to coordinate the information that they have over a long period of time or do they just do it once and then it breaks down? And so you know these are.

C

There are all sorts of games that I feature in here: they're dif, there's a zero order or zero player game, which is uh a set of games that involve uh like different point processes and the game of life is actually considered a zero player game.

C

Then you have single player games which are games against nature, and um this is where you have some natural process, a stochastic process, or it could be deterministic which isn't very fun but like the stochastic process, and then you have an agent that's playing against that stochastic process.

C

So the agent's trying to predict what the stochastic process will do and, of course, being a stochastic process, that's hard to predict, but the agent will develop some sort of strategy for that and then there's a payoff. If the agent is correct, they get a payoff of one if they're yeah.

C

Actually, if there's a, I think if they came out how this works, but there's a payoff that they get for some sort of correct choice and then for a neutral choice which would be like something: that's not correct, but something that doesn't hurt the agent it's a zero and then, if it's something wrong, it hurts the agent it's a negative one. So then you can add this payoff of over time. So this collective payoff is, you know, there's an average payoff, a cumulative payoff.

C

You can calculate it in different ways and then you can say I see where that process unfolds it the way it did because that payoff is, you know, there's a maybe there's some mechanism in the agent, that's able to predict um something in nature, and but it only is able to do it sporadically.

C

So you can kind of understand things.

D

Like that, and that's actually.

C

Quite valuable at the molecular scale where there's a lot of there are a lot of processes that involve like recognition, but you know like say, for example, cells when you look at like things like uh translation or uh binding sites. Things like that, it's a processor there's a lot of this going on and not all of it actually unfolds in the same way. Sometimes you know there's just a lot of stuff in the cell. That needs to be bound, and if enough of it is bound you get this activation.

C

So not every agent is going to have a fair payoff, and so that's that's the idea behind this, and then there are these uh two. You know the end player games, which are like two or greater number of players, and there are a number of examples from behavioral biology and other. You know other types of rational agent behaviors that we can use and apply to this developmental context.

C

So in this case there were two examples: the prisoner's dilemma, iterated prisoner's dilemma which I've described previously, then there's this knots and crosses, which are the first mover games, and this is something we published a couple papers on. We've published a paper on this in embryos which is in this paper, origin of the embryo. That was done a couple years ago, where I like one of my contributions to that was to put in one of these first player games.

C

So the idea is that you have these different cell lineages that uh they differentiate and then they divide and the idea is there in this confined space and the lineage that divides first is the first mover and then the one that divides. Second, is the second mover and the first mover in general constrains the second mover's behavior. So if the first mover divides, um they have more space to expand. The second mover then, can divide and expand and then so on and so forth. So this might explain timing in a lineage tree.

C

You could also apply this to the connectome, and so how does the connectome get assembled? And I actually use some data from c elegans, both cell differentiation data and synaptic data, to show how this connectome assembles taking this uh principle of first mover dynamics and deriving some rules or some scenarios where the data fit in. So the data get classified by these different rules and they fit these strategies.

C

So then there are other types of things like systems, equilibrium, complexity, so they're different types of combinatorial games, evolutionary more explicit evolutionary games and what we call spatial games, which are things that look at space as a set of potential strategies. So you know agents will obtain a specific spatial structure as a collective strategy to defend against invaders.

C

uh There are other things like that, and then there are the system, broader systems, level phenomena which I go through. So that's that paper and I I like the idea, particularly of uh natural agents, but some of the games. I mean this is like a pre-print, so we can expand out. We can come up with new games or kind of discuss more about some of these things.

C

I was thinking even of what we might call an evo divo game, which is where we apply uh evolutionary game theory more explicitly to the developmental context and look at how that uh maybe explains some things we see in the evolution of development, a couple of uh okay, so a suggestion right, people look at only 932 papers in web of science was that for, uh oh, that was for model organisms or yes, okay, thank you. Yeah could extract email addresses, so maybe yeah look.

E

C

Web of science and see what they have if we have some interesting things there and then differentiation wave is a simple purpose. Okay, did you want to expand on that.

C

Yeah I mean it's this part here, the part about, like you, know, purpose and natural agents. It's an interesting topic of stone papers yeah. So you have to look at that a little bit more. I can't remember exactly even the paper, the origin of the embryo that I was involved in. I can't remember if you get to a point where you can't remember, you remember kind of what every paper is about, but like kind of bernie.

E

C

You I'm sure dick knows that it's like.

C

Anyways, um so that's good um any comments or questions at this point.

C

Okay, so um next thing I wanted to move on to is a couple things for 2022.

C

We had a couple things that kind of ended. At the end of the year we had the submissions document and the major tasks which haven't been updated. You know as regularly as I would like, but that's okay, um so this is our like last year. This is our submissions list. So at the beginning of 2021 we started this list and uh we had 34 things in it from last year, and this is like things that were proposed or that we did for different venues.

C

Some things never got anywhere and some things got accepted and presented.

C

Some things were rejected, and so you know there are all these possibilities, but we wanted to assemble them in this document, so this was for 2021 and maybe in he had a comment on this once it was headed open and someone said, is that something you're going to go through, because this looks pretty um illegible? um Jim I mean it is. If, if you look at it like this, we usually zoom in on a couple lines, but um but anyways maybe doing this for 2022 and starting off with some of the new submissions.

C

So 2022- and this is 2021.

C

So let's go to 2021 here, okay, so we had a lot of things and a lot. Some of these things will carry over to 2022, um and so that's something that so the neuro ips was something that was. I think people didn't. I don't think anyone here sent anything to them, but I think that's something that a couple people, especially my other group, uh participated in, and so there and same with neuromatch we had a narrow match.

C

They had some participation in neuromatch as well, so that's good um and then this dynamic stage 2022, which was green- that was presented. So that's pretty good all right. So we have some things here. uh Some of these items that never uh we didn't really do as much within 2021 as we would have liked. uh Maybe we'll get some of these and pull these over.

C

These submissions- these are the ones on the test of williamson's. uh Here's this one williamson's symbiosis. This is the idea of.

C

D

It's an old idea that.

C

um We've talked about in the past, like I had a row here.

E

Anyways, um maybe.

C

We'll do this later, oh now there we go okay, so let me get this so we have that one. We have a couple, others that involve diatom movement, and this is something I know we were. I presented some stuff a few weeks ago on that, um so that that might go somewhere this year. um This steve mcgrew's book uh eye of nature. We still need to go through that and see what we can do with it.

C

We talked about publishing it or you know, building upon it in some way, and then this quantitative comparison of archaea and shape droplets, which is still you know. I haven't heard from my knock on that. I still wanted to get in touch with him about the presenting on it, but I haven't heard back so um we'll see you know present in the main meeting here um I know susan presented. I think it was last meeting that we had of 2021, and that was a very good talk and that was on some of her work on.

C

There was a biophysics.

E

C

Well, I can't remember exactly what the title was, but it was a pretty good talk and we had a lot of there's some follow-up from that as well.

C

Okay, so add one dimensional ising is wolfram pattern generator. So that's a good one. Let's see this one here, I'll fix the formatting later and then line 26 basil area, which is, I guess, line 26 from 2021, and that's this one here.

C

I don't know this one yeah smoother jerky, so that's already in the list. So we have that in the list uh 2022 list we have some yeah. So this is the different diatom things um and then the one dimensionalizing is wolfram pattern. Generator yeah, there's some interesting things going on with pattern generation um like you know, in terms of both like kind of thinking, about how it's generated and then looking at some of the outputs of that visualization so again like this is something that you know. This list is like just very basic ideas.

C

We can go over this when we want and kind of you know, figure out what we want to do with it and where we want to submit things and so forth. So that's good.

C

uh Let's see! No, I don't want to do that.

C

Okay, I'll fix that later. The other thing is major tasks for 2020 2021, which I guess now is 20 20 to 2022. So I'll change that later- um and so we have all these cards here and I didn't want to have to go through them point by point, but I just wanted people to be aware that these are still here, and so some of these things are finished.

C

I don't I'm not logged in in the so that's okay we'll do that later. But some of these are finished. Some of the these two are finished, for example, or at least partially finished. You know I don't want to say they're not going to revisit that again. It's just that. You know that for now it's finished a lot of things here to do to possible things.

C

So this kind of ties into this submissions document, where we kind of plan things out, so, if you're looking for something to do a way to contribute, these are things that we can revisit and kind of make up. You know kind of find something to do. We have a lot of things that we've talked about. This is the way to carry over some of those ideas and align them with people, people's interests.

C

Okay, fish ladder is submitted, good, I'll I'll, go back and and um modify this uh later. um Okay. So the last thing I want to talk about today.

C

Okay, so last thing I want to talk about today is get into some papers, and I know I talked about some things already, but uh let's see I had some uh interesting things to come up. I know dick has sent me some papers in the pet over the break. So thank you for that, um and so one of the first thing I want to talk about is this microscopy folder. I have a couple interesting papers in here to go over um and following up on our microscopy interest in the group.

C

uh The first paper is from bmc bioinformatics. This is quick, piv uh the technique name, and so this is efficient. 3D particle image, veloci velocimetry software applied to quantifying cellular migration during embryogenesis. So this is velocimetry software. This is on, uh I guess, on a data set and they're actually developing the software around it and uh trying to get uh to sort of quantify cellular migration. So in this case you're not just extracting you know segmenting parts of cells, you're tracking them as they move in in movies.

C

So it's a series of images, you're tracking the movement.

C

So the background here is a technical development of imaging techniques, and life sciences has enabled the three-dimensional recording of living samplers at increasingly temporal resolutions, meaning that you can not only capture the movement in as like a movie. But you can capture that movie in smaller and smaller time uh or.

B

The sampling- oh, that's right, trying to try to speak to something because I think you're.

A

B

Are you screaming something.

C

Okay, it was muted, yeah, okay, no sound, okay um yeah. So let me go back to this, um so this is the yeah. So this is uh the technical development of imaging techniques. So the idea is that you have these videos, but they also have a higher sampling rate that you know you can have- and we've talked about this before, and so there are some machine learning techniques that allow us to take like series of images, time, series of images that are sampled at a fairly low sampling rate and increase those increase, the sampling rate.

C

Basically there you know the idea is that the algorithm will take the images in sequence and then infer the uh images in between and so in a normal time series uh analysis.

C

You can only do so much of that before you experience this aliasing effect and so in, and I don't know what the rule is in machine learning, but suffice it to say lower you know, sampling rates will not give you the data that you need to in for higher sampling rates or more continuous action.

C

um Unless you really kind of you know, work at it, and I I don't know but anyways this paper is a dynamic 3d data sets of developing organisms. So they have. These they've collected the data, especially for this. They call it dynamic. 3D allow for time resolve quantitative analysis, morphogenetic changes in three dimensions, so they have three like. I guess they have a three-dimensional data set where they collect data from like three spatial dimensions and then integrate that and then do the analysis on that.

C

So this generates terabytes of image data. So it's a lot of data, a particle image, veloce velocimetry. I I'm having trouble with that word today, piv as a robust and segmentation free technique. So this does not involve segmentation. This particular technique suitable for quantifying collective cellular migration on data sets with differing labeling schemes. This paper presents the implementation of an efficient, 3d piv package using julia, which is a nice scientific programming language that uh we've in my other group, we've talked about about it around the edges, but it's a very nice language.

C

It's not python, but it's something. That's really enables a lot of scientific computing, and so this package is called quick piv. Our software is focused on optimizing, cpu performance and ensuring the robustness of the analysis. So this this quick piv is three times faster than the python implementation host in an open piv. So this quick viv uses julia, and so our software is also faster than the fastest 2d pv piv package in open piv written in c plus.

C

By applying this to the 3d data sets of the embryogenesis of tribolium canastium, we obtain vector fields that recapitulate the migration movements of gastrulation, which is a phase of development in these embryos, both in nuclear.

E

And actin-labeled.

C

Embryos, so they have again a lot of the cell tracking that we do uh ultimately involves cells that have something labeled in the cell. So it could be the nucleus, it could be actin. It could be something in the cell that you can use as a marker and you can hook onto that to segment them, but in this case they're just using these labels as a way to sort of orient their analysis.

C

D

C

I say label I mean like they do this in a molecular uh way: they they use some sort of fluorescent protein or some sort of antibody stain that binds to something and makes it fluoresce. So it's like this nice fluorescent contrast in the image and so um they're able to detect translations and non-segmentable biological image data. So there are some of our images we've chosen.

C

Some of the data sets that we have because they're easy to segment, but in this case they're not doing segmentation they're doing this kind of just using these different labels and tracking them, and so they don't have to segment the data here and they're able to detect things in three dimensions: spatial features as you will, so that's good.

C

um So, let's see if they have anything about this embryo, let's see what it looks like.

D

C

Sometimes they put the images towards the bottom. I don't know if they have.

C

I hope that here we go so this is uh pictures of the embryo, so this is piv. Analysis was were performed on the actin signal of a double hemizygous transgenic embryo before and during so before. The top during is the bottom of gastrulation, so this is a and this is before- and this is during gastrulation.

C

So, as you can see, there's this gastrulation process where you get this sort of folding of the tissue, and you can see it's picking it out in this area here. So this is the image, that's being analyzed. That's kind of you know giving us differences in this. We have the arrows here. So it's showing the movement of cells and these arrows are, you know you can see a difference here between the two samples.

C

So that's what it's doing. It's picking up these uh labels and seeing how they move in the image in the video images and then it's creating these arrows that show the direction and the extent of the movement. So you can see that it has this there's a difference here in the different samples and then you, of course you can create numbers out of this and actually do some more analysis on it and then pi view is also performed for the same time.

C

Points on the nuclear signal when b and then the similarity selective average vector fields are shown next to the actin vector fields which I'm not sure what the difference is here, but you can see that you have differences again in the embryo and then these vector fields that result so, okay. So that's good, um so yeah! This looks pretty good. This is a nice uh and then they show this graph here and see the orientation similarity between each pair of vectors and the two channels is computed through their normalized dot product.

C

So this is the euclidean error here, the normalized dot product- and it kind of shows you this in the plot and then three patterns of cell migration can be distinguished in this data set. So the first is segmentable and trackable, okay and then the second is segmental, segmentable and non-trackable, which is this one snt and then non-segmentable and non-trackable nuclei. So.

E

These are classes of nuclei.

C

So you have a segmentable and trackable nucleis, meaning you could both track and segment these these uh nuclei, meaning that they're pretty obvious to the algorithm and they're pretty easy to sort of segment and and understand as a little dot.

C

Segmentable and non-trackable means that you can segment these images here, but it's hard to track them over time and then non-segmentable non-trackable are these nuclei.

D

C

Do neither in every as you can see, they're pretty. um You know you can't really make them out. You can make them out with the eye, maybe but not with an algorithm. So that's that's what we have when we're. You know dealing with this kind of data, so I think this is an interesting paper. The other one I want to talk about is this label-free three-dimensional imaging of c elegans with visible optical coherence microscopy.

C

So I think susan talked about this technique in her talk last year, um optical coherence, microscopy is this ocm method um and so fast, label-free high-resolution three-dimensional. Imaging platforms are crucial for high throughput in vivo time-lapse studies.

C

So this means that we want to look at these things live and they want to look at a lot of data and we want to look at them over time as a time lapse, and so this is the kind of thing you want to do for c elegans, because c elegans has a nice mode of development where you can do this and it's easy to analyze and easy to understand what you're getting so, um despite the needs methods combining all these characteristics have been lacking in this case, they're presenting this label-free imaging technique for three-dimensional data sets sub-micrometer spatial resolution and it's something they call visible optical coherence, microscopy or vissocia.

C

So vis-ocm is a versatile optical, imaging method which we introduced recently for tomography of cell cultures and tissue samples. So susan mentioned it as oct or optical coherence, tomography, there's, of course, a larger class, which is just the general microscopy, but it's the same thing more or less.

E

It's imaging with infrared light: oh good, oh hi, susan! How are you into the depth of the issue? Yeah? Sorry, I'm late! Oh.

C

Good, no problem.

C

Yeah, so this is uh yeah that so our method is based on fourier domain optical coherence, tomography and then this kind of tells you what that is by operating in the invisible wavelength range and using a high and a objective. Bizocm attains a lateral and axial resolutions below one micrometer.

C

Additionally, we use a vessel illumination, offering an extended depth of 40 microns, which is micrometers people call microns as well. We demonstrate the vis-o-cm image properties of while rapid, imaging, full-size living c elegans down to the subcellular level.

C

So that means that we can look at some pictures, perhaps and if we go into the pictures, this is the setup here. So this is the broadband light source and they've characterized the intensity.

C

Then this is the sample mounted on a plate, so they have these cell cultures that they look at, and then this is the setup here, the illumination detection, and then this is the spectrometer that they use for this. These sort of recorded spectrum with a subtracted background, and then you get this profile there. Anything you want to say about this. Susan.

E

um No, I haven't gotten into um optical clearance tomography like I want to it's just that it's a um well, it's a tomography. You do depth images and it can go to two to three millimeters, depending on how you set it up. Usually they say two millimeters depth, so this is a little shallower.

E

C

Are you familiar with the different types of light source you might use for it or.

E

Like yeah there's uh one, that's um spectrometry, which you use a broadband light source and then there's a swept source where you break out the different frequencies and you more or less get the same type of image or you it's more or less the same technique, except that one's a broadband light source and one is sort of incremental, so you're separating out the the bats.

E

And then you do similar processing with um the different frequencies. Okay, post-processing yeah, yeah.

C

So you see the step before you transform here.

E

Oh definitely need the 40 transforms to work with this and um some interesting tenser math that does um sparse sets, which I haven't looked into either very advanced linear, algebra.

C

Yeah, that's that's good. um I think dick had to leave and I had a comment from him in the in the chat. uh We were talking about pi the piv method. Before is it a problem? uh Real pi via flow run, a moving diatom.

C

So piv was this other technique where I know susan missed it, but it's this uh technique where you can look at different things like different markers in the in the uh you know,.

E

Say like a embryo.

C

Or a cell culture, and you can uh basically track them over time and get a flow map which is like a map of arrows, and what dick is saying here in the chat is that you could create a pivot floor on a moving diatom.

C

uh You could use like you know some sort of track, maybe a tracking die, or maybe some sort of fluorescent tracking technique to look at the flows around the diatom itself. So, instead of using something in the cell and making it fluorescent for us, you could use something around it. I think that's an interesting point, because this piv technique looks kind of interesting and, of course we have the computational tools to to complement.

E

So it's very hard to find a dye that works at the infrared frequencies that does that that do optical coherence, tomography um light tends to go. That type of light tends to go through air and water.

C

E

Readily so finding um a dye that works at those frequencies is difficult, they're, more or less water windows.

E

Dr sharif wouldn't like this, but I'm not remembering exact frequencies, they're 870, nanometers, 1100, nanometers and 1300, and something nanometers, and some of them are used simply because communication systems are at those frequencies and this the hardware for that is readily available.

C

E

I mean that's sort of my extent of the knowledge here. I'm.

C

Yeah yeah, that's that's fine, yeah uh yeah! It's! So that's the image there and I don't know if there are any other images they have. Sometimes they have an image of the okay. Here we go. 3D volume rendered images of cut sections of a young adult wild type c elegans. So this is our adult c elegans. This is where it's not a mutant, it's a wild type and you can see that they've, characterized the entire worm from head to tail. This is the this is what a worm looks like under a microscope.

C

You get the it's it's a model, so it's not like doesn't show a lot of the features on the outside here, but uh it basically looks like this, and you have a cross section here, so you're able to walk through the cellular structure um in d and e, so you can see that there's this cross section with sort of cell bodies all along this axis.

C

uh You have difference. uh This f is a transverse section, which is where you cut it from top to bottom dorsal to ventral, and you can see this sort of section where you have cells are organized. You know cut in that way and so b and c are what uh b and c are sections along the x-axis.

C

So those are just sections along this way from uh head to tail and you can see the structure in there. So this is this technique. Will slice up this c elegans in a number of ways that allow you to see the different anatomical views and walk through them? As you know, if you've done microscopy or even fmri, which is a technique for looking at anatomy, you get these slices, it walks through the slices you can explore through the slices and it looks like you know, you're moving through some organ or some organism.

C

You know from one side to the other, and you know you have these discrete slices, but in this case you, you know, they've been able to characterize all this pretty well and.

E

C

Scale of 100 microns here 110 microns and 63 microns. So this is this: this little cube here would be these dimensions.

C

So, along the surface of a c elegans, you know it might be like a couple of uh millimeters long and then the width is not even 110 microns and then the height is about 60 microns you ever take, and uh so that's the spatial scale we're looking at. So it's pretty fine grained and you can see that it's coming up with some pretty nice images.

E

Major advantage of this technique is that you can do histological imaging without harming the animal like they're, live and already said that, but it also doesn't harm tend to harm things. um I've done axolotl eggs at early stage and the actual legs have hatched, and in the case of the female that hatch, she had her own eggs and hatched babies that were normal yeah. I've already tried this or my friend elizabeth has.

E

It didn't seem to harm them at all.

C

Yeah so like the uh one of the alternatives for c elegans is something called cryo-em, which is where they freeze the sample and then they go through and they get these really nice images problem, of course, is you have to kill c elegans to do that? You freeze it. You know in liquid nitrogen, I believe so. This is a probably an uh advantage here as well, um and so this is the pharynx the intestine.

C

So these are just like. You know different parts of the worm. You can see it's pretty high resolution where you get like you can make out the vulva and oocytes germ cells. I mean these are things you would see like you know under maybe like a more a normal microscope, but getting this. This is a little bit higher resolution I think than normally.

C

um You would normally get in some cases, um so yeah they're able to do a 3d model of this, so they're able to pull this together and build it and so uh yeah. So this is uh just a benchmark. I guess this isn't like anything like where they're trying to discover something, but this is something that they're able to build a 3d model of c elegans, using high contrast down to the subcellular level, where you can actually identify things inside the cells and yeah.

C

I think that's good, um so that's that's our microscopy and then I wanted to talk about one more thing before we go. um I don't know if I brought it up here today, but I'm not sure if we want to get into it.

C

Well, I don't think I included it today, but um I think I might talk about maybe a physics paper here before we go um so this is an interesting paper scale, dependent irreversibility in living matter, so that we talked about active matter last time, and this is about living matter. So, let's get in a little bit into this. I just wanted to go over a little bit before we go so the abstracts is a defining feature of living matter. Is the ability to harness energy to self-organize multi-scale structures.

C

His functions are facilitated by irreversible non-equilibrium dynamics. While progress has been made in elucidating the underlying principles, what remains unclear is the role that thermodynamics plays in shaping these structures and ends their ensuing functions.

C

So here we unravel a fundamentally thermodynamic connection between the physical energy, dissipation, sustaining non-equal non-equilibrium system and a measure of statistical irreversibility or the arrow of time can provide quantitative insight into the mechanisms of non-equilibrium activity across scales. So a non-equilibrium system might be an organism. This statistical irreversibility is the arrow of time. So the idea is that you know you have this organism or this embryo that's developing or this organism that's living and time is basically entropy acting upon the system.

C

It's either you know developing and growing or it's you know sustaining itself where it's dying and falling apart, and this involves energy flows. So you need a lot of energy when you're growing or sustaining- and you know energy is dissipated when you're dying and you don't have any incoming energy. It's a very you know abstract way to look at it, but so they want to understand this mechanism across scales and so in an organism it would be like the cells or it could be the dna or it could be like other organelles in the cells.

C

All these things are driven by the same process.

C

Specifically, we introduce a multi-scale irreversibility metric demonstrate how it could be used to extract model independent estimates of dissipative time scales, so they use this metric of looking at things that are irreversible. With respect to this, the flow of time demonstrate how it could be used to extract these estimates of dissipative time scales like so. You know what is dissipative just means like it's. uh You know, there's an energy flow there using this metric. We measure the dissipation time scale of a multicellular structure, the active meiosin cortex, which is a thing with muscle.

C

It's a muscle uh thing that involves muscle building and things like that and.

D

Further observe.

C

That the irreversibility metric maintains a monotonic relationship with the underlying biological non-equilibrium activity. Additionally, the irreversibility metric can detect shifts in the dissipative's time scales when we introduce spatiotemporal patterns of biochemical, signaling proteins upstream of ectomyosin activation. Our experimental measurements are complemented by a theoretical analysis of a generic class of non-equilibrium dynamics, which is this these. uh You know the problem. We use something like differential equations to look at the behavior of this over time, eliciting how dissipative time scales manifest in multi-skill or reversibility.

E

Have they included gravity in this anywhere.

C

I don't know it's a good question. I think they're interested in largely in the energy flows, but that's that's interesting.

E

Microgravity and gravity and yeah it's another component of this.

C

Yeah yeah, so you know, I don't know how they. This is interesting, like some of these models of like entropy, they don't really consider like things like biomechanics or other types of uh things. So I don't know um how they're looking at this. I don't think they're considering gravity but or a tree. I don't know how they treat gravity in it.

E

Yeah, I was just thinking about that.

C

E

Well, the other day yeah how these dead reactions react differently in microgravity, for instance, yeah right. So if it fits in with this topic, really.

C

Right right, yeah, yeah, actually something maybe to follow up on. I don't know how people were thinking about this like.

C

Like especially like we've talked about like uh active materials- and you know I don't know like what the state of that is. If people really consider things like gravity in those models or not but um or a differential behavior of gravity, usually I think maybe they just assume a specific gravity.

C

E

Right, it's just that um I'd like to see them included. I know some people did a model of chemical interactions and I asked them well, did you include gravity and they said why yeah really well.

C

I'll tell you.

E

What it makes a difference, yeah.

C

I would imagine yeah and especially.

A

C

Especially if you're considering differences in gravity, I mean people probably just assume like one g, which is you know what you have on earth and that's that and it's like a vacuum. You know it's like uh the whole joke about the uh spherical cone of vacuum or spherical, something in a vacuum. That's like the physics standard physics model, uh yeah yeah, so I mean a lot of this is, like you know: they're modeling, these type of uh biological molecular biological systems.

C

So, looking at the energetic efficiency of molecular motors, for example, um you know- and it's just kind of looking at the different uh forces but again, like you, know, they're not considering gravity necessarily.

E

Yeah but they're um the motors, the molecular motors will operate differently in microgravity microtubules, don't form in the same way there's no convection in microgravity. So that complicates a microtubule's life. Shall we say it? Doesn't they don't get convection to help with uh the replenishing of the plastic.

C

E

The way I I see this, but it makes a difference yeah a huge difference, starting on a small scale, leading up to astronauts having a bad time in microgravity.

C

C

So a lot of times they consider like with these models. They also consider like just a closed system. So, like you know, it's like you kind of fix the parameters and you see what it behaves like under a closed system, but a lot of life is actually an open system which is kind of an interesting uh apparel. You know interesting thing. I don't know how you model equilibrium as an or model thermodynamics is an open system, but um you know it does make.

C

So any non-equilibrium process is accompanied by the irrever uh irrevocable loss of energy to its environment, often in the form of heat which uh commensurally uh commensuratively produces entropy in the environment. So again, this is in a closed system. Now, when you have, you know, obviously you're losing uh you know, you're generating entropy as you sort of as an organism behaves or as it doesn't replenish it well, it can replenish its energy, but you're still losing this entropy. So it's an inconvenient way to think about.

C

Like you know, this energy balance is to think about it as a closed system, but in an open system I mean you still get entropy, but then there's this question of. What's you know? How is how is a system able to grow and generate things and do things? That's not.

C

You know that sort of you know there's an entropy loss, but there's also a lot of energy coming in. So it's that you know it's one thing I don't. I don't really know how they're I know they're, probably thinking about it as a closed system and kind of like isolating the mechanism, but there's a lot more going on there that we can consider- and so again we have our equations here for thermodynamics and so active meiosis and cortex is a multi-scale dissipative structure.

C

That's their main argument here, and so this is something they were able to measure, and this is the ectomyosin cortex. This is uh these little meiosis molecules here so and you can see that they've got different positional fluctuations and position. Increment autocorrelation functions of these endogenous cortical, granules and they're able to probe this for force fluctuations in the cortex. So this is the ectomyosin cortex.

C

These are the molecules here and they're doing different perturbations to it they're able to they're the granules here. The f actin molecules the meiosis and molecules.

E

What's it granule.

C

I don't know exactly, but that's these balls in here I guess they're, just like with embedded within. So if you see in this image the red are the molecules, these white things, I guess, are the granules.

E

Yeah yeah: well, it's a movement of them, but what do they mean by granule? That's that question. Yeah.

C

C

Yeah um yeah, it's good! You have to dig into the paper for this, but.

E

um Okay, yeah well, like I can look at it. I captured it um all the time good.

C

Yeah, so this is, and then this is time irreversibility reveals the dissipative time, skills and levels of non-equilibrium activity and pattern cortex. So you can see that they're looking at these dissipative time scales, they're looking at the fluctuations and the intensity here, looking at frequency analysis and then lag time, which is uh d, which is this estimated irreversibility of this parameter for five different stages as a function of lag time.

C

Inset here, because the time scale, the peak, irreversibility or yeah peak reversibility is a function of row oscillation period.

C

So this is something that they've derived out of this uh graph, and this is the oscillation period s irreversibility time scale. It goes up as the row oscillation period increases. So that's an interesting paper. I don't want to get too much. You know I probably.

E

C

Hour on it, it's not, uh but it's an interesting way to look at the different problems, and um I was looking for another thing that I was going to show. But I was I couldn't find it, but next week I'll probably show that anyways.

E

Yeah, I'm sorry I'm late, I was like I would have what was the first paper you went through or.

C

It was that imaging paper was the um I can send them along. It was this piv technique, which is like a technique for imaging uh different way. You image different uh fluorescent markers and you don't segment the cells. You just track the fluorescent markers in a cell over time, so you have these movies and then you can track them. You can create vector fields which allow you to get a handle on movement and in collective movement and like they looked at an embryo actually, so that was a nice way to do it.

C

Yeah, okay! Well, everyone thanks for attending our first meeting of the year, I think, was pretty successful. If you have any papers, you want us to present in the meeting. You can do that and present it yourself. If you want, you can also present on different topics. I know there's a lot of stuff that we talk about every week, but you know one of the things about this is we'd like to see people present, maybe a short thing that you know. If they're have a topic, they want to really kind of explore.

C

You know, uh besides just kind of presenting papers like the way we do. We can kind of lay them out a little bit more. So.

E

Well, I'm working on a on a paper right now, yeah after I get it finished, I could do a short version of it or something for you. Good yeah, yeah, yeah,.

C

Yeah yeah, we'll also be in touch with respect to google summer of code, we're going to be working on that more in the next few weeks. So we'll be talking more about that. Okay, thank you. Everyone see you next week.

E

Bye, everybody.