DevoWorm Lab Meetings, 22 Mar 2021

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: DevoWorm (2021, Meeting 11): Onboarding, Math of DevoWorm, Simulating Developmental Processes.

Description

DevoWorm Onboarding Guide, Emerging GSoC-related Projects, Mathematics of DevoWorm, Superresolution, Cellular Automata and Artificial Chemistries, and Blastoids. Attendees: R Tharun Gowda, Jesse Parent, Aayush Kumar, Yash Vadi, Krishna Katyal, Bradly Alicea, Mayukh Deb, Mainak Deb, Akshay Nair, Susan Crawford-Young, Richard Gordon, Tom Portegys, Vrutik Rabadia, Debojyoti Chakraborty, and Shruti Raj Vansh Singh

A

Yeah, that's the only thing that seems to work so hey.

B

Bradley, where are.

C

You, how are you.

B

I'm fine it's a long time. You're you're prospecting: oh yeah! How are you.

C

Yeah, I'm fine.

B

Busy with some of the tasks like I got some internship just doing that: okay, good yeah, yeah and currently working on the digital bachelor's project. Okay, yeah- and can you tell me what is the specific uh task of the project like uh what is our primary motive to doing this project? I asked puncher about that, like he told me that to detecting several uh diatom uh like specific movement and all other patterns in that project, so is that the motive.

C

Yeah I'll talk about a little bit later um in more detail, but yeah we'll talk about it but yeah. Basically, that's the the idea, um yeah so nice to see you again, uh susan hello,.

A

Hi, I actually, I bought a rather expensive microphone, see this that they all seem to be about the same price, so this is called blue okay. So now I have to answer the phone.

C

C

How are you guys doing.

C

Yeah, I'm doing really good good.

D

C

And I've been so I've been communicating with a number of you on slack about your proposals and other things and I'm getting to them, I'm working on it as we go along, and so I mean I'm just giving comments in documents and uh answering messages, so you'll be getting those soon. um So welcome to the meeting, uh I don't know how many I mean people will be coming in.

C

I guess uh over the next, maybe 10 minutes or so usually that's how it works and we'll just get right into some of the materials here, um hello, tom, hey. How are you good? How are you doing all right.

C

It said tom p, so I was a little disoriented, but I knew that yes,.

E

Yeah, I don't have a camera on this machine. I tried one with the camera, but it didn't work. Oh yeah, maybe I'll try and switch over.

C

Okay, well yeah welcome.

E

Well, yeah gonna be uh good to see everybody.

C

E

C

And jessie's here, hello, jesse.

C

uh I don't know well uh surety's, not here right now, but like tom has he's worked with the group uh over a number of years and he's one of the people who's done a lot of work on cellular automata and the morphozoic platform. So we've talked a little bit about that in recent meetings, um so you know he's doing a lot of stuff with artificial intelligence with like biomimetics. He gave a talk in the group uh last year, so some of you are probably haven't.

C

We hadn't joined yet and so he's uh interested in that area. So if you want to he'll, be in the slack channel uh for open or for diva worm, if you want to ping them, if you have like, if you're interested in like the cellular automata area or some artificial intelligence topics, it might not says hello. So.

A

Welcome to the meeting.

C

um I we have a number of things today. uh I wanted to point out. First of all, let me share my screen.

C

We have uh akshay hello, actually.

C

So we have a new publication out. uh It was yesterday, I think it went live and this is a periodicity in the embryo paper. So this is usual myself and jesse. Were the authors on this and susan uh didn't want authorship but she's in the uh um acknowledgements and uh we this so this is finally on biosystems.

C

This is the special issue on waves in development, and so this is the final sort of stub for the publication. You know goes through a series of stages here. The first stage was as a preprint and we had that out for a while, and then we revised the preprint and then we did the final version which is in biosystems, and that was there was like a impress version of that, and now it's the final version.

C

So it's it's a pretty good paper. It's about uh the timing of uh cell division and cell differentiation in diff in two different embryos, c elegans, which is the nematode and indoneria, which is a zebrafish, and those are two model organisms, but they're, very different modes of development and different things going on in their development.

C

And so you know we can look at those two models and then we can also simulate cell division using a simulating data set that we created for this purpose and looking at sort of the speed of uh division and and that sort of thing so uh you'll have to read the paper to get the details on that. But it's a uh I. I hope that people enjoy this paper. uh You know it.

C

It was something that we were gonna do maybe two years ago and I got put on the shelf and then finally, there was an opportunity to go forward with it. So we took the opportunity- and now it's out.

C

uh Okay need to get access yeah, if, let's see, uh if you go to the devo worm, uh weebly site thanks tom, because the diva warm weebly site actually have the biosystems version which is pretty similar to it. But it's um you know: you'll get a sense of what's going on a little bit crumb um yeah, so that's periodicity in the embryo, um I I can post a version of it in the chat in the slack channel as well.

C

um Next thing I'd like to talk about. Actually, this is something I pulled up. I had this in the uh in the queue of things that maybe need to be done. I'll talk about that in a little bit. This is a paper called morphogenetic patterns in the theory of deep learning- and I have a couple people on here as authors right now, but we can add more authors if people want to contribute so the idea this is it's about sort of deep learning and how deep learning networks can tell us something about uh development.

C

Specifically more. You know the creation of morphogenetic patterns and maybe some other phenomena. So we talk about. You know some of the aspects of uh deep learning in gans and then you know kind of going through, maybe how some of those things might provide insights into development?

C

um It's it's pretty rough. It's a abstract still with some references, but you know we'll flush this out over time, um and you know I I just wanted to bring this up because I want you know if I want to give people a chance, maybe to get involved with it and see what we're doing under the hood. This is something that I've I've shared with a couple people in the slack channel and it's been a while since we revisited it.

C

So there isn't much here right now, but I'm trying to get this together so maybe can submit it to a conference or somewhere else. So we could probably submit it to a conference as the abstract version, but we might also want to do a full paper on it.

C

Okay, so uh next thing I'll do is talk about the um onboarding guide. So for those of you who are looking to join the group as a member of the that, maybe is like a gsoc person or you know doing a proposal in gsoc we're looking to get involved in the group. We have this onboarding guide.

C

That's now available- and this is uh let's see mayor, contributed to this and krishna contributed this, and I pulled a lot of stuff together here and there's a lot of stuff that we've had in different places, and so the purpose of this document is to give people an idea of what's going on, indeed in in diva worm. So this is, uh I have to reload it okay.

C

So the purpose of this document is to give people sort of an essential guide or give a worm, and you know the different parts of the project and what's going on, but with an eye towards you know like gsoc and contributing. So this is uh sort of a little bit on the open room foundation.

C

We have some links here if you're, not in the slack. Here's a way to get into the slack.

A

C

A

C

Which is different than the diva worm github? This is the openworm website, url and so forth, and then we have some resources for some of the open worm projects. So you know, diva worm is one open worm project. There are other projects like c302 and some other uh projects that are, you know doing a lot of different things at the worm. Looking at you know simulating the activity of of neurons looking at the connectome looking at some of the other things going on movement, behaviors, and so those can be found here.

C

I need to add a couple more resources here, but- um and this is a talk that I gave at artificial life, 2020 on open worm and so you'll find those things very informative. Then we get into the diva worm group and some of our things. We have a group website. We have our own github repository.

C

We have devo learn which is an organization devoted more to like machine learning and deep learning uh and uh analyzing like developmental data, and so we have all these things here and then we have some introductory slides from krishna on. uh You know how to maybe contribute to open source, uh how we can how we participate in gsoc 2021, so we participate through incf openworm. We pre we're.

C

We belong to open one, but we participate in google summer code through incf and so uh that, hopefully, that clarifies things for people gives you some ideas about what is open source. This might be a little bit too. You know low level for some people contributing or applying to gsoc, but I think, as a general resource, it's a good way for people to kind of you know understand some people who come to the group don't really know what an open source contribution is.

C

So we want to make that clear to people and then here are ways. Maybe you can contribute to diva learn. This is just a process of how to put a contribution into play. uh You know going through looking at. What's going on uh trying to find an issue to address uh you know, maybe you create an issue.

C

Maybe you get in contact with people, but you then work on the issue and we had a couple issues this week where people worked on them and submitted pull requests, and so now their pull requests are in and um so now you're formal contributors to the project. um Then there are the project descriptions. So we had a question at the beginning of the meeting about uh digital basil area. um I know 3.1 is popular.

C

This is upgrading divaler and this is the diva learn software that my oak is is maintaining, um and this is his uh g-stock project from last year. This project revolves around working on that uh uh software platform, which is essentially a pre-trained model for analyzing developmental data. So you know analyzing data for embryos, the c elegans embryo, but other embryos as well- and you know developing techniques for analyzing those data.

C

So uh the this project, though, is upgrading the capabilities of that package, and so I think a lot of you are already aware of maybe some of the things that that entails, uh there's a formal, a project, description for all these projects on neurostars.

C

So I think a lot of you are familiar with that, but digital bacillaria actually is. um This is a project that started where we had this diatom, it's a it. It's a a single cell colony, so it's a bunch of cells that are behaving together and interacting. It's called bacillaria.

C

It's a group of diatoms that are these long rod shaped diatoms that exist in colonies and they move around in different ways in the water column, because diatoms are marine microorganisms, and so we have a lot of video of these organisms and and diatoms are actually fairly common in uh aquatic environments.

C

uh So you know, if you anywhere in the world, you can find them.

C

You can isolate them and with some luck you can even culture them in a dish and what we're interested in here is looking at these microorganisms getting microscopy images which we have a library of and in this project your job would be to analyze these images, so you'd be doing some segmentation, you'd be doing some annotation and then working with the video and then we'd be doing some things down the pipeline, which you might also be involved in, depending on your skill set such as analyzing the data, and you know maybe building simulations or visualizations of the data, and so that's that's what that project is the the diatom world.

C

Is you know they don't have a lot of sort of digital uh models of organisms here we have in in diva worm the reason it's called evil worm is because we focused on c elegans, which is a very uh well characterized model organism, so c elegans, there's a lot of secondary data floating around there's a lot of data about different aspects of c elegans. uh Surprisingly little about c elegans in the wild, but um by comparison uh diatoms are pretty poorly characterized in that way.

C

So our goal here is to sort of build a digital diatom.

C

uh So it's a it's an organism: that's not a model organism, necessarily that where people study it to study disease or other things, uh I mean people study diatoms for like biofuels and things like that, but there's no digital characterization of it so and I have in this onboarding guide. I have a couple of links, uh so we, the this project, has been uh worked on previously by oswald singh and he's developed. This platform called diva warm ai, which actually has some of the models for digital bacillary, but also for some other models.

C

Machine learning, type models for analyzing. You know microscopy images and things like that. There's also a link to the github repository for digital basil area, which is you know not in the newest state possible, but we have some references there, so it hasn't been updated in a while is what I'm saying. uh There's some resources, though here for you, if you're interested in working on this and maybe doing an application or working with the data, we have some raw video data, which is microscopy data. So we have microscopy.

C

That's in the form of video files, so you'd have to take the video files apart and analyze the individual images or you know if you want to recognize things through video analysis.

C

You know there are different approaches you can use to under to segment the images and then we also have temp delimited versions of the data. So these are data that were extracted in in our previous year's work and those are available for people to look.

F

C

Maybe use to validate their results or whatever we have a paper here which is uh on the biowork. I have a net talk that talks about a lot of the analysis. That's already been done, so it gives you a sense of what we're trying to do at least to some extent, um at least in in terms of the image segmentation and then there's this recent presentation and basil area movement.

C

This is the presentation that we gave to neuromatch, which is a conference, an online conference, and this had to do with looking at bacillary behavior, and so this is a model of behavior for basil area, and so you can find all those links here.

C

Finally, there's this digital microsphere project, which is the uh it's a project where we use a special type of microscope where we capture embry and embryo from like many different sides at once, so we're observing it from many different project, projection angles or points of view. And then we take those points of view and we want to stitch them together into an atlas or some sort of like visualization that we can explore, and so there are two papers here on this.

C

One is on the uh what we call ball microscopy and susan is actually the person who's uh heading up this effort to acquire this embryo data through this using this ball microscope. So she puts an embryo in the in the stage and then she has all these mic these these microscope. These cameras around different.

C

You know sides of the embryo and she's able to get a bunch of different angles at once, and then the idea is to stitch these images together into an atlas of some type, and this paper appears actually describes his vision um in terms of sort of like google maps but google embryo, and this is about 10 or 12 years old, but I think you'll get the idea of what we're trying to do it.

C

There's a you know: there's a notion of like creating like a map or a like a you know, something you can explore a surface you can explore and there are different ways you can do this, so it isn't just you know, creating something that looks like google maps. It could be any kind of like you know you could take like a sphere and something that you could move around and turn with a cursor.

C

It could be a flat map. It could be some other type of visualization, but the idea would be to take these data and project them to something that's explorable um and then so there are a lot of things that this can lead to a lot of different insights to be made a lot of different ways to display the data.

C

um So there are a lot there's a lot of potential here. um So that's that's all. That's almost all that I wanted to talk about with the onboarding guide, except that I'm also developing some of the biological resources here.

C

So what we also want to have here at the end of this onboarding guide is, I want to add some references in on on diatoms, because we don't have that yet, and I think that would be useful to people, but we do actually have- and this is from years past, we do have some references to c elegans biology.

C

So if you're not up to speed on your c elegans biology, you can put it in here um and and there's a general faq, so you can find out more about like some of the uh some of this language is like. uh I need to be more specific. These are c elegans data sets, but I think we're coming along on this.

C

I think this would be a good resource for people just joining new york and, if you're interested in contributing resources to this, um please feel free to you know, make a pull request or contact ian slack or you know whatever, and we can uh keep improving this onboarding guide.

C

So what do we have a bunch of comments here so after the onboarding guide link? uh Let's see uh yes, can you please um upload the recording of this meeting? Yes, that happens every week um jewel of the sea.

C

Devos says uh susan says how big are the basil area, so they're about 90 microns long, uh maybe a little bit more, maybe a bit less depending on the well.

A

C

They're they're, that long, you know it depends on the diatom they're diatoms of many different morphologies. Some of them are round some of them. Are these long uh rods? Rod shaped uh things like two categories round and rot yeah. There are several different categories actually, but those are the two main ones. So you have you know different uh genie and species that are, you know they have different shapes they're, characterized by sort of silicate cell walls.

C

So it's made of silica, which is different from a lot of other organisms, um but they have you know they have this. They have some really interesting properties, as uh as organisms go, so they have these silica cell walls and they have their own. You know cell division dynamics and they have their own growth dynamics and they know you.

A

Know this isn't really.

C

Developmental biology, but it kind of is because these colonies grow and they move so um the digital vascularity project isn't really focused on the development so much, but um and as much as they have a developmental stage, but um I'll put some more information about diatoms up on the um in the onboarding guide. So what we can get some of that up there and make sure people are on the same page in terms of what's going on there.

B

So, like final aim of the project is to create a uh digital version of the open one like the open one project, uh they created a cybernetic uh one. So for the digital bachelor's project, are we going to create a simulation of the same kind of.

C

Thing uh I think we yeah, I think, we'll- probably have something similar. I think a better way to think of it, though, is like a digital characterization.

C

So you know we can actually take these organisms and we can image them and we have. Then we can get some information about their morphology and we put that into a digital format, meaning we can attach numbers to it. We can have like a model of its shape. It's size, it's like maybe location during movement.

C

We have all these different parameters that we can store and then we can reconstruct this, maybe in a simulation and that's kind of what we're doing with open worm with c elegans as we're trying to characterize all the neurons and the cells in the morphology of the worm and then we're trying to get.

C

You know a better characterization of the worm and its behavior and all this other stuff and people talk about you know uh making a sort of you know, building a digital organism. Well, digital organism is a lot of work and, of course, you know you're not going to get like down to the level of like genes, or you know particle physics, but you're going to get this model that that is somewhat like approximating what you would see in nature.

B

Like a approximate digital.

B

Like by this deep learning and machine learning stuff and after that, we are going to reconstruct this in a digital form.

C

Yeah yeah and that's I mean that would be the ultimate goal and of course you know it's a long term goal. It's not going to be done in one year, but that's it. Yeah.

B

Yeah, like I think that there is a lack of data like uh we need some more data regarding this. So uh can you think like? How can we collect like if there was be 100? I think there was around 100 images for the diatoms, which are labeled.

C

Yeah yeah, I mean we're doing a very a very limited scope here and we're actually getting the images from a collaborator. So I mean there are different ways you can. There are other, but there are data sets online for not for diatoms so much but for other model organisms, and we.

A

C

Something called evo zoo, which is a sort of repository of different images that exist, so people will image, uh you know, do high resolution imaging of different organisms and they'll, you know, make the data public, and so we have like a repository of those resources and some other resources that we've kind of put in one place and then we've processed those and and created, like you know, numeric descriptions of those things, and so those.

A

C

Those are all available as well, and um I don't know yeah. I I need to make sure that the devos stuff is is clearly demarcated in the onboarding guide, but.

A

I think that's.

C

All those are all good questions.

B

Yeah- and I think it's a very good aim to that, uh like understanding the simple organism, first and after that, going to more difficult one yeah, I see your digital microsphere project as well. I see some of the recordings uh on the youtube and I think it's a really great project that we are going to try to create a map of the um like uh embryo morphology.

A

B

uh Try to understand it by this, like constructing the what you said that google earth it's the same kind of thing like a map, but I understand.

C

Yeah yeah, so the idea would be to take the data and make like a visualization that you can, because the idea is that you know usually in a in a study. You know if you took like a an embryo when you put it on a slide and you took microscopy images. You would get like slices of the uh specimen, but it would be at one angle, so it would be, one orientation would be like sort of the top of it and then yeah. This would give you like a view, but also like over time.

C

So in an embryo things are always changing, and so the idea is that you can image the thing as it's sort of changing over time. um You know if you have different samples or different stages of development.

C

You can get that information over time and you can get different orientations of it at certain time stages and you can make connections between different stages of time. So you know if it's like an early stage embryo versus a later stage embryo.

C

You know not only what's going on in terms of the divisions but, like you know, is it turning. You know if, if there's something going on like rotationally and capture, that or or you know, cells moving around at the top or the bottom of the structure, because they often do this where they move around the embryo during development. So all those things you want to be able to capture as well um yeah there's the divorm youtube account.

C

We have uh yeah length 70 to 200 micro uh microns, with five to eight microns. That's for the basilaria size.

B

I think I think it's a good uh resource for the like.

C

Yeah yeah, it's a good resource for diatoms um yeah, so so we've got some people interested in in proposing projects on that, um and so susan had a couple things to say about the ball microscope. The ball microscope has a one millimeter field of view, so just wondering if I could get 3d images.

C

So these are you know you can also make yeah three-dimensional images as well. uh You know, depending on how you you can play with the images we can get a lot of information out of these images. The ball microscope can be adjusted so that it has several centimeter field of view as well. So we can get very wide angle shots in the ball microscope.

C

I don't have a picture of it, but I've shown in previous meetings. It's basically this.

B

Yeah there is an image. I see that uh that it's. uh I think that some of your members sent the image of the microscope and it's like.

A

B

A

A football if we can.

B

Imagine that is a footwork into the microscope. If we put a uh like a sample of the embryo- and we can look at that like there was some pain kind of like, like a pain or pencil kind of structure around it and via it, the information is going to be a computer or something like that.

A

Yeah yeah the uh pencil structures on it are actually the microscope cameras. Also yeah and there's I've got sorry how many um I can go up to 14 microscopes with that structure, but right right now I only have nine.

A

C

Yeah, that's great um so yeah and then uh akshay asked uh yes susan. I was actually planning to do sort of a mercure projection approach which we may stitch all the embryo images together.

C

So yes, an example would be like a map of the world where you have latitude and longitude and then uh the coordinates would be like a 3d projection, yeah, so uh yeah, that's that's! A good example is like a map where you take like the surface of the sphere, which is the earth's surface and you pull it off and flatten it out, and so you need to do.

C

You know you can do that, but you also need to have a coordinate projection, because when you flatten it out you're going to have a lot of distortions and so they're, actually, aside from the workloader projection, there are other types of projections that people might be able to do in their projections that are better in terms of capturing the information at the top and the bottom at the poles of the image and so like on a on a a regular flat map of the earth.

C

For example, in some projections you got a lot of distortion at the poles. Things appear to be much bigger than they are and things that in the equatorial regions appear to be much smaller, which is a problem if you want to know what's going on at the top of the image and there's a lot of the polls in in development. So it's um you know it's important to get that information, but there.

A

C

So there are other projections that you can use, but I think yeah, using these kind of 2d project, coordinate projections as a way to approach and analyzing the data. um You know there are other ways as well, like you know, uh being able to uh there. There are a lot of mathematical models that one can use um for doing different things, um and we can. We can talk about those if you're interested, um but I I don't have any good references right now, but um those exist, and so, uh let's see, akshay, also asked otherwise.

C

An alternate approach would be to segment the images using opencv to obtain the embryo part only later on. We can use later on or can be used to obtain each cell of the embryo. After that, the stitching can be done of the obtained cells to make a 3d plot.

C

So I've put some data, I don't know if it's in the onboarding guide, uh it should be in the onboarding guide. I've made a like a link to some of the data.

C

uh Yeah there's some data in here. I believe. um Maybe I don't have it in this one. Well, there there are some data that I've made available and I will put the link in the onboarding guide and uh this. This is like a sample of data that susan gave me and it's just like a sequence of events.

C

So, like you know, you have the sphere and it's not like every angle. It's just like a single. It's it's an earlier version of this. Where there's like a flipping, uh she has called what she calls a flipping microscope, which is where you, you know. Basically, the embryo rotates in water, it it's turned upside down and then it turns upright, and so by doing that you can see most of the surface of the embryo, but it rotates over time.

C

So you have to like uh map these images to some sort of rotational movement, and so someone kicked out debo. I don't know what happened, but.

B

Maybe can you turn off the like the anonymous motor, something like that? I think uh I think uh it will like everyone, can have the moderation workload so yeah like google me, you can only take control.

C

I'm not sure why it's doing that, but.

C

Yeah, I don't know why I did that it. I don't think, there's anything I I have on my end that I can.

B

C

Okay, yeah all right, sorry, uh so yeah susan can I have your email of that. We have any presentations today. We may have some presentations uh so yeah. Does anyone have anything to present today or.

C

I don't know if any anyone doing like any g sock stuff has any presentations they want to do. Last.

F

C

C

F

So basically, what I did was: firstly, I accumulated.

F

Using various methods such as binary, thresholding, candy, educators, any taxes and manually, so as you can see, the individual cells on the right, which are segmented uh lightly in white background, and then I went ahead and.

G

F

G

F

Deep learning architecture, so, basically, in this architecture we have, I replace the normal combinations of residual residual blocks to improve the long term connections flow include the learning of the model, uh as uh as we go as we like, as we go deeper.

H

And deeper into the.

F

Framework uh there might be lots of information with the user of normal translation.

G

F

So I basically replace them with a residual.

H

F

Residual and identity flow, and then we I had the results here to make. If it is in the.

F

Wherein, if you closely look on the left, the top side, you have an option to upload a particular image which you want to analyze and then.

A

When you click on click to analyze button, the gui.

F

Runs in the background and generates the segmentation mask and then displays the input image, along with the prediction uh to the new color to the end user.

C

Okay, that's very good, so this is uh this is the gui, and this is uh so. We have the divorm ai, but it's you know it's still in a work in progress, so I could see this as being something we could put into the divorm ai platform.

C

So I'm not really sure.

D

A

The images that you are.

F

Showing, as prediction.

D

Work is pretty good, but, like uh the objective of our uh problem is just to clear: is that we need? uh We don't only need to like uh segment these using binary masks. We need to analyze them as well. Then we have to find parameters like the height of each individual cell. How many cells you can identify like uh how many, for example,.

A

If you are taking or some sort of boundary.

D

Box, how many boxes are there per frame so because, like we have to take all these frames in continuation, and then we have to analyze like what is the uh predictive movement for this uh particular mom? That is being shown in the video frames so like uh whatever you have done uh this? This is a good starting, but you have to like evolve with something so that we can analyze it like. You can't analyze the individual cell in the diagram. The segmentation is pretty good, but uh I think you'll get my point.

F

Yeah yeah, like I, was just working on the prototype uh last week. That's why you know just to lay out the groundwork.

A

F

So that I can build upon it later so like on developing the jio platform as well unit architecture, and now I have.

C

Yeah, that was good. uh Thank you for clarifying um some of the aims here uh so uh worked on this in 2019 and he's done a lot of. uh He knows you know kind of how this would proceed.

C

um Yeah you're showing the the masking, so I think it's masking, but also the annotation of the cells, individual cells and uh being able to get those, because you know the whole cup that's a whole colony of cells there and so being able to get the individual. Because what will happen is the individual cells will move around relative to one another so having the individual cells uh marked out and being able to annotate them? uh Even you know, because they do they don't really change shape, but they change position and so um so good work.

C

um Anyone else want to present anything or any have any questions about this.

F

F

D

Generated annotation for the.

F

Folder one and then I just started prototyping another block from that one. That's fine! The thing is um how good are the iou scores or the losses like how.

A

Accuracy, which is pretty good yes,.

F

We can like easily get 95 if we expand it.

F

We want to use multiple models.

F

C

Yeah, thank you. um So that's that's uh again, that's a good uh job and we're. We have a number of approaches that people are trying on these different uh data sets. So uh if you have anything to present, you know if you're doing work on putting your proposal together or you have something that you're doing you want to show to the group. This is a good time to do it.

C

If we don't have anyone else, I can I can uh move on to the next order of business.

C

Comment in the chat here: okay- pretty interesting! Yes, very good! Thank you. It's very interesting, um so I want to talk next about this. uh Well, we can talk a little bit about the submissions, but I did have this thing that uh I got contacted in by someone in openworm in another works on another project but they're interested in putting together this thing.

C

They call the mathematics of open worm, and so I'm trying to put together something called the mathematics of diva worm and uh what this is is it's like um it's a way to sort of boil down a lot of the topics that we're doing in the group to sets of equations or models that people can sort of summarize and understand.

C

um So these are some of the ones that I have here, we'll probably add more. um I have a couple of different categories. I have developmental function, a spatial complexity and self-organization.

C

Those are two categories and then, uh in the spatial complexity category I have this model of an embryo where I have this set of parameters that we this is sort of the model we do when we analyze embryo data. We, you know, we have what they call cell tracking data, which is being able to track the cells and assign a specific three-dimensional, coordinate position to each cell, and then you have things like time and division angle.

C

So the time is when you, you know capture the image in terms of how the embryo is unfolding, and then the division angle is when you have two cells that divide and start to move apart from one another. You know what is the angle of that, and so those are. Those are like five parameters that I've put together here and we've. We have a paper on this from 2014. So if you go to the davidwarm website, you'll see the link to the paper, but we have.

C

We can have other parameters here as well, and so this is one sort of approach. Another thing that describes, uh uh maybe spatial complexity, is this idea of a cellular automata. So this is a von neumann neighborhood and this is the radius of the von neumann neighborhood. So this is the equation for the radius of that neighborhood. So you can have von neumann neighborhoods, which are the immediate cardinal direction, neighbor cells.

C

So it's like the cells to the right to the front, to the back and to the left. You know you can have many different orders of neighborhoods, so this is an order. One neighborhood with a radius of I think one, but you can have much larger von neumann neighborhoods. So this is just a characterization of this um idea. This concept of a nearest neighbor, neighborhood and.

A

C

I would love what I'd like to do eventually is to take these. uh You know make sort of like figures that describe each of these categories and then like write a paper or a short paper describing these things, and so you know I don't want to just leave the visualization there. I want to actually describe what it means, and so uh this is seems a little bit like out there now, but there's going to be some description of it later.

C

Developmental function is another example, and for that we have three different uh sorts of data structures. We have complex networks where, in development we have node attachments, so these are like our embryo networks or networks of cells or something where we're adding nodes over developmental time.

C

We also have neural networks, which of course, we use in things like deep learning, but are also like in the connectome. You see neural networks, which are a little configured a little bit differently, but they, you know, that's the basic structure and you have lineage trees. So you have this. This is a chart that describes cells when they divide and they form new cells or daughter cells, and then they divide again and they form daughter cells, and this recursive process is described by a lineage tree.

C

So what I wanted to do with this is to sort of put together like the major equations and data structures involved in development and then describe them, and the person who I talked to in openworm said that you know. Maybe we would submit this to something like worm base or worm book as sort of a document for the community of c elegans researchers, but it would be more, you know, general of general interest to people interested in like computational development and how to think computationally about development.

C

So this is another thing, that's sort of like start, you know we're starting out on it and if people want to be involved, they can get involved. They can, you know, contribute something to it.

C

So then that brings us to the submissions list, which is the submissions that we have and the submissions we might be making in the near future to different uh journals or to different uh conferences or to other venues.

C

So, and those of you have been in the meetings before kind of know, the drill there's this list of reference, this list of things. These are things that may happen or may not happen. We just need to kind of move them forward.

C

um So this is the uh these green uh rows are things that are done, so we have a couple things that are already done. I started this list at the beginning of the calendar year, so we're already getting there.

C

So we have uh a couple of things here: evolution 2021 we've been talking about submitting some abstracts there. I think krishna has one on kill. The winners have uh killed the winners, which is something he presented on several weeks ago, um and we also have this euler paths for life, which is something I've been working on. I showed it in the meeting as well, if you're thinking about submitting to virtual evolution.

C

This is the link to uh go to the submission portal. It should be open now, but the deadline is april. 30Th, so you'll want to work on. uh You know your abstract and submit it by then um the cost is you know it's not as much as it would be if it were in person, but you can check out the costs at the link at the meeting site. So uh again we have you know and people can submit if they feel like they want us have something they might want to submit to this conference.

C

They can maybe put it on this list or go ahead and submit it yourself, but I um I I actually like to revisit this kill the winners submission uh if krishna, I don't know if he's still here here, he is. If krishna wants to talk about that in the next week or two, maybe we'll get back and look at that.

C

um Actually, following up on the last one, jesse says I'm interested in how the math models work. Is that on github anywhere? ah The answer? Isn't we don't have a github repository on this, but this will be sort of like a next step. So stay tuned, uh krishna, we're gonna, say something: they're, muted,.

I

uh Hey hello, everyone hi bradley nice to see you so yeah I'll, be. uh I can present that thing um just we have another paper coming up. That was on the bias, so I I like to present it is. Can I present it today, yeah.

C

You can go ahead and present it if you want.

I

Okay, you can complete with the concerted lines. Then I'll start.

C

Okay, yeah yeah, so let me go through this really quickly, um so yeah we have that. We have also the diva learned paper, which is something I've put on the back burner. Unfortunately, but I'm going to work on it and this describes our pla our divalent platform, so those of you contributing to google summer code are pushing to the diva learn repository and we have.

C

We have a bunch of new people who have uh issued, pull or have fulfilled, pull requests. They don't know if we have anything on the pull request list here, but this is the divo learn repository where a lot of you have been contributing.

C

uh Thank you for that uh and, if you you know their pull requests open right now, myoca is taking care of all those pull requests. Thank you for being a maintainer, my hook, uh it's a hard. You know it's it's hard work, but it's good to be able to maintain these resources, and so well, maybe next week we'll do a little bit more on on the different pull requests um on this. But uh suffice it to say that's a main part of the steve learn platform. Well, we also have a lot of other things here.

C

We have the general biological model, which is a machine learning model for analyzing different types of biological data. We have uh the c elegans divo learn, which is it's a web app to support the c elegans part of the diva library. So this is part of this divo uh diva worm ai initiative, and we have a repository of model organisms here which needs to be worked on because we don't really have a lot of descriptors in here, but there's a lot of stuff. That's coming up on this platform.

C

So there's the software, but there's also the platform, and this paper is about this platform and it's still we're still working on it. But if you want to be involved, you know you can start by contributing to the divo learn software, and so uh we also have this growth form and theory of deep learning paper, which I showed.

A

C

We don't really have a target venue for that, but it's um you know it'll happen. um We also have this bacillary non-normal, cognition paper that is coming up. We have to write a paper by april 30th, I'm working on that paper. um It's we would put a an at. We submitted an abstract to a book chapter which got accepted, and now we have to produce a paper and that's.

A

Where a lot of this.

C

Bacillary analysis is going into things like this. This is uh open norm. Oh there's, an open room poster that uh someone else in the open worm organization is interested in submitting to the international c elegans conference.

C

I think that's something that we can just say: d worm wants to be involved with, but then maybe later down the line, we'll actually put some content into that poster, and that is that poster the international c elegans conference.

C

The deadline for abstracts is march 25th, but the conference is in june, so we'll be talking about that in the next couple months. um We also have uh this boring billion book, which is this idea about uh deep evolution and why things were static for like a billion years of earth history. So that's still being worked on.

C

uh I submitted a proposal for the devil, learn a talk on the devil, learn platform to the incf neuroinformatics assembly. So that's coming up in next month. I think and we'll be talking more about that in the next coming week in a couple next coming weeks, uh preparing materials for that and that'll hopefully be something that people can attend online or you know, however, they decide to host it um and then there's this mathematics of evil diva worm, which is this worm book thing, which is pretty rough right now, but we're getting it together.

C

And then I wanted to point out for a lot of machine learning. People there's the deadline for their ips, which is on may 19th for an abstract and may 26th for a full paper. If you have you know, if you want to write a paper, time is pretty short on that, but you can submit something either to that or to one of the workshops which are have a little bit later deadlines.

C

So that's the paper review. I just wanted to review those deadlines and now krishna. Would you like to present.

I

So hello, everyone.

I

Can you see my screen.

C

I

I

Okay, so here it is.

I

And it's on biasing, artificial intelligence. So I'll just give you a quick, you can say a quick. You can say recap of what other things that we are doing here so uh in psychologists have considered that there are more than you know, 200 pies, in human brain. uh We, you know, we tend.

A

To make a lot of assumptions- and you know.

I

We work on them without you can say, checking them and confirming them. So that's where a lot of stereotypes and a lot of bias comes into play, so the.

I

A

Efficient and as.

I

Good as it should be so the only seven types of price is, you can say, exclusion by sample price measurement price. So then, for example, if a person is designing an algorithm- and you know he made a perception, he wants to see the things that he like to see. Then there is a quotation that you see the word as you want it to comes into play. We have measurement bias like whenever the data that is reported to us is incorrect.

I

It is not uh accurate, then we have record bias which comes in the labeling stage.

I

I

I

For example, if you are feeding professions so it was found that in google tran translation, when men were linked to doctors, women were linked to nurses and it should not be there because both men and women can be doctor. So that is something that is something actually like. Some of the machine translation problems when men were targeted as engineers, so women were not targeted.

A

I

Instead, they were targeted as teachers, so there is nothing wrong with teacher or engineer, but the translation.

A

I

Happening was not you know, it was not fit for real world and, for example, that measurement was anyone who has looked from computer vision know that we usually have a drastic drop of performance when our algorithm works on real-world data, because the.

A

I

Are having is, you know, is very good resolution data with good lightning and the sharpness and everything of.

A

I

And this is, you can say the summary of the paper that we are working on and I want to present one more thing.

I

C

Well, I think it's a very good point that, like even like a lot of the stuff we do in the group with respect to data is some of it's high resolution, but we also have to remember that we're dealing with very small samples relative to the actual biology that's happening. So uh I think everyone should keep that in mind as they're doing their work. uh You know, but that's fine yeah. We we can talk about more about that in meetings to come so yeah.

D

E

D

Studying one point like one of the main biases that I have encountered like, while working with the.

D

On a particular set of people like, I am not learning something. How.

G

Magic or the world is dealing with right now, like.

D

G

D

Take can fake generations prefer anything just for, like only.

I

One set of faces like.

D

A

Because all the images.

I

That we were having, were you, know, clicked in the laboratory. They were having white.

A

Backgrounds and then a real.

I

uh Data was put, it was, you know you can see, they were not similar to the data that was given for the.

H

I

H

The lightning conditions and the.

I

Quality image, quality and the background, and everything is really affect the data. You may be getting loss and good accuracy while trying, but when it comes to real life, the whole the family changes and the whole picture is somewhat very different, and that is the thing that is.

I

But how they are related to the real world and the quality of the data is very much free ignored. So that's it any questions.

J

uh Not a question, but I really like what you're doing krishna um so good work.

G

Yeah, at least we are doing it, yeah.

J

And there's this like there's, I have to have to do of give it more of an update I'll do a bit later this week on the new york celebration element, computing stuff, which is uh related to you, know, bias and ethics and ai.

J

um This is like, obviously, this all tied together for that and for real for people who don't know much about that. um We uh we part with with the other group. We we uh we're working on some ethics related stuff, so if you're interested in their ethics and bias, sort of things um just mention it in slack, and uh we can talk more about there. But uh this is you know.

J

The paper is um is definitely was one part of that, and I like, where it's going at listening to the biases is yeah it it's something that we could involve to be familiar with familiar with going forward. So it's good.

I

Yeah, so I have to present one more thing. It was regarding that.

I

I

I

C

You can say.

A

Intermediate here.

I

So it is known as gradient, so uh we are considering the gradient for probably if this point has to be considered, then it's horizontal and uh vector gradient should be considered.

I

I

I

C

Okay, uh what what is the so this is basically up sampling data yeah, okay, okay, uh why don't you? Can you send me a link to the slides I'd like to look them over.

G

Yeah sure yeah.

C

All right, yeah that looks pretty good. uh We I think we can find a home for that. uh You know I don't know like we'll keep working on it and we'll see. Thank you any questions about that.

C

I know krishna's audio was a little off on it, but um it's like I mean you know. I think you got the idea um so yeah, I think that's you know. uh People have next week. If people have things that they're working on they want to present, please feel free to present them put them on the list. Let's see what we have in the chat, we have a lot of different things here that people have been posting.

C

uh So susan says axolotl was a model organism, so it's a salamander xenopus is a um it's a frog, yeah and akshay says my bio is weak. That's why we have. I want to put some more uh tutorial information up on the onboarding guide, because I think people maybe need a refresher in biology. It helps um and we don't have any- I mean they're for model organisms. You can look them up by their name that you see. If you see a name of a model organism, you know you'll be able to find out more like.

C

There are a lot of uh references online wikipedia just for a very basic level, understanding of it, um but uh you know so this is we have a lot of these model organisms. We've talked about this um okay, this is uh this person does not exist. This is an example of a gan where they generate people who don't exist and they're like generate faces from you know, just existing human faces, but they're generating new faces and people don't exist uh yeah.

C

So this is uh bias that twitter exhibits, twitter crops, images to zoom into faces and it's generally more biased towards white faces. So that's a you know. It's a design feature of some of the. This is kind of what we're talking about and then nvidia uses super sampling on games. This is a link from throne, so this is a link to what they're using super sampling for so some very good points in the chat. Thank you for contributing to the chat. uh Finally, I'd like to get I'd like to turn to some papers.

C

I was hoping you had more time for this, but and I'm not going to go through too much more, but uh I wanted to go over some points that I had prepared here. uh First of all, we continuing our discussion of cellular automata. I wanted to point people to this. uh There's been some interesting work on morphogenesis in minecraft, and so this is something that's using three-dimensional cellular, automata coupled to a neural network. So it's something they call neural cellular, automata and there's a new paper here from the er on the archive on this topic.

C

uh Sebastian reese's one of the authors and it talks about how they show that neural cellular automata can learn to grow. Complex 3d objects in minecraft. So for those who are knowledgeable, what minecraft is it's a three-dimensional environment?

C

It's a virtual environment um where you can build things with blocks, so it's kind of like virtual legos and you can put them together and build like all sorts of things and you can build avatars and you can interact with each other in it in in the environment and what they're doing here is they're, taking these blocks and they're using the uh that block structure to generate 3d, artifacts and they're you're, basically evolving them using this neural, cellular, automata or they're they're forming them.

C

I don't know if they actually use an evolutionary algorithm but they're using this. This approach to build these artifacts and they're actually also to show that you can build functional, minecraft machines that regenerate when you cut them in half. So this is like what you see with some worms.

C

You see this with flatworms, this extreme ability to regenerate but other organisms. If you cut them in half, they regenerate the rest of their body. So if you cut a single uh organism that exhibits this property in half, you can end up with two separate organisms, and so you know this is an interesting advance.

C

If you want to know more, you can read the paper on that they also have.

C

I think I posted it in the slack channel, there's a competition for um open-ended evolution in minecraft, as well so they're using minecraft for all sorts of interesting things these days, and we should talk more about how to leverage that in future meetings.

C

um So, let's see, I also had another question that was raised after last week's meeting by dick gordon and we talked about some papers on the environment and it kept using the word environment and the papers kept using the word environment and he had a question in the chat and he kind of tried to clarify it after the meeting.

C

The idea is that environment we talk about in development as being this thing that, like an uh embryo, exists in or that organisms experience. But what is it? What is its relevance and how do you define it, and it might seem like a very elementary question and dick is, you know, uh he's done a lot of work in biology and development. So he knows he. You know you think he would know what the answer is, but I don't think anyone knows what the answer is.

C

Actually it's very unclear because people throw the term environment around and it's like what does it mean and so a lot of times people will throw this term around and not really define it. We're not really know what it does and so we're you know we kind of targeted two different areas.

C

The first is that environment is permissive, so environment allows for things to happen in the organism, so that may be some very general type of information like changes in the environment like from hot to cold or stresses so, like you know, starvation versus food availability or intensities, so you might have a temperature gradient where you might get really hot. You know super hot or super cold, and so those things can be transduced into the biological system. It might permit certain things to happen. It might certain things happening, but it's this idea that it's permissive.

C

It allows for things, and so, uh but the second is that environment is instructive, so this is different in the sense that environment actually has some specific information to give to the biological system, so that might include, like ratios of different signaling molecules, often times they're signaling pathways in cells or in in small microorganisms that allow for things called like quorum, sensing or other types of collective behavior, and that also, you know, involves patterns or codes that might be involved in the environment.

C

Like you know, fluctuations in the environment are interpreted in a certain there might be patterns in, in that uh sequence of events that trigger things in the biological system, so the environment is also instructive, and so this this is sort of trying to work this problem out- and I don't know if people have things you want to add to this. You can talk about it in the slack, but I think it's an interesting sort of idea to think about, um but you know that's, maybe something we can talk about later in future meetings.

C

uh I'd also like to finally kind of get into some of the papers. I don't want to get too deeply into these papers. um Jesse sent me this paper on. Nothing in evolution makes sense, except in the white of parasites, and this is uh by a couple people in who actually contribute to the artificial life, literature, susan, stepney and paulie and hogawig, and uh this is talking about sort of the role of parasites in evolution, and so uh they talk about you know.

C

The emergence of parasites and evolve in evolving systems appears to be inevitable, and so that means that parasites play this critical role in evolutionary systems.

C

So this is something they talk about spatial pattern, formation um being a factor in, like you know, or parasites, being a factor in spatial spatial pattern formation. um It also slows down replication. Sometimes it involves things like co-evolutionary, arms races, and so there are a lot of things that parasites are involved in their parasitism. More generally, which is.

A

C

One organism or one entity uh benefiting off of another entity, so there are different types of interact: interactive relationships, there's parasitism, there's uh mutualism and there are different things like that: parasitism is where you have one entity sort of living off of another entity without contributing anything, and so there's a lot of interesting stuff that this is actually.

C

I think it may have some simulation work in it. I can't seem to get the thing to scroll down there. We go.

A

C

Yeah they have a lot of different, um so they're, using this string, automatic chemistry so they're using an artificial chemistry to model some of this stuff. So it's really kind of an interesting approach to parasitism and they look at the population dynamics.

C

So you can see that they're mapping the population and number of species over time and they're looking at their. You know: they're modeling their hypotheses using this platform and here's some. uh You know some image screenshots of how they're evolving this artificial chemistry.

C

So this is what it looks like you're, just evolving agents in this artificial chemistry and you're, putting together you're getting these results um from different chemical reactions, and so I'm not really an expert at artificial chemistry by far but just to let you know that these things exist in the world and that you can do cool things with them.

C

Then there's this other paper on modeling developmental sequences, this neurodevelopmental sequences. I only bring this up because it's is something that's been out for a while, but I found it this week. They basically look at neural development in 18 mammalian species and they look at like different heterochronic changes in brain evolution.

C

So what that means is that they look at the timing of the development of the brains and all these different mammal species, and they look at what's going on so they're, looking at like how neurons form and connections grow and change over developmental time and how connectivity is refined and they have all these data that show that their chain, their timing, changes and the onset of growth and the duration of growth and other things that actually lead to a lot of the changes you see in different species, and so this is a sort of a catalog of all these changes.

C

It's a really uh complete paper- and I mean this. This actually is something we might look into anymore in the other group.

A

C

Doing some work on developmental ai- and this is something that might be very useful to that, but I think this is also useful for understanding uh development in terms of the brain and then finally, we have a couple of papers here that, uh though, a couple of things that are kind of coming out now that are really exciting. uh There's some wet lab work uh on an exuteral mouse embryogenesis model, so they actually have this exuderal mouse embryogenesis from pregastrulational weight organogenesis.

C

So this is a mouse model that is looking at formation of an embryo and then looking at differentiation within that embryo, and so this is done in a lab. This is done in in cell culture.

C

We've talked about what cell culture is um and they're looking at some of these changes that are happening in the embryo from the time when there's like this mass of cells that are starting starting to take shape to the formation of hind limbs- and so you know, you're, actually getting not only tissues but the formation of different parts of the phenotype, and so this is really interesting work.

C

um They talk ab. This is very technical overview here. I don't know if I have the uh this is actually uh something related to this. This is a news and views articles. This is easier to understand.

C

This is the first complete model of the human embryo, so in this case, they're not doing what they did with the mouse so much, but they're, actually taking a human embryo and they're actually able to grow blastocyst in a dish and so they're able to grow these human blastocyst-like structures.

C

So this is the blastocyst where it's a very simple version of of you know, what's going to happen uh in the formation of the human embryo you're going to have this inner cell mass and you have this trifecta derm layer, so you have a very, very early stage uh developmental structure here and um so studying early development in humans has been a challenge because it's hard to get, I mean their ethical considerations, but it's also hard to get these things to grow in a culture, and so you know and to really study it like.

C

We would like to that's why we use model organisms, in fact, because we can't really study these things in humans.

C

One of the reasons why we use a lot of different model organisms, but now there are two papers in nature that make key advances in this field, allowing us to like actually reprogram cells from adult tissues, to form these or to use human embryonic stem cells and to form these blastocysts in a dish, and so basically they're going to get these blastoids, which are these sort of blastocyst, aggregations or cells that behave like a human blastocyst and so you're able to generate what the advantage is here.

C

And so you can generate these without having to uh deal with a lot of the ethical constraints or with a lot of the you know, the consequences of trying to get a human embryo to grow into a blastocyst outside of the womb, and so they we've talked about. uh You know different sorts of organoids before organoids gastroloids, and this is a blastoid.

C

So this is uh really exciting work and you know they just grow it in this chamber and they let the cells aggregate and do this pattern form basic pattern formation and they want to study what happens during this process. And so this is a good. This is a new some more advances in this model, so I don't have the full papers are listed in the references of this so often what they do is they have this overview paper and they cite the original papers in this.

C

But it's a nice accessible way to understand it, and then this paper is a little bit deeper into the details of modeling embryogenesis in a dish but they're doing this in mouse, and so these are a number of different models in mammalian systems which are much different than our c elegans and our other models that we're doing in our group, but they're, very important to understand. If you want to understand development, so we have some questions and I'll take those before we go um so susan says I have some seashells.

C

Should I take some 3d images of them? Yes, please, if you can I'd love to see those mine acts as axolotls can regenerate. Yes, maxolotls are a key model organism and regeneration studies. If you go to the regeneration, literature you'll find these axolotls. A lot uh uh says environment is adaptive, which is true, um so you know we're trying to figure out what the role of environment is, but environment itself is an adaptive.

C

It drives adaptation in the organism, so I mean we're trying to piece this together. You know the language is very uh broad in general and we want to kind of get this together. Can you post the link to the parasite paper yeah I'll post it in the slack channel? um That would be a good thing for people to read. I think um dicks, oh dicks, here hello, dick anyone willing to learn a note basic. So if you want to get involved with endnote, we want to put these papers in a bibliography of some type.

C

uh Surety says: was the issue resolved in the code: axolotl's development, constant temperature, uh oxygen and distilled water? What does environment do, and so this is the thing uh about the the question about the environment. Funding was cut for human embryos decades ago, yeah.

A

C

They have don't really fund that area, so surety collected 100 papers for the paper. I'm writing. Please be patient and hexabattles, regenerate spinal cord and heart. Yes, axolotls are very good model organism for looking at a lot of things in in in organogenesis as well. um So so thank you for meeting. I know we're probably at the end of our time here um and I just want to say uh if you know we'll be going over some of the stuff on slack, we should have some more discussions on slack or via email.

C

I send out the weekly email list and so, if you're interested in submitting a paper to present during the meeting or if you have something to present, please bring it to the meeting or send it to me in advance, and we can do that any other questions before we go any other.

C

C

Okay, our ancestors look like axolotls. Why can't we regenerate limbs? I think, that's a major question: the regenerative medicine community. I know that, like I've worked with people who are, I worked with one person who did like work on newts and other thing.

C

You know in other neural systems, and you know it's still hard for people to understand like why we can't regenerate like you know, lizards and amphibians, have some and fishes as well have some uh quite uh impressive abilities in terms of regeneration, but we don't have that, and there are a lot of theories as to why, but no one's really, I think, really uncovered the key factors and why that is.

C

I mean it's going to deserve like a a presentation to get into the reasons why I don't wanna.

C

So jesse a note basic is free capacity for fifty thousand records. So that's so there are no more questions. uh Thank you for meeting um have a good week.