Yaki Setty on synthetic biology and agent-based modeling

  • cover play_arrow

    PLAY EPISODE


Season 2014
Season 2014
Description arrow_drop_down

Description

Can you grow an organ inside a computer , and would it teach you something biology alone cannot? Computational biologist Yaki Setty describes how agent-based models of stem cell development can reconstruct the pancreas, the C. elegans gonad, and even parts of the brain from first principles, revealing emergent properties that no single experiment could predict. Subscribe for more from the Convergent Science Network podcast series. Yaki Setty joins Paul Verschure and Tony Prescott at the BCBT summer school to present his approach to synthetic organ development using autonomous agent-based modeling. Each cell in the simulation is defined by biologically justified state diagrams , covering differentiation, proliferation, movement, and environmental sensing , with every parameter traceable to published experimental data. The environment is modeled as a three-dimensional grid of voxels containing chemical gradients governed by differential equations, and cells interact with this environment and with each other to produce emergent organ structures. The discussion walks through three applications of increasing complexity. The pancreas model, with over 150 cell states, reproduces the characteristic cauliflower-like morphology of pancreatic tissue and demonstrates how blood vessel scaffolding guides cell aggregation. The C. elegans gonad model achieves quantitative predictions about cell numbers, zone lengths, and cell cycle ratios with far fewer states, validated against experimental measurements within weeks rather than years. The conversation also touches on extending these methods to neural development, where the same platform and principles apply but the complexity of cell types and connectivity presents new challenges. Key topics include how autonomous agent models differ from conventional computational approaches, why all available biological data should be incorporated rather than held back for testing, how mutations serve as the primary validation strategy for these models, what the relationship is between stem cell stemness and differentiation potential, why morphological benchmarks like cauliflower structure are difficult to quantify rigorously, and how these simulations could eventually model disease processes by tracing developmental history back to its origins. Part of the Convergent Science Network podcast series from the BCBT Summer School.

Tagged as:

About the author call_made

CSN Podcasts

Both the triumphs of humanity and its most evil deeds have resulted from collaboration. In a time where humanity is required to aspire to the former and minimize the latter, the question arises of how collaboration arises and why it fails. Surprisingly, this phenomenon, so central to who we are, is not well understood. Hence, a collaborative effort is required to understand collaboration in its full biological, psychological, sociological, cultural, and economic complexity and to translate this understanding into operational impact. This series of podcasts is one step toward achieving these complementary goals. The Collaboration Podcast presents interviews with people who are central orchestrators of collaboration in various domains including business, government, science, art, health, sustainability, and the military. The discussions were conducted by Prof. Dr. Paul F.M.J. Verschure and members of the Program Advisory Committee of the Ernst Strungmann Forum on Collaboration (https://www.esforum.de/forums/ESF32_Collaboration.html) during 2021 and had the goal to sketch a map of opportunities, challenges, and obstacles in human collaboration. The forum took place in May 2022, and now we would like to share this series of interviews with a broader audience. The full report of the Forum will be published in 2023 by MIT Press. The podcast was produced by the Convergent Science Network (https://www.convergentsciencenetwork.org/). Context: The stability of social systems depends critically on realizing sustainable methods of “collaboration,” yet how and by which means collaboration is achieved is not clearly understood; neither are the conditions or processes that lead to its breakdown or failure. Collaboration can be understood as cooperation between agents toward mutually constructed goals. Part of the reason for our lack of understanding is that the phenomenon of collaboration is, by nature, a highly multidisciplinary problem, and effective research into its complexities has been difficult to achieve across the broad range of scientific and technical disciplines involved. The need for a fundamental understanding of collaboration, however, has become increasingly important. Not only does humankind demand answers as it attempts to address critical challenges at multiple scales (e.g., climate change, migration, enhanced automation, social and economic inequality), but ever-increasing technological and economic means of interconnecting people and societies are disrupting long-established, familiar patterns of how we interact. Radical technological changes that are ongoing have the potential to reshape collaboration in ways that are currently hard to predict or influence (e.g., by altering configurations in interaction, information creation, and modes of communication). On one hand, such changes could disrupt hitherto stable forms of collaboration by affecting critical communication channels and traditional roles, as can be observed in the rapidly changing patterns in governance, commerce, and social interaction. Conversely, technology could lead to the emergence of novel, successful forms of collaboration that deviate from traditional “hierarchical” architectures. Evidence of this can be seen in areas as diverse as highly automated manufacturing plants, the open science movement, collaborative software repositories, user-centered services, and the sharing of economy-based modes of organization. Without a fundamental understanding of the mechanisms, processes, and boundary conditions of collaboration, it is not possible to evaluate or predict which of these possible scenarios are sustainable or even plausible. The Forum “How Collaboration Arises and Why it Fails” (May 8–13, 2022, Location: Frankfurt am Main, Germany) Chairs: Andreas Roepstorff and Paul Verschure Program Advisory Committee: Jenna Bednar, Julia R. Lupp, Bhavani R. Rao , Andreas Roepstorff, Ferdinand von Siemens, and Paul Verschure

More posts

Timestamp

  • fast_forward00:00:03 - This is the Convergent Science Network podcast.
  • fast_forward00:00:08 - Leading researchers in the domain of neuroscience, brain theory and technology
  • fast_forward00:00:12 - are interviewed by Paul Vershoor and Tony Prescott.
  • fast_forward00:00:17 - Ready to roll? Okay, this is Paul Vershoor with the Convergent Science Network
  • fast_forward00:00:21 - together with Tony Prescott.
  • fast_forward00:00:24 - And our guest today is Yaki Setti, who was speaking at the BCBT Summer School
  • fast_forward00:00:29 - on actually synthetically building different kinds of organs.
  • fast_forward00:00:36 - So, Jaki, how do we build a synthetic organ?
  • fast_forward00:00:41 - Well, first we have to know the biology. We have to know what God created or
  • fast_forward00:00:46 - what biology created to rebuild it.
  • fast_forward00:00:50 - We need to know the biology better than the biologist and better than the biology
  • fast_forward00:00:54 - knows itself because we need to reconstruct it. So in a way,
  • fast_forward00:00:58 - I don't want to sound too sophisticated, but we're playing kind of creator.
  • fast_forward00:01:05 - We create something. We use the computer as a platform to play with the data,
  • fast_forward00:01:09 - put the data into the computer and play with it.
  • fast_forward00:01:13 - And we know what the desired output is.
  • fast_forward00:01:17 - We don't know all about it, but we know if something is missing,
  • fast_forward00:01:20 - we will see that the output will be wrong.
  • fast_forward00:01:24 - So that's the... But no, this doesn't just happen by accident, right?
  • fast_forward00:01:29 - It's not that you just have, let's say, a random set of ingredients like some
  • fast_forward00:01:33 - big minestrone and now that comes an organ like a pancreas.
  • fast_forward00:01:36 - So the first example you discussed was to construct, to really grow in silico a pancreas.
  • fast_forward00:01:44 - And then you showed to us that that model of the developed pancreas shared many
  • fast_forward00:01:50 - features with the biological pancreas. So the pancreas is actually built from
  • fast_forward00:01:56 - different kinds of cells, right?
  • fast_forward00:01:58 - And the morphology of the structure will change over time in some developmental process.
  • fast_forward00:02:02 - So how do we exactly model that? How do you model such a cell and how do these
  • fast_forward00:02:08 - cells then collectively form something that you could call a pancreas?
  • fast_forward00:02:11 - So in each cell, you put the biological data from the papers.
  • fast_forward00:02:15 - So you have a block diagram.
  • fast_forward00:02:18 - Which blocks and arrows, and each arrow is a transit between the two states of the block.
  • fast_forward00:02:26 - So you take the cell and you define the state it may be in, and this is correlated to the biology.
  • fast_forward00:02:32 - You have to justify every block and every arrow you put in this diagram.
  • fast_forward00:02:39 - So for your cell model, how many states can the cell attain?
  • fast_forward00:02:41 - Well, it depends which model it is. The pancreas model, it's the most complicated,
  • fast_forward00:02:46 - and it had over 150 states all over.
  • fast_forward00:02:52 - The number of the differentiation state was, I believe, like 10,
  • fast_forward00:02:59 - which is more or less the number of markers that the biologists found.
  • fast_forward00:03:06 - So for each state or stage, the biologists have defined we have a state in our
  • fast_forward00:03:13 - differentiation component.
  • fast_forward00:03:15 - And it's the same for other components in the system.
  • fast_forward00:03:19 - For example, the proliferation component has five states, which it is correlated
  • fast_forward00:03:24 - with the five stages of cell cycle.
  • fast_forward00:03:26 - So we are trying to put into the model and
  • fast_forward00:03:30 - to formulate as many biological data as possible
  • fast_forward00:03:33 - and to be able to
  • fast_forward00:03:37 - justify everything for example if we put an
  • fast_forward00:03:39 - assumption or an error or a block in the diagram
  • fast_forward00:03:42 - so each one of that have this we
  • fast_forward00:03:45 - have to we are able to say okay this data was taken from this paper and that's
  • fast_forward00:03:51 - why we constrain the model to be biologically plausible or accurate or at least
  • fast_forward00:03:57 - justified right but then so we have the cell the cell can have a large number of states
  • fast_forward00:04:03 - then there are certain inputs to that cell that
  • fast_forward00:04:06 - will make it actually decide which state it attains now
  • fast_forward00:04:09 - that transformation of inputs to the
  • fast_forward00:04:12 - state expressed that's more like a logical lookup table like if my inputs have
  • fast_forward00:04:18 - the following characteristics then in a deterministic fashion i turn into state
  • fast_forward00:04:22 - one or is that transfer function following more the kind of messenger systems
  • fast_forward00:04:28 - that you would find within a cell.
  • fast_forward00:04:30 - Yeah, it is temporal decision.
  • fast_forward00:04:33 - It's based on time and space.
  • fast_forward00:04:35 - You will get some kind of message at a certain point if you are exactly at the
  • fast_forward00:04:42 - same certain position and has the same specific message.
  • fast_forward00:04:47 - So it is deterministic in that kind of perspective if you look at the individual cell. all.
  • fast_forward00:04:55 - But if you look at the population level, you'll see that there are a lot of different groups.
  • fast_forward00:05:01 - A lot of stochastic decisions are made. For example, if cell number one made
  • fast_forward00:05:06 - a decision at one point and cell number two made a decision at another point,
  • fast_forward00:05:11 - and they switch roles, so your simulation will be a little bit different.
  • fast_forward00:05:15 - It is equivalent to embryos, to human beings.
  • fast_forward00:05:20 - All of us have two hands, have two eyes, have one head, two legs,
  • fast_forward00:05:24 - but still there is a lot of variety, and part of it is the genetic issue that
  • fast_forward00:05:32 - we take with us, and part of it is part of our development.
  • fast_forward00:05:35 - So this is why two brothers are not the same, and even identical twins are not identical.
  • fast_forward00:05:42 - So, I mean, your model of a cell is a kind of informational model where you
  • fast_forward00:05:46 - describe the states it can be in, the kind of switches within it,
  • fast_forward00:05:51 - whether the switches can be thrown back, and so on.
  • fast_forward00:05:54 - And it's sensitive both to external, I guess, chemicals that can affect the
  • fast_forward00:06:02 - membrane and therefore change the internal state, but also internal mechanisms
  • fast_forward00:06:06 - that can throw switches. Does that summarize it?
  • fast_forward00:06:09 - Yes, that's more or less it. The cell itself, we look at this as a three component,
  • fast_forward00:06:13 - the component of the cell and a component of external sensors,
  • fast_forward00:06:17 - which we call the membrane or the receptors.
  • fast_forward00:06:19 - And there are the receptors and there are the factor, it's not only the genetic
  • fast_forward00:06:24 - ones it's not only genes, it can be genes in different states.
  • fast_forward00:06:27 - That would summarize it nicely.
  • fast_forward00:06:29 - And the specific thing that is exciting about this of course is that you can
  • fast_forward00:06:34 - model stem cells which have the exciting property that they can turn into other kinds of cells.
  • fast_forward00:06:42 - Yeah, they can differentiate. So perhaps you could because this audience of
  • fast_forward00:06:48 - BCBT is mainly interested in neuron cells.
  • fast_forward00:06:52 - Perhaps you could fill us in a bit more about the different kinds of stem cells
  • fast_forward00:06:55 - and what we know about them. Well, it depends.
  • fast_forward00:06:59 - Okay, stem cells in...
  • fast_forward00:07:02 - The core of stem cell research is cells that can differentiate and proliferate
  • fast_forward00:07:06 - and carry their own data.
  • fast_forward00:07:10 - But lately, a lot of research was done and they claimed that stem cells,
  • fast_forward00:07:16 - this naïve definition is not enough.
  • fast_forward00:07:19 - And there is a notion of stemsness.
  • fast_forward00:07:23 - It's kind of how much stem this cell is.
  • fast_forward00:07:27 - And it means that even in a growth tree, if it can differentiate to three different cells.
  • fast_forward00:07:37 - Or can get any kind of cell like the embryo cell like the Zygote,
  • fast_forward00:07:43 - The first cell is like the original cell,
  • fast_forward00:07:46 - and this is more or less the new notion of stem cells, but the original one
  • fast_forward00:07:52 - is a cell population that can maintain itself and differentiate.
  • fast_forward00:08:00 - And in almost any organ, you can find stem cells.
  • fast_forward00:08:03 - And even you can, I understand that you can take cells which aren't stem cells,
  • fast_forward00:08:08 - and by stressing them, I think I heard this about blood cells,
  • fast_forward00:08:11 - was it? You could maybe turn them back into stem cells.
  • fast_forward00:08:13 - Well, I don't like the term turn them back.
  • fast_forward00:08:15 - You can turn them into new stage. You can turn the stemness mechanism of the cells on.
  • fast_forward00:08:23 - They start to behave like a stem cell. Right. And so the behaviors of the cells
  • fast_forward00:08:27 - in your model, can you just summarize the things that they can do?
  • fast_forward00:08:29 - They can do the two most important mechanisms of differentiating.
  • fast_forward00:08:38 - They can go from one state to a variety of states. There is kind of a tree of
  • fast_forward00:08:44 - decision that the cell can make.
  • fast_forward00:08:46 - And in parallel, orthogonal component would be to proliferate.
  • fast_forward00:08:51 - They can create new instance of the same cell. And this is enough to maintain a stem cell population.
  • fast_forward00:09:00 - But in our model there is an additional requirement
  • fast_forward00:09:04 - and this is their position in space they can move to another place this is how
  • fast_forward00:09:10 - we can create structure of organs and we can create different population and
  • fast_forward00:09:17 - as we have a front end we visualize the model you can see it,
  • fast_forward00:09:23 - the user can view it and can interact with it sometimes you can even say Say,
  • fast_forward00:09:27 - okay, I want cell number five,
  • fast_forward00:09:30 - six, and seven to be erased from the simulation and see what's going on.
  • fast_forward00:09:34 - Or to change the environmental effect at runtime.
  • fast_forward00:09:39 - So at the larval stage or when the model is young, it's not an adult yet,
  • fast_forward00:09:45 - I want to keep the environment as it should be.
  • fast_forward00:09:50 - But at some point, I want to decide that, okay, after three days,
  • fast_forward00:09:54 - I want to change the environment. I want to inject a new material.
  • fast_forward00:09:57 - Or you want to kill 10% of the cell and then to see what the simulation suggests the result will be.
  • fast_forward00:10:05 - Right. So your model consists of a number of initial cells, maybe even just
  • fast_forward00:10:10 - one, and then an environment in which the cell lives and it's communicating with the environment.
  • fast_forward00:10:15 - And with each other. And with each other, yeah. So in terms of the movement
  • fast_forward00:10:20 - of the cell, it's able to do things like move up gradients, chemical gradients.
  • fast_forward00:10:25 - Yeah, it senses the gradients in your environment and can move towards it.
  • fast_forward00:10:29 - So can you just summarize what the environment looks like from the point of
  • fast_forward00:10:32 - view of simulation? So you have chemical gradients in it, for instance.
  • fast_forward00:10:35 - Do you have physical structures beyond that? Yeah, okay. The environment is
  • fast_forward00:10:38 - divided into voxels, three-dimensional pixels, boxes in the space.
  • fast_forward00:10:43 - And inside there is a gradient of chemicals that exist in each part.
  • fast_forward00:10:49 - In many cases, the decision what are the values is based on the differential
  • fast_forward00:10:55 - equations that determine it.
  • fast_forward00:10:58 - Determine the value and it changes over time.
  • fast_forward00:11:02 - So the environment itself is a model of its own.
  • fast_forward00:11:06 - Then in each one of these voxels, there is a cell or a sphere that senses what
  • fast_forward00:11:12 - are the chemicals in the atmosphere or in the surrounding environment and act based on that.
  • fast_forward00:11:18 - This concept is inspired by... it's a rather old concept that didn't get a lot
  • fast_forward00:11:25 - of attention in science. It's called autonomous agent.
  • fast_forward00:11:29 - Autonomous agent is taking out of artificial intelligence. During the 60s,
  • fast_forward00:11:34 - they defined this concept as a machine, an object, or agent that interact with this,
  • fast_forward00:11:42 - environment and decide on its next move based on its current state and the environment.
  • fast_forward00:11:48 - So you consider the whole model to be an agent-based model? It's agent-based
  • fast_forward00:11:52 - and it's autonomous agent-based model as adapted to biology.
  • fast_forward00:11:57 - This This is like the computer science view of the models, yeah.
  • fast_forward00:12:03 - So the first model you described was to replicate the pancreas.
  • fast_forward00:12:10 - And historically, that was your first model? Yeah, that was the first model.
  • fast_forward00:12:14 - And what was the specific sort of scientific question that led you and the group
  • fast_forward00:12:18 - you were in to think that the pancreas was a good place to go for this?
  • fast_forward00:12:22 - Well, I wanted to study about the diabetes.
  • fast_forward00:12:25 - And I started research about the diabetes. and then I got to the organ that
  • fast_forward00:12:29 - is controlling the diabetes.
  • fast_forward00:12:31 - And then my view was that you can't study the organ without knowing the process
  • fast_forward00:12:38 - that brought him to this state,
  • fast_forward00:12:42 - and you need to understand the structure and how it was formed.
  • fast_forward00:12:46 - So you need to start by the first day where the organ develops in order to understand
  • fast_forward00:12:52 - what went wrong by the end.
  • fast_forward00:12:54 - And if you notice during the lecture, I haven't even talked about the diabetes
  • fast_forward00:12:59 - because we didn't get this far.
  • fast_forward00:13:02 - There were a lot of open questions, a lot of interesting questions that just emerge as we go.
  • fast_forward00:13:10 - But I envision that models like that will cover the whole process,
  • fast_forward00:13:13 - including the functionality in the end.
  • fast_forward00:13:15 - And then you can study diseases and everything and you can track all the way
  • fast_forward00:13:21 - back to the history and see if something went wrong on day zero.
  • fast_forward00:13:27 - And affected what we see on day 100.
  • fast_forward00:13:31 - And I guess many of the diseases that you're interested in have a genetic component,
  • fast_forward00:13:36 - which perhaps you'll then be able to recreate in the model once the model is
  • fast_forward00:13:40 - sufficiently complete.
  • fast_forward00:13:41 - Yes, but it's not only the genetic background.
  • fast_forward00:13:45 - It's like the thing that the organ experienced.
  • fast_forward00:13:50 - And cause this disease. How far does the model have to develop before you'll
  • fast_forward00:13:56 - be able to ask these kinds of questions with it? It still has like...
  • fast_forward00:14:00 - Right now it covers all the developmental birth. This is 15 days.
  • fast_forward00:14:08 - And we need to add expression of the factors, of the hormones to that,
  • fast_forward00:14:16 - and then we can investigate it. I believe in a few years' time.
  • fast_forward00:14:21 - So you're close to having something that would be useful in a disease model.
  • fast_forward00:14:26 - Yeah, but we are close to having kind of a complete model, but it doesn't necessarily
  • fast_forward00:14:33 - mean that we know everything we need.
  • fast_forward00:14:35 - Because once you finish modeling the organ and added all the functionality,
  • fast_forward00:14:41 - you still need to verify that it is consistent with biology.
  • fast_forward00:14:44 - So at each stage when you're developing the model, you're using some data from
  • fast_forward00:14:50 - the biology to test the functionality.
  • fast_forward00:14:53 - For instance, I think you were showing with the pancreas, you were showing what
  • fast_forward00:14:57 - happens with the interaction with blood vessels and how the model produced similar
  • fast_forward00:15:02 - outputs to what people observed in experiments.
  • fast_forward00:15:06 - And that methodology whereby you take these biological data and use them to
  • fast_forward00:15:13 - test the validity of the model.
  • fast_forward00:15:16 - Do you consider that the methodology that you have there is finalized, fully refined?
  • fast_forward00:15:24 - How do you know when you've done enough tests? Or is it open-ended, I guess?
  • fast_forward00:15:30 - Yeah, you never have enough tests because the data keeps streaming into your desktop.
  • fast_forward00:15:35 - So there are more papers published and there is new data and you have to refine
  • fast_forward00:15:39 - the model as much as you need.
  • fast_forward00:15:41 - Okay, I guess there's another way of putting that. What's the minimal set of
  • fast_forward00:15:45 - biological constraints that you feel you need in order to say that the model is useful?
  • fast_forward00:15:52 - If you manage to reproduce the biology and a few mutations, so it's not enough
  • fast_forward00:15:59 - to have the biology reproduced.
  • fast_forward00:16:01 - You still need to show that you agree with the known data. And then you can use it for prediction.
  • fast_forward00:16:09 - I guess I'm thinking sort of when people are building a machine learning system
  • fast_forward00:16:15 - for instance, something that.
  • fast_forward00:16:18 - Learn to recognize spoken language and create written language,
  • fast_forward00:16:24 - then they might take a data set and use half of it for training, half of it for testing.
  • fast_forward00:16:29 - Do you use those kind of methodologies of deliberately leaving out some of the data?
  • fast_forward00:16:36 - No, I would include all the data and then start testing.
  • fast_forward00:16:41 - But is there a risk then that your model isn't sufficiently...
  • fast_forward00:16:46 - Well, you can build the model to fit those data, but then you don't have enough
  • fast_forward00:16:52 - test points to really check whether the model is...
  • fast_forward00:16:56 - I believe that you should get all the available data into your model,
  • fast_forward00:17:00 - then start taking the next step and see if you can get more data once you have the model.
  • fast_forward00:17:08 - Would that be because you need what data there is to constrain the model and
  • fast_forward00:17:14 - build it? I think that the difference is that I'm working with biology.
  • fast_forward00:17:18 - Leaving out some of the biological data makes no sense.
  • fast_forward00:17:22 - If you know an evidence on the system, you must agree with it.
  • fast_forward00:17:26 - What can be done is that experiments that were done in the past and you are
  • fast_forward00:17:34 - not aware of or you prefer to test to keep them as a test case, that can be done.
  • fast_forward00:17:41 - But for the wild-type simulation, you can't leave data out.
  • fast_forward00:17:49 - So the data set wouldn't be sufficiently rich for this kind of,
  • fast_forward00:17:53 - to have two sets of data, one which you used to build the model,
  • fast_forward00:17:56 - one which you could use to tell.
  • fast_forward00:17:57 - In analogy, we can say that you take the wild-type data, the type of the normal
  • fast_forward00:18:04 - growth, normal development, and you test the mutation.
  • fast_forward00:18:09 - So that's a separation you can do
  • fast_forward00:18:11 - that's what you're doing with the models that's what I'm doing with the models
  • fast_forward00:18:14 - but you can take the existing data and say roughly this I keep out and this
  • fast_forward00:18:20 - I keep in you have to keep something in mind before you do it and I think this
  • fast_forward00:18:25 - separation of mutation and real data can be can be equivalent to what you described,
  • fast_forward00:18:32 - so to come back a bit to the cell model that we're using.
  • fast_forward00:18:37 - And naively you might think about that you might model a
  • fast_forward00:18:40 - cell as let's say an autonomous entity with a
  • fast_forward00:18:43 - membrane that sort of is is is motile it moves through some substrate it might
  • fast_forward00:18:48 - have interactions with other cells adhere to it and so on but in the simulation
  • fast_forward00:18:53 - that's not really what a cell is right in the simulation the cells in the end
  • fast_forward00:18:58 - defined operationally as the state of.
  • fast_forward00:19:03 - A little volume of space, right? So you take the whole volume that you want
  • fast_forward00:19:08 - to simulate and you sort of cut it up in a lot of little cubes and then what
  • fast_forward00:19:15 - you call a cell is the state of each of those cubes.
  • fast_forward00:19:18 - Well, it depends how you... No, no, I don't like this way of looking.
  • fast_forward00:19:22 - I disagree with it because there are a lot of empty cubes.
  • fast_forward00:19:27 - What type of cell is that? Is that the null cell?
  • fast_forward00:19:31 - You can tell me, right? No, I don't believe in the null cell.
  • fast_forward00:19:35 - Cell is an existing entity.
  • fast_forward00:19:37 - A null cell or a null object can be a mathematically defined object,
  • fast_forward00:19:45 - but it can't be a biological defined object.
  • fast_forward00:19:47 - Sure, but in terms of the caricature I gave of the technical approach,
  • fast_forward00:19:53 - the implementation is, I think, reasonably accurate.
  • fast_forward00:19:57 - Yes. In terms of implementation. The implementation, you can look at it that
  • fast_forward00:20:00 - way, but you'll have to add the temporal component into it. Of course.
  • fast_forward00:20:08 - So it's kind of, you can think of it as a kind of a three-dimensional space
  • fast_forward00:20:13 - or two-dimensional space divided into cubes or pixels and it changes over time.
  • fast_forward00:20:19 - So it's like a three-dimensional cellular automaton.
  • fast_forward00:20:23 - Yes. You can think about it as a collection of three-dimensional cellular automaton.
  • fast_forward00:20:27 - Right. But we have chosen to separate between the environment and the cell.
  • fast_forward00:20:34 - And this is because this is how it acts in biology.
  • fast_forward00:20:37 - So it's hard to me to, okay, it's like saying you're not a human being.
  • fast_forward00:20:43 - You are kind of a pixel and you are just moving from one place to another.
  • fast_forward00:20:49 - But no, that's not the truth.
  • fast_forward00:20:50 - That's not the case. If you're doing something, if we are talking or something,
  • fast_forward00:20:54 - there is something behind this action.
  • fast_forward00:20:58 - We're not just talking, we're having conversation. You say something, I reply.
  • fast_forward00:21:03 - So to think about it just as movement and things in the space,
  • fast_forward00:21:09 - I think it will be simplification of reality.
  • fast_forward00:21:13 - But yes, if you prefer to see the technology challenge or the implementation.
  • fast_forward00:21:19 - Just to define what we're doing, right?
  • fast_forward00:21:21 - Because we're interpreting the outcome of an algorithm, right?
  • fast_forward00:21:25 - And the algorithm is just, if you want, deciding what state a certain voxel
  • fast_forward00:21:31 - in that simulation should take, a certain little bit of this three-dimensional volume.
  • fast_forward00:21:37 - And then with that, you model these systems, these organs, or C.
  • fast_forward00:21:42 - Elegans in the brain. These are the three examples we looked at.
  • fast_forward00:21:45 - So then the question becomes, okay, when is then such a simulation,
  • fast_forward00:21:49 - let's say, defendable? When is it plausible?
  • fast_forward00:21:53 - When is it not more our interpretation?
  • fast_forward00:21:58 - As opposed to what's really going on in the simulation, right?
  • fast_forward00:22:01 - So for the pancreas, you showed us that we have to look at a system that essentially
  • fast_forward00:22:07 - at the stage in development, your modeling expresses three cell types, right?
  • fast_forward00:22:11 - And these three cell types cluster in very specific forms, no?
  • fast_forward00:22:15 - And then what you want to recover in your simulation is, again,
  • fast_forward00:22:18 - the generation of these three cell types and this kind of clustering.
  • fast_forward00:22:21 - Is that correct? That's correct.
  • fast_forward00:22:23 - Okay. And then I could argue, well, and then the structuring of this pancreas
  • fast_forward00:22:28 - was also very much predicated of the role as you told us on the role of the
  • fast_forward00:22:33 - blood vessels right so the blood vessels providing some sort of scaffold,
  • fast_forward00:22:38 - in which that very much guides the coagulation of the cells yeah,
  • fast_forward00:22:44 - so but then but this is a well-known biological fact it's not something that
  • fast_forward00:22:49 - we sure but you could still argue the biologists were not really able to tell
  • fast_forward00:22:54 - you yet how this then happened in all its details, right?
  • fast_forward00:22:57 - How sort of, let's say, specific parameters might affect this.
  • fast_forward00:23:01 - But then you could, as we saw earlier, right? Your cells, as you said,
  • fast_forward00:23:04 - have dozens, hundreds of states.
  • fast_forward00:23:09 - Large populations of them give rise to specific structures.
  • fast_forward00:23:14 - But what we try to explain are the three cell types of the pancreas forming,
  • fast_forward00:23:18 - okay, like cauliflower-type structures. So, I could say, well,
  • fast_forward00:23:23 - this is a super powerful model.
  • fast_forward00:23:25 - You have so many parameters you can play with, this can never go wrong.
  • fast_forward00:23:29 - You should always hit the jackpot with that, right? Yes, but all the parameters
  • fast_forward00:23:35 - in the model are constrained by biology.
  • fast_forward00:23:38 - So I believe that what is in the model, it's in the biology.
  • fast_forward00:23:42 - So maybe the biology is a superpower model.
  • fast_forward00:23:45 - You can think about it. It's right because you can find the pathways in cellular
  • fast_forward00:23:50 - decision that you can bypass.
  • fast_forward00:23:52 - You have a few pathways that leads to the same result.
  • fast_forward00:23:57 - So let's ignore the model and think about the biology.
  • fast_forward00:24:02 - Why the biology of two paths that goes to the same destination.
  • fast_forward00:24:06 - So another way of validating the model is to create variants of it which don't
  • fast_forward00:24:12 - fit the biological data.
  • fast_forward00:24:13 - So just take the one which you think is accurate and flip a few of the arrows
  • fast_forward00:24:18 - or whatever and see if they do anything similar. And that would tell you how.
  • fast_forward00:24:24 - How valid it is to say that that specific model is needed to generate the data that you see.
  • fast_forward00:24:31 - Now, in some sense, that's what you're doing when you get your mutations,
  • fast_forward00:24:34 - but then you're saying this is an actual mutation that happens in nature,
  • fast_forward00:24:38 - and I can see what happens in the model.
  • fast_forward00:24:40 - But have you done experiments where you just flip components or leave components
  • fast_forward00:24:45 - out and see how that works? Yes, of course.
  • fast_forward00:24:48 - One of the first in-silico experiments we did was just to take the nucleus,
  • fast_forward00:24:55 - the inner decision of the cell, and just make a big spaghetti,
  • fast_forward00:25:01 - like Erdos way of testing, taking the arrows and connect them randomly to other
  • fast_forward00:25:08 - states and to see what's going on, like the craziest stuff.
  • fast_forward00:25:12 - It's just like taking the DNA and just take the operon and point it to express
  • fast_forward00:25:22 - a different gene, just like a crazy scientist did.
  • fast_forward00:25:26 - And yes, this disabled everything in the simulation, but at least it's proven
  • fast_forward00:25:32 - that that data is really necessary for the modeling.
  • fast_forward00:25:36 - But what I'm trying to say, and maybe I'm not clear enough, we are trying that
  • fast_forward00:25:42 - the models will be realistic as possible and to take off a pathway or take off
  • fast_forward00:25:48 - genes just because we don't understand what they do,
  • fast_forward00:25:50 - or there are other genes that do the same, that does the same.
  • fast_forward00:25:55 - I wouldn't recommend that, because then you might miss point.
  • fast_forward00:26:00 - In the model. The idea is to put
  • fast_forward00:26:02 - into the model as many biological data as possible, not the minimal one.
  • fast_forward00:26:07 - Well, there are two issues for me on this, right? On the one hand,
  • fast_forward00:26:10 - how do you really quantify a cauliflower, right?
  • fast_forward00:26:14 - So in some sense, in the presentation, it was a bit more sort of by eye,
  • fast_forward00:26:19 - like, well, so it looks sort of similar, right?
  • fast_forward00:26:22 - And I think we would all agree that we can do better than that, right?
  • fast_forward00:26:26 - And also on top of that, that cauliflowers have fractal-like structures,
  • fast_forward00:26:29 - which raises interesting questions about development, right, and morphogenesis.
  • fast_forward00:26:34 - So how then do you really quantify that similarity between the structure that
  • fast_forward00:26:40 - you see in your simulation and that you find in biology? It's not easy.
  • fast_forward00:26:46 - Life is hard. Life is hard, yeah. And one of the hurdles in this issue is that
  • fast_forward00:26:52 - there are not that many data, biological data.
  • fast_forward00:26:56 - It's not that I can ask you, ask a biologist, please give me how many cells
  • fast_forward00:27:00 - there are in this cauliflower and what are the fractals.
  • fast_forward00:27:03 - This will be a very naive way to see the experiments.
  • fast_forward00:27:08 - Usually, that's what we get. That's the data. No, but still,
  • fast_forward00:27:11 - in the literature, you will find these characterized.
  • fast_forward00:27:14 - There will be images. There will be microscopic analysis, electromicroscopy.
  • fast_forward00:27:20 - There will be different levels of analysis. You will hardly find it in mice.
  • fast_forward00:27:23 - You will find it in, I think, I didn't come across too many.
  • fast_forward00:27:27 - Or you might have the inter-cauliflower spacing.
  • fast_forward00:27:31 - Yes. What I did, for example, is doing some of image analysis,
  • fast_forward00:27:37 - trying to compare not exactly the population, but to split between the color
  • fast_forward00:27:44 - of the domains of the color.
  • fast_forward00:27:47 - Or how many pixels carry red color and how many green colored pixels there are
  • fast_forward00:27:54 - in histology picture and to compare it with a cell count in the simulation.
  • fast_forward00:27:59 - So it's kind of making... Okay, but you're saying you couldn't really do that
  • fast_forward00:28:04 - yet because the data wasn't there. The data is not there.
  • fast_forward00:28:07 - The data is very dirty in this... Okay, but which parameter in your model would,
  • fast_forward00:28:13 - for instance, control the inter-cauliflower spacing?
  • fast_forward00:28:16 - There is no single parameter. There are a few parameters and all related to the biology.
  • fast_forward00:28:21 - For example, the flow of the gradient of the blood vessels secreting the factors in the environment.
  • fast_forward00:28:35 - That would be a very important parameter. And indeed, when we played with this
  • fast_forward00:28:40 - parameter, we see that we generated a lot of different structures that may and may not be correct.
  • fast_forward00:28:49 - But this is something to be tested in the future when we can control the blood
  • fast_forward00:28:53 - vessel, the way the blood vessel secretes factors to the environment. Right.
  • fast_forward00:28:59 - So then after the pancreas, we looked at the C. elegans, right?
  • fast_forward00:29:05 - And there in particular, you looked at the hermaphrodites who are basically producing other C.
  • fast_forward00:29:12 - Elegans by cell fertilization.
  • fast_forward00:29:16 - And that C. elegans, the way you described it, just works a bit like a sausage machine, right?
  • fast_forward00:29:20 - So you have sort of stem cells generating the cells that move through the body.
  • fast_forward00:29:27 - They then sort of get transformed into eggs on a new stage, and then they pass
  • fast_forward00:29:31 - through a sperm phase, or they got merged with sperm, and then there you have…
  • fast_forward00:29:35 - They get fertilized. Then you have your fertilized egg, right?
  • fast_forward00:29:38 - But it is a big part of the C. elegans, this reproductive organ.
  • fast_forward00:29:43 - Sure. But it's not all of it. It still have a neuron system and locomotive system.
  • fast_forward00:29:49 - But you sort of, you modeled this sausage machine that sort of is spitting out,
  • fast_forward00:29:53 - if you want, these new C. elegans, right? Yes, this is the granite.
  • fast_forward00:29:56 - This is the productive organ of the C. elegans. And so which elements of the
  • fast_forward00:30:01 - assimilation of the pancreas could you carry over to this C. elegans system?
  • fast_forward00:30:08 - By means of platform, I used the same platform. Principles were.
  • fast_forward00:30:12 - And the same principle was implemented.
  • fast_forward00:30:15 - Which is? And the way we construct the cell and the cell, different components,
  • fast_forward00:30:22 - the nucleus and the membrane and all the autonomous agent concept we discussed
  • fast_forward00:30:28 - earlier was just adopted to this model.
  • fast_forward00:30:33 - We had to change the name of the genes.
  • fast_forward00:30:35 - But this is basically it. and the connections between the states. Or the blocks.
  • fast_forward00:30:40 - Yeah, and the way they affect each other. So if we think about it as a graph,
  • fast_forward00:30:45 - we just had a different graph between the nodes and the neurons.
  • fast_forward00:30:50 - But I would assume that those are the number of states of the cells who have changed.
  • fast_forward00:30:54 - Yes, yes. If in the pancreas we had like 120 states in the cell against,
  • fast_forward00:31:00 - we had like 20, which is like five times less.
  • fast_forward00:31:05 - So, in the environment in which the C. elegans is growing, what is that environment?
  • fast_forward00:31:11 - Because the pancreas is growing in a very different kind of environment from
  • fast_forward00:31:15 - a... Yes, for example, in the pancreas model, we had a very complicated network of blood vessels.
  • fast_forward00:31:25 - In the C. elegans, it's very easy. At the tip cell, there is one cell that secretes
  • fast_forward00:31:30 - a factor to the environment.
  • fast_forward00:31:32 - That's it. So it was much easier to get a more accurate description of it.
  • fast_forward00:31:38 - And it was even modeled as a using order and order differential equation at the stationary phase.
  • fast_forward00:31:48 - So it's just like a chemical that is being secreted in a gradient.
  • fast_forward00:31:53 - It is much easier to capture.
  • fast_forward00:31:56 - So it's like one cell of the blood vessels of the pancreas model.
  • fast_forward00:32:00 - It's much easier. And this This is why it was much easier to generate good prediction
  • fast_forward00:32:09 - or testable predictions while in the pancreas.
  • fast_forward00:32:13 - Every prediction I had had to wait like five years until the biologists find
  • fast_forward00:32:18 - the right technique to test it.
  • fast_forward00:32:20 - In C. elegans it was tested within weeks.
  • fast_forward00:32:23 - But in the pancreas, I think the benchmark was very much the morphology of the structure.
  • fast_forward00:32:29 - That's correct, right? So what would be that benchmark for C elegance?
  • fast_forward00:32:34 - I think it's, okay, it's more accuracy because most of the, if in the pancreas
  • fast_forward00:32:41 - model, I had to use a lot of hand waving and say, okay, look, this is similar.
  • fast_forward00:32:47 - In the C. elegans model, I can talk exactly about the lengths of areas and the
  • fast_forward00:32:54 - lengths of zone and number of cells and cell cycle.
  • fast_forward00:32:57 - And indeed, out of all this refinement and fine-tuning of the model,
  • fast_forward00:33:03 - it turns out, for example, even in the small details, what is the different
  • fast_forward00:33:09 - cell cycle between the adult and the young C.
  • fast_forward00:33:16 - Elegans? During development, what is the ratio between the cell cycle at the two faces?
  • fast_forward00:33:23 - And we suggested in the model 1 to 5. And then biologists think it's 1 to 3.5.
  • fast_forward00:33:31 - So it's the same range. And we didn't take it into account.
  • fast_forward00:33:35 - We just try to make it as realistic as possible and to collaborate it with time.
  • fast_forward00:33:42 - And this all tiny detail just emerged. So that would mean for C.
  • fast_forward00:33:48 - Elegans, the benchmark is more like what kind of cells do you have in C.
  • fast_forward00:33:54 - Elegans and what position at what point in time?
  • fast_forward00:33:58 - And the number of cells. Sure, yeah, exactly. So you can start talking about
  • fast_forward00:34:02 - the numbers and quantity aspects while in the pancreas it will just look,
  • fast_forward00:34:08 - it looks like a cauliflower.
  • fast_forward00:34:10 - It's similar. But in the first C. elegans, we talk in the end about three or
  • fast_forward00:34:16 - four cell types, the stages, right? So why is that a hard problem?
  • fast_forward00:34:23 - Why this is a hard problem? Because it's not all about the cell type.
  • fast_forward00:34:30 - What controls them? What is the network? This is harder.
  • fast_forward00:34:34 - How the external signal affects the development. That's a hard problem.
  • fast_forward00:34:39 - It's hard to know what happens there. And not all is known.
  • fast_forward00:34:44 - Many, many issues there are open. For example, what causes cell death?
  • fast_forward00:34:53 - Is it autonomous? Is it inner decision?
  • fast_forward00:34:56 - Is there an external signal that kills them?
  • fast_forward00:34:59 - We know that 30% of the cell do not make it to death.
  • fast_forward00:35:05 - 30%, it's a lot of them. It's like third.
  • fast_forward00:35:10 - So we know that they don't make it. What happens when they go to apoptosis?
  • fast_forward00:35:14 - You mentioned this also in your talk, that this was one of the insights that
  • fast_forward00:35:18 - you had. But apoptosis of cells, so the cell death, is an active process,
  • fast_forward00:35:25 - right? This is really a regulated process.
  • fast_forward00:35:27 - There are receptors sitting at the outside of cells that are directly driving apoptosis.
  • fast_forward00:35:33 - So it's not like, well, in your simulation, you were telling us you were inducing cell death.
  • fast_forward00:35:39 - But you're saying, okay, if cells pass through this zone, there's a 30% chance they won't come out.
  • fast_forward00:35:44 - That's because apoptosis is very active in the adult. Thank you for watching!
  • fast_forward00:35:50 - In the developing organ, it's less active.
  • fast_forward00:35:54 - Usually, you know, how many cells a human, an adult is losing in his lifetime,
  • fast_forward00:36:00 - the embryo keep growing.
  • fast_forward00:36:03 - Usually, developing systems has less apoptosis than the adult.
  • fast_forward00:36:07 - Well, they better, or else it won't develop, right?
  • fast_forward00:36:10 - No, what I'm trying to get to here is that in C.
  • fast_forward00:36:15 - Elegans, we would assume that also their apoptosis is actively regulated.
  • fast_forward00:36:19 - For instance, depending on environmental factors, apoptosis might be more or
  • fast_forward00:36:23 - less, right? Depending on, let's say, the stress on the animal.
  • fast_forward00:36:26 - Well, in those elements, you sort
  • fast_forward00:36:29 - of, you are not really including in your simulation. No, we left it out.
  • fast_forward00:36:34 - We just flipped a coin at that stage. Exactly.
  • fast_forward00:36:37 - This is an extension to the model. I'm not sure there is enough data about apoptosis
  • fast_forward00:36:43 - and C. elegans at this stage.
  • fast_forward00:36:44 - I'm not sure about it. I don't want to commit. it but if there
  • fast_forward00:36:47 - is it can be formulated and added to the model okay i'm
  • fast_forward00:36:51 - still trying to understand how you build and tune these models so um in
  • fast_forward00:36:55 - the case of c elegans you have what you described a
  • fast_forward00:36:57 - very simple environment with a chemical gradient and then
  • fast_forward00:37:00 - you have your uh cell model for your
  • fast_forward00:37:03 - initial stem cell which uh presumably
  • fast_forward00:37:06 - when you set it up has some of the relevant details from
  • fast_forward00:37:09 - the biology but not all and then
  • fast_forward00:37:12 - i imagine you drive the model you see what happens
  • fast_forward00:37:15 - and you see well it's not building a worm for me yet and
  • fast_forward00:37:18 - then you say okay what else is what's wrong with this what
  • fast_forward00:37:21 - can i do different what you go back to the library and with much more many more
  • fast_forward00:37:26 - uh so so it's a very iterative process where you you run models and all the
  • fast_forward00:37:32 - time the models aren't working and gradually they get closer and closer to what
  • fast_forward00:37:36 - you would like to see yes okay and at that point you You say,
  • fast_forward00:37:39 - well, I think I've got a reasonably good working model of the wild-type worm.
  • fast_forward00:37:48 - Then you go to a different part of the library and say, let's look at mutants
  • fast_forward00:37:51 - and say, let's see what happens if we now flip the circuit. Do we get mutants?
  • fast_forward00:37:56 - So the validation stage is quite a strong one, really, because I think you already
  • fast_forward00:38:03 - have something that captures the development of the living natural animal.
  • fast_forward00:38:09 - And then you can mutate it in ways that you'd be fairly confident match real
  • fast_forward00:38:15 - mutations. Yeah, but that's a very important part of the modeling process.
  • fast_forward00:38:20 - But once you complete it, this is the really interesting part.
  • fast_forward00:38:24 - You start to play with it. You start to risk it. You can play the crazy scientist.
  • fast_forward00:38:29 - You can do whatever you have in your mind. And then you can get predictions.
  • fast_forward00:38:33 - And some of them are testable.
  • fast_forward00:38:36 - And for some of them you can tell whether they are plausible
  • fast_forward00:38:39 - and where whether they are testable some of
  • fast_forward00:38:41 - them are just crazy enough too crazy to be tested so some of the predictions
  • fast_forward00:38:48 - you make from the model turn out to be wrong yeah and then very often then you
  • fast_forward00:38:53 - go to stage one go back to the library so so the the methodology doesn't change
  • fast_forward00:38:59 - that much i guess you get your basic working model,
  • fast_forward00:39:01 - You try it out with some mutants, some it works with, some it doesn't.
  • fast_forward00:39:04 - You say, well, it didn't work with that mutant. I need to refine the model again. Yes.
  • fast_forward00:39:08 - There are a version of models all the time.
  • fast_forward00:39:12 - And this is a never-ending process because data flows in and as the data came
  • fast_forward00:39:18 - to your bench, you just need to rethink all the model.
  • fast_forward00:39:22 - Sometimes a small paper can change the whole model.
  • fast_forward00:39:25 - What I find interesting is that apparently you live in a field of developmental
  • fast_forward00:39:29 - biology where the literature is pretty clear because apparently you can just
  • fast_forward00:39:34 - go to the library and pick out the papers and get the parameters for your simulation and reality.
  • fast_forward00:39:39 - At least the field in which we exist, it's never like that, right?
  • fast_forward00:39:43 - You find the papers, they're contradictory, they're incomplete,
  • fast_forward00:39:46 - you have to meditate on this stuff.
  • fast_forward00:39:47 - They have to talk to the specialist to actually understand what the hell they're writing about, right?
  • fast_forward00:39:51 - So is it really fair to just say, okay, then I go to the library,
  • fast_forward00:39:55 - get the papers, I know my parameters.
  • fast_forward00:39:57 - I cannot believe that's how it works. That's not fair at all.
  • fast_forward00:40:00 - So how does it really happen? You know, every three papers have four different options.
  • fast_forward00:40:04 - Exactly right, yeah. But the modeling approach allows you to change between hypotheses.
  • fast_forward00:40:09 - And try to check different cellular mechanisms.
  • fast_forward00:40:13 - So you can play out scenarios, right? That's the whole point.
  • fast_forward00:40:15 - That's the whole point of modeling in general, I think.
  • fast_forward00:40:18 - But what the modeling, my modeling or the modeling I presented offer is kind
  • fast_forward00:40:26 - of a nicer visuals way for developmental systems,
  • fast_forward00:40:33 - in particular at the population level.
  • fast_forward00:40:37 - It's not a single pathway that we are testing.
  • fast_forward00:40:40 - It's not one cell. It's the whole population.
  • fast_forward00:40:43 - And it's the whole population over time and space.
  • fast_forward00:40:47 - So you can get a lot of new insight and ask a lot of new questions.
  • fast_forward00:40:53 - So that means also the model can summarize a lot of data.
  • fast_forward00:40:56 - It can help you to play out different scenarios to see, okay,
  • fast_forward00:40:59 - what if all this missing information would have this characteristic?
  • fast_forward00:41:04 - It could also highlight what you don't know. Because it summarizes it. Sure.
  • fast_forward00:41:09 - But now, in the end, if you want to turn that model into a theory,
  • fast_forward00:41:13 - you must explain something and you must make predictions, right?
  • fast_forward00:41:17 - So at this stage, if we look at the model of C. elegans, what do you think have
  • fast_forward00:41:21 - you really explained and what is a testable prediction of that?
  • fast_forward00:41:24 - Yeah, we explained a lot of the interplay between cell cycle and differentiation.
  • fast_forward00:41:31 - And one of the insights I presented is that if you change the ratio of cell
  • fast_forward00:41:39 - cycle between the larva and the adult,
  • fast_forward00:41:42 - then you get different behaviors. And one of them is that your stem cell population
  • fast_forward00:41:49 - gets shorter or smaller once you reduce it.
  • fast_forward00:41:54 - It seems a bit naive and it seems a bit straightforward, but the scientific
  • fast_forward00:42:01 - community never thought of looking at the consequence of the cell cycle at the
  • fast_forward00:42:08 - larva stage on the adult stage.
  • fast_forward00:42:12 - So this is kind of question that we can easily answer
  • fast_forward00:42:16 - or ask what happens if you change something
  • fast_forward00:42:19 - at day zero what happened in day five right and it's
  • fast_forward00:42:22 - not that easy to to test in the in the lab because the you change something
  • fast_forward00:42:29 - and you need to wait a few days it's better to have kind of what where to look
  • fast_forward00:42:34 - at not not what the answer is but where where to look and search and research.
  • fast_forward00:42:41 - And in the system right and what that
  • fast_forward00:42:44 - may give us an answer about it so then
  • fast_forward00:42:47 - the last example you you analyzed and presented was
  • fast_forward00:42:50 - neurogenesis right the development of the nervous system and again you roughly
  • fast_forward00:42:55 - took the same modeling framework the same cell model but now again changing
  • fast_forward00:42:59 - again the states changing the
  • fast_forward00:43:02 - the expression path of the control pathways ways to build a piece of brain,
  • fast_forward00:43:08 - it's not a piece of brain it's a piece of the nerve.
  • fast_forward00:43:12 - System because what you see there what will be later the nerves so what did
  • fast_forward00:43:19 - you exactly learn from that exercise what did you find.
  • fast_forward00:43:22 - Before we go there, in terms of actually building that model,
  • fast_forward00:43:27 - from the previous two models, were you able to take little networks of the blocks
  • fast_forward00:43:34 - from within those models and use them again?
  • fast_forward00:43:36 - Because we're interested this week in how evolution has reused some basic processes,
  • fast_forward00:43:44 - perhaps in very different ways, sort of regulatory dream networks that might
  • fast_forward00:43:48 - be involved in body patterning and inversion,
  • fast_forward00:43:50 - but reused again in building bits of brain.
  • fast_forward00:43:53 - And I wonder if there are examples of that in your work that you can say there
  • fast_forward00:43:58 - was a network in one part of my cell and I was just able to use it in the pancreas, in the C.
  • fast_forward00:44:04 - Elegans and in the mouse brain. First, I'd like to give you some statistics that may explain it.
  • fast_forward00:44:11 - The first model, the pancreas, it took five years to develop.
  • fast_forward00:44:15 - The second model, the C. elegans, it took two years.
  • fast_forward00:44:21 - And the model of the brain, of the neuronal migration, took three months.
  • fast_forward00:44:27 - So we short the time of developing the model itself. off.
  • fast_forward00:44:32 - The analysis keeps taking a long time and looking for the prediction is a long period,
  • fast_forward00:44:39 - but to develop, since we have a good view of what the modeling framework and
  • fast_forward00:44:46 - the approach is, it takes less time to develop the model.
  • fast_forward00:44:51 - So the basic concept and the basic approach, as time goes, as the research continues,
  • fast_forward00:44:58 - it's much easier to implement it.
  • fast_forward00:45:01 - And this concept of autonomous agent and the concept of having this sensor unit
  • fast_forward00:45:07 - and the internal switch unit and the differentiation and proliferation are being
  • fast_forward00:45:12 - controlled by these two components seems to work.
  • fast_forward00:45:17 - So this seems like a general concept in developmental biology.
  • fast_forward00:45:21 - I'm very, very careful here because I don't want to claim, hey,
  • fast_forward00:45:25 - I have the holy grail, I'm having the way to model.
  • fast_forward00:45:28 - But it seems very beneficial. we have we are
  • fast_forward00:45:32 - encouraged by the results we don't it
  • fast_forward00:45:35 - doesn't mean we hit the jackpot you know it's kind
  • fast_forward00:45:37 - of we still have to we still have to to do
  • fast_forward00:45:41 - a lot but what was the benchmark what was the benchmark in this case what did
  • fast_forward00:45:44 - you replicate and how accurate was that replication and in the neuron we replicate
  • fast_forward00:45:49 - the way neurons migrate from their birthplace to the place where they are going
  • fast_forward00:45:55 - to going to become part of the nerve system.
  • fast_forward00:46:00 - So it's from the core of the brain to the surface.
  • fast_forward00:46:05 - It is guided by fibers, by the gallial fibers.
  • fast_forward00:46:08 - And at the first stage, they use the gallial fiber to track the road.
  • fast_forward00:46:15 - Then something happens and they stop following this gallium fiber,
  • fast_forward00:46:22 - moving randomly in space.
  • fast_forward00:46:23 - Then they reattach to the gallium fiber and just putting layer on top of the
  • fast_forward00:46:29 - layer of neurons on the surface of the brain.
  • fast_forward00:46:33 - But now, so in your sim, so what we're really talking about is also really the
  • fast_forward00:46:36 - migration of cells along these guiding axes of glia or certain kinds of gradients
  • fast_forward00:46:44 - that might attract them or repel them.
  • fast_forward00:46:47 - So that means in that setup,
  • fast_forward00:46:50 - collisions is an issue right that's cells if
  • fast_forward00:46:54 - these migration patterns get disrupted cells might not
  • fast_forward00:46:57 - be able to migrate across certain other cell population it's in all cases they
  • fast_forward00:47:02 - do not they do not migrate one towards the other and one across the other they
  • fast_forward00:47:07 - if they see that there is a cell in a neighboring in a neighboring pixel they
  • fast_forward00:47:12 - won't move to it no way but in reality.
  • fast_forward00:47:16 - Cells have to go through layers of cells that might have embedded themselves
  • fast_forward00:47:23 - earlier, right? Because you always go to the outside.
  • fast_forward00:47:26 - Yeah, but you need to think, okay, it's not exactly three-dimensional,
  • fast_forward00:47:31 - not exactly two-dimensional. It's somewhere in the middle.
  • fast_forward00:47:34 - So there is a layer of cells. They climb above it.
  • fast_forward00:47:38 - They're creating another layer and they are walking on top of them and then
  • fast_forward00:47:44 - they will place themselves on the same layer. So it's kind of two slices.
  • fast_forward00:47:48 - So they go over and... What I was trying to get at, Zoltan Molniar was here
  • fast_forward00:47:55 - early this week talking also about development of the brain.
  • fast_forward00:47:58 - And a typical pattern there is that you have certain subpopulations moving from
  • fast_forward00:48:03 - these neural plates at different points in time and also having to cross through.
  • fast_forward00:48:08 - They cross through other populations. And they cross through other populations.
  • fast_forward00:48:12 - And what I wanted to ask you about is that I would...
  • fast_forward00:48:16 - My claim would be that with your simulation technique, you cannot handle that because.
  • fast_forward00:48:22 - One location in space has just one specific state.
  • fast_forward00:48:27 - So the dynamics of, let's say, migration and possible collision and obstruction
  • fast_forward00:48:33 - and so on is not something that you can capture that way, or would you think
  • fast_forward00:48:36 - that's too negative interpretation?
  • fast_forward00:48:38 - I think you've just defined a research question. Okay. It's enough to be a PhD thesis.
  • fast_forward00:48:44 - This is something to investigate. So I might have a chance to get my PhD done.
  • fast_forward00:48:48 - That might work. You're accepted. Okay, great.
  • fast_forward00:48:52 - But you still didn't fully answer my question, and maybe I don't know if it was a bad question.
  • fast_forward00:48:57 - But you said that for the brain model,
  • fast_forward00:49:01 - you keep the same structure of the previous two models for the cell.
  • fast_forward00:49:05 - But what I was interested in, all these cells are doing similar behaviors of
  • fast_forward00:49:09 - differentiating and migrating.
  • fast_forward00:49:11 - Are there some bits of network within the decision-making parts of the cell that you could reuse?
  • fast_forward00:49:18 - Yeah, the proliferation, the cell cycle is the same to all the models.
  • fast_forward00:49:28 - There is the decision about asymmetric and symmetric proliferation,
  • fast_forward00:49:32 - which is different between them.
  • fast_forward00:49:34 - But the proliferation component is the same, and the differentiation component is very similar.
  • fast_forward00:49:41 - So this is exciting in terms of when we think about evolution and we think,
  • fast_forward00:49:47 - well, brain neurons are very different from C.
  • fast_forward00:49:50 - Elegans neurons, but actually they share a lot of the same chemical machinery
  • fast_forward00:49:54 - for doing what they do. Yeah.
  • fast_forward00:49:57 - You have to change the specificities of what they're doing, but say networks can work internally.
  • fast_forward00:50:02 - Yeah, I think this is the discussion that we had at the end of the talk,
  • fast_forward00:50:05 - that you think that you're surprised that the biology is simple building blocks,
  • fast_forward00:50:11 - and I see it as the reality.
  • fast_forward00:50:13 - It's kind of, yeah, it's surprising that the biology carries simple building
  • fast_forward00:50:20 - blocks, but I think this is the way it is.
  • fast_forward00:50:22 - I don't think it is as complicated as we'd like to. Not wanting to agree with
  • fast_forward00:50:26 - either of you, I could argue, wait, you guys are both deeply confused because
  • fast_forward00:50:32 - we're scientists. Of course we're deeply confused.
  • fast_forward00:50:36 - Right. But the point is that in your case, what carried over between these phenomena
  • fast_forward00:50:43 - was a modeling strategy, an algorithm, if you want.
  • fast_forward00:50:47 - But that technology as such doesn't tell you much about guiding principles because,
  • fast_forward00:50:53 - as you said yourself, you don't have all these principles explicitly defined.
  • fast_forward00:50:59 - And Tony is seeking, let's say, common principles across all these different things.
  • fast_forward00:51:05 - Like the way you would grow an organ, a pancreas would include principles that
  • fast_forward00:51:10 - you would carry over to having a C.
  • fast_forward00:51:13 - Elegans sausage machine. But I think we said that there were some intrinsic
  • fast_forward00:51:17 - cellular networks that you could copy whole chunks of the network across from
  • fast_forward00:51:23 - these two different cell types.
  • fast_forward00:51:25 - If you prefer to think of it as a mechanism of tears, that this is a motive
  • fast_forward00:51:30 - of a mechanism of a cell, this is another way to look at it.
  • fast_forward00:51:34 - Because, okay, but then we have to, if you gentlemen are so optimistic about
  • fast_forward00:51:38 - this, you must declare from me. We're scientists, we must be optimistic.
  • fast_forward00:51:41 - Right here and right now, what the common principles are between pancreas,
  • fast_forward00:51:47 - C. elegans, and this piece of brain.
  • fast_forward00:51:50 - So they both have proliferation, differentiation. They both have a kind of component
  • fast_forward00:51:55 - of external sensor recognition or signaling, and they both have kind of internal
  • fast_forward00:52:01 - inner cellular decision.
  • fast_forward00:52:03 - And the decision of the mechanism must be defined by all the three components.
  • fast_forward00:52:08 - It will be wrong to say that the environment and the extracellular signaling
  • fast_forward00:52:13 - is not important, And it will be even more wrong to say that the internal genes is not effective.
  • fast_forward00:52:21 - And it will be wrong to say that this has no proliferation or no differentiation.
  • fast_forward00:52:26 - So this is necessary components for all the three elements, whatever it is,
  • fast_forward00:52:33 - whether it is the pancreas in mice or the C.
  • fast_forward00:52:36 - Elegans, germline development, or the neural development. Yeah,
  • fast_forward00:52:40 - but the risk is, of course, that you say, look, E.
  • fast_forward00:52:42 - Coli can have flagella and it can sort of flop around in some gradient and humans
  • fast_forward00:52:47 - walk and they share common principles because they navigate.
  • fast_forward00:52:50 - So the point here is that, Fritz, I could argue that in case of the pancreas
  • fast_forward00:52:55 - and the brain, we might have all sorts of regulatory genes that really tightly
  • fast_forward00:52:59 - orchestrate this process,
  • fast_forward00:53:01 - while the same might not hold for the complete pathway of the C.
  • fast_forward00:53:06 - Elegans or the trajectory that cells go through in C.
  • fast_forward00:53:09 - Elegans. you will not have single regulatory genes guiding and orchestrating that process.
  • fast_forward00:53:14 - Well, it's never a single gene. It's kind of a combination of many genes.
  • fast_forward00:53:19 - It's between one and many.
  • fast_forward00:53:22 - But with the bacteria, with the E. coli flagella, it's a different system.
  • fast_forward00:53:29 - It's not developing. It's hardly proliferating. If you want to see it.
  • fast_forward00:53:32 - I meant something else with this. It was like seeing a similarity between the flagella of E.
  • fast_forward00:53:38 - Coli in our legs because they're both used for navigation and say,
  • fast_forward00:53:42 - ah, this is now possible to have a common principle. Maybe they are.
  • fast_forward00:53:47 - I do, roll it out. It seems a long shot, though. So those abilities are probably
  • fast_forward00:53:52 - in our genome. They're just not expressed.
  • fast_forward00:53:55 - And I think what you're saying is it's not surprising because we're all descended
  • fast_forward00:53:59 - from a common ancestor and C. elegans and the mouse and ourselves.
  • fast_forward00:54:05 - In that common ancestor, there were stem cells. They migrated.
  • fast_forward00:54:09 - They built complex bodies.
  • fast_forward00:54:10 - And so what more do you need than those mechanisms? So most of it's already going to be there.
  • fast_forward00:54:17 - And you say you're not surprised. I'm a little bit surprised because I think
  • fast_forward00:54:21 - that maybe in rodent brains, there's more that stem cells do.
  • fast_forward00:54:25 - But for you, that's not a qualitative shift in what they do.
  • fast_forward00:54:29 - It's just a bit more richness, perhaps.
  • fast_forward00:54:31 - Yeah, it's more richness. I don't think it's more complex.
  • fast_forward00:54:34 - So look, now that you gentlemen refuse to see the light, I'm shining on this
  • fast_forward00:54:38 - issue. Maybe we should get to the finish line.
  • fast_forward00:54:40 - And the question there is, so Yaki, as you also shared with us,
  • fast_forward00:54:43 - it's actually interesting. saying the community of people doing your kind of
  • fast_forward00:54:48 - work, like modeling these developmental, it's actually rather small. It's very small. Yeah.
  • fast_forward00:54:53 - So you're sort of chipping away at this, making progress.
  • fast_forward00:54:58 - So, but in that sense, given what you've seen and learned in,
  • fast_forward00:55:02 - in this sort of as a solo agent in this, this field of developmental biology,
  • fast_forward00:55:07 - what would be Jackie's law that we should adhere to in trying to understand
  • fast_forward00:55:11 - the developing biological system? Well, keep it dynamic.
  • fast_forward00:55:15 - Look at the dynamics. I think that is the most important.
  • fast_forward00:55:18 - I think many people in this field fail.
  • fast_forward00:55:22 - To look at, to think dynamic, to think that an event now has effect in the future of the animal.
  • fast_forward00:55:29 - And if you look at that, you'll find an all new world of results.
  • fast_forward00:55:34 - That would be my law. Okay.
  • fast_forward00:55:36 - Now, Tony likes traveling, and so I ship him around the world.
  • fast_forward00:55:40 - So when are you coming to Israel?
  • fast_forward00:55:41 - Exactly. I'm going to tell you that. I can tell you this exactly.
  • fast_forward00:55:44 - Four years from now, he's going to come to Israel, and he's going to meet up
  • fast_forward00:55:49 - with you, and he's going to check a prediction he's going to make today.
  • fast_forward00:55:52 - Tony is going to ask you, look, four years ago, you made a specific prediction
  • fast_forward00:55:55 - that you would observe X in your simulation.
  • fast_forward00:55:59 - And today, I want to know, four years in the future from now,
  • fast_forward00:56:02 - whether this was validated.
  • fast_forward00:56:05 - So, what's the one prediction you're going to make for us today that you will see tested by then?
  • fast_forward00:56:10 - Yeah, well, I'll give a different kind of prediction.
  • fast_forward00:56:12 - I'll predict that more people will do my kind of modeling.
  • fast_forward00:56:17 - Come on, you can do better than that. It has to be about the mental biology. No, no.
  • fast_forward00:56:22 - Developmental biology. Okay, so I predict that more people in development biology will do.
  • fast_forward00:56:27 - No, I really think that what I'd like to see is not prediction in the lab that
  • fast_forward00:56:34 - I've predicted and was accurate.
  • fast_forward00:56:37 - I would love to see it comes out of the model, but I would love more to see
  • fast_forward00:56:41 - people doing my kind of science. and maybe in the long run there will be a model
  • fast_forward00:56:47 - of a human being that we can play before we go do anything.
  • fast_forward00:56:52 - That can be something. This would be my vision.
  • fast_forward00:56:56 - I'm not looking into a particular prediction that will be validated.
  • fast_forward00:57:02 - All right. Giacchiosetti, thank you very much for this conversation. You're welcome.
  • fast_forward00:57:07 - Antonio, you're a bastard. You're supposed to make it difficult for him.
  • fast_forward00:57:10 - You're not supposed to agree with him. The CSN podcast was produced by the Convergent
  • fast_forward00:57:14 - Science Network of Biometrics and Biohybrid Systems, a project funded by the
  • fast_forward00:57:20 - European Sevens Research Framework Program.
  • fast_forward00:57:24 - For more interviews, recorded lectures, or upcoming conferences in the field
  • fast_forward00:57:30 - of biometrics and biohybrid systems, go to csnnetwork.com.
  • fast_forward00:57:36 - Music.

Be the first to leave a comment

Leave a comment

Your email address will not be published. Required fields are marked *

Exploring the convergence of neuroscience, robotics, and AI through conversations with leading researchers since 2010.

A project of the Convergent Science Network Foundation.

© CSN Podcasts. Developed by IMCreativeWEBC

0%

Login to enjoy full advantages

Please login or subscribe to continue.

Go Premium!

Enjoy the full advantage of the premium access.

Stop following

Unfollow Cancel

Cancel subscription

Are you sure you want to cancel your subscription? You will lose your Premium access and stored playlists.

Go back Confirm cancellation