Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Episode 27: The Connectome
27 Plays4 years ago
30 min episode
A connectome is a representation of every connection between neurons in the brain. Recent brain-slicing technology, in addition to image recognition tools, has begun to make this science-fiction idea become a reality. Rolf and Joe discuss the recent publication of the largest completed connectome to date, that of the fruit fly drosophilia. The database for the connectome is publicly available, and includes huge amounts of data about every one of the approximately 25,000 neurons mapped to date.
Paper:
A connectome of the adult drosophila central brain (2020)
OR access the database yourself at:
https://neuprint.janelia.org/
Recommended
Transcript
Introduction and Podcast Purpose
00:00:10
Speaker
Welcome to Cognation. This is your host, Rolf Nelson, coming to you from Providence, Rhode Island. And this is me, Joe Hardy, coming to you from El Cerrito, California.
Exploring the Connectome
00:00:21
Speaker
And today we're going to do a special shorter length podlet, commute length podcast, about 30 minutes. And today's topic is the connectome, the set of connections between neurons in the brain. The reason this comes up as a topic is because just recently there has been a publication of a full connectome for half of a fruit fly brain, which has been a major advance in
00:00:50
Speaker
in the science and something we're really excited about.
Comparing Connectome and Genome Projects
00:00:56
Speaker
Yeah, most of you probably have heard of the Human Genome Project with the goal of mapping all of the genes in the human body. And the Connectome is sort of named as an analogy to the genome with the idea that
00:01:16
Speaker
The connectome is all about all the different connections between neurons, specifically in the brain, but it could be anywhere in the body. But in this case, it's really about the brain. And the idea behind that is that if you can think about a model of the brain and understanding how it works, it would be very helpful to know what the units of that model are and how those units are connected.
00:01:43
Speaker
Yeah, so we can think of the connectome as essentially a wiring diagram for the brain.
Complexity of the Human Brain
00:01:49
Speaker
So in the human brain, there are approximately 85 billion neurons and over a trillion connections between them. So this is a massive project. And the idea that you could plot every single one of those neurons and understand the relation between all the neurons in the brain is a pretty ambitious project.
00:02:12
Speaker
This has been undertaken before in smaller creatures. The first creature to ever have its entire connectome mapped was the C. elegans. So a model animal, just a tiny little worm that's about a millimeter long, has had its entire connectome mapped. Now this creature has somewhere on the order of 300 neurons.
00:02:37
Speaker
So think of the difference in complexity between a creature like this and a brain like ours and its orders and orders of magnitude more complex. So this first connectome was mapped out in the 80s.
00:02:53
Speaker
And now, of course, we've got much more sophisticated tools.
Mapping the Fruit Fly Connectome
00:02:58
Speaker
We've got a lot of image recognition tools that can help put slices back together and understand how the 3D structure of the brain works. So the next step in this is the fruit fly, so Drosophila, which is a big animal model. The fruit fly has about 25,000 neurons.
00:03:23
Speaker
know, substantially more complex than a C. elegans worm, but not nearly to the level of human complexity. Yeah, so I think, you know, this, this model of the brain, you know, with about 25,000 neurons in this Hemi brain, you know, basically, you know, is much closer to the human brain than C. elegans is, but it's still very far away. So we're still talking about
00:03:52
Speaker
orders of magnitude less complexity. But it's very, very complex. And if you think about all of the connections that are happening there, there's a lot. There's a lot. Millions of pre-synaptic sites and tens of millions of post-synaptic connections.
00:04:15
Speaker
Yeah, that's right. That's a good point. So I said 25,000 neurons in the fruit fly, and you're right. So it's 25,000 in just the hemi brain, so just about half of the brain that they're mapping. And even with those 25,000 neurons, they're somewhere around 20 million connections between them. And this is a very, very tiny brain. I mean, how big a fruit fly is. I think the size of this is about 250 micrometers cubed.
00:04:44
Speaker
It's pretty sensitive and tiny little brain. Yeah, and if you think about the size of that brain, part of the challenge of getting these
00:04:57
Speaker
connections mapped is to actually image this brain.
Research Collaboration and Tools
00:05:00
Speaker
So they had to, part of the work that this group was doing, and this is a group out of the Janalia research campus at the Howard Hughes Medical Institute and some researchers at Google research, as well as a number of other sites. It's a big, big multi-site project with a lot of collaborators. You know, it's an intersection between
00:05:25
Speaker
biologists, neuroscientists, a lot of computer science goes into this, and a variety of other disciplines as well.
00:05:34
Speaker
Right, so this is definitely a big coordinated project. And I guess this is not unrelated to the human connectome project, which is, at this point, not quite at the level of mapping every single neuron in the brain. But it's been an ambitious project. When did this start? Around 2010 or so? 2008, you remember? No, that sounds about right. I don't remember exactly what that sounds about right.
00:06:04
Speaker
Somewhere around then, and it's been an ongoing project to try and map connections between different brain regions, mostly using fMRI. So those of you who are aware of limitations of fMRI machines, so we're not getting quite the kind of resolution. We're not going to be able to see individual neurons in something like that. So a broader scale, ambitious project with a lot of different labs involved, but doesn't get the sort of detail that something like this does.
00:06:33
Speaker
Yeah. And so if you're thinking about how do you image every neuron in the brain of a fly, well, you need to chop it up into little bits and then put every one of those little bits in a microscope. Right. So the first thing I should say is you can't do this while the fly is alive or you can't, I mean, you wouldn't be able to do this sort of technique with a living human.
00:06:55
Speaker
Right. Yeah, exactly. And in this case, that's just when they say like the fly brain, they're talking about like literally one fly in this case, because it has to be one individual because in order, because the slices obviously need to be continuous. You need to actually have all the connections in place for that one individual. So this is one female fruit fly.
00:07:15
Speaker
So this is like, it's like the, uh, it's like mapping the genome is, is of course, it's going to be different for different people. So you have to have one single model that you're sort of basing that standard genome on. So same with, um, the connectome too, right? Yep. Yeah, exactly. Exactly. And so then, and then they have to figure out, you know, a bunch of different things they had to figure out how to do here.
00:07:37
Speaker
to make it happen. One of the things they had to do was like how to figure out how to slice this brain, like without messing up the connections, you know, so that you could sort of draw the line between one slice and the next slice. And so they did a bunch of work on figuring out like how to get the best, like hot knife to like slice through this tiny little brain.
00:07:59
Speaker
Right. So we've always sliced brains to understand the physiology of brains, but this is kind of a new, this calls for pretty precise instruments. Yeah, absolutely. And then you have to basically be able to like stitch those slices back together and figure out like
00:08:20
Speaker
how each, you know, and that, you know, within like a certain, like, if you think about what the microscope is doing, it's creating like these volumes, you know, these like what are called voxels, which is like a little, like a pixel, but in volume. So it's like creates these voxels that have, you know, these little, uh, 3d image, um, cubes in them. And you need to be able to connect those.
00:08:46
Speaker
So I guess thinking of scale, too. So a voxel, usually a voxel in an fMRI experiment is about a millimeter cubed. So that would represent the smallest resolution that you could get in fMRI. And a fruit fly brain, in this case, is much smaller than that. Right, exactly. I mean, the whole brain is just going to be not much bigger than one voxel. Yeah.
00:09:15
Speaker
Uh, uh, of the FMRI. So yeah. So really, really small little things. Yeah. And you go look into those images and you try to basically figure out how within each of those, every individual cell needs to be segmented. And then over from voxel to voxel to voxel, the connections between them, like how those, because these neurons, you know, how we say I have long traces.
00:09:44
Speaker
the cells are extensive and connect over space. It's almost like a plumber having to figure out what goes with what and connect them up. You can see how it gets followed, the entire neuron including all the synapses and the axon, how it all gets traced through all of these different slices. I think in this case, there were 37 slices.
00:10:14
Speaker
Which is impressive, the fact that you could cut this tiny little piece into 37 slices and image them closely enough so that you can locate all of these neurons and the synapses and see how they connect to each other. Absolutely, yeah. And it required a lot of artificial intelligence work to basically automatically segment and then identify these cells.
00:10:41
Speaker
I mean, this is something that would have never been possible in the 80s or 90s because you would have had to do these things by eye. So you'd have to look at them and try and match them up. And this is just computationally way too much information for a person to look at and map up. So you need some sort of automatic process that does this.
00:11:01
Speaker
Right. And then, you know, of course, you know, those automatic processes do depend on labeled data. So at some point somebody had to label all of these cells, make decisions about these types of cells and what these cells sort of look like and how the shapes of them were so that the machine could learn from that and then make decisions on these novel, novel voxels. But then, you know, a lot of these they needed to, there's a lot of, uh,
00:11:28
Speaker
proofreading that needed to be done. So both computers and humans had to go back in and check and double check and triple check to make sure that these segments and these labels, these identifications, the names we were giving to the cells were the correct names. So there was a lot of manual work involved in that. 25,000 cells, each of which needs to be segmented and labeled correctly.
00:11:58
Speaker
A lot of double, triple checking by humans as well. So I think it probably, I don't think they said in here how many hours it took, but it must have taken. Good Lord, a lot. Hundreds of years of human effort, if you think about in total.
00:12:16
Speaker
Yeah, and so for those of you who are interested in seeing some of the results of this, so there is a publicly available database that you can look through.
Challenges and Scale of Connectome Data
00:12:27
Speaker
Now, you can imagine that the amount of information that something like this generates is huge. So I think this generates something like 20 terabytes of data, just 20 terabytes describing what's going on in that fruit fly brain.
00:12:44
Speaker
And yeah, just to give a sense of that, they gave a nice little, little detail in the paper, which was if you have like a gigabit fiber connection. So a gigabit per second of data being transferred, it still takes a couple of days to download this database.
00:13:01
Speaker
So, not in substantial amount of data here and. If you go on the, if you go on the website that contains their that contains the data visualization and we'll put that on the show notes, you can see.
00:13:17
Speaker
some really amazing stuff. So number one, so there's a whole bunch of different cells that they're identifying here. So they have to figure out the morphology of each different kind of cell in the fruit fly brain, which is quite a few. I don't know how many different cells are there in the fruit fly brain? A whole bunch. I don't know. I mean, I think they didn't, they said how many they were,
00:13:46
Speaker
you know, how many classifications they had in there. And they also mentioned that, you know, there's disagreement about that because you have lumpers, you know, and, and splitters. So some people like to have some sorts of ambiguity where it's a little, it's a little tough to tell, especially at that small scale. Yeah, absolutely. So, so, and some people prefer to, uh, you know, say that two cells that look similar are the same cell type and other prefer to say they're actually different cell types.
00:14:15
Speaker
Now the cool thing about the database is you can actually go in there and you can select any brain location that they've mapped out. So clusters of cells that may connect to other clusters of cells. And you can see exactly what the strength of the connections between those areas is. And you can also see connections between every individual cell and where it connects to. So it really gives a lot of amazing detail on this. So just combing through it is actually kind of fun to do.
00:14:43
Speaker
Yeah. And it's a totally open source. So if you want to go in, if you're so inclined to that type of person and you have those skills, you can go in and do your own analysis on either the images or any of the resultant data that they've, that they've, uh,
00:14:59
Speaker
tabulated and that's really the goal of this. One of the goals is to create this data set that other researchers can then go to use to do analyses and generate hypotheses or test hypotheses.
00:15:22
Speaker
That's the basics of what they're doing in this paper. Maybe we can talk a little bit about some of the limitations of the Connectome and Connectomics, and then of course some of the potential and some of the cool stuff that you can do with it.
00:15:44
Speaker
So first, what does this tell us? So what are we learning from having, if we could have an entire connectome of the human brain, would we know everything about the brain? Is that all there is? No, of course not. I mean, there's a whole lot that just the morphological connectivity doesn't tell you
00:16:15
Speaker
You know, all you need to know about the physiological connectivity. So, you know, that doesn't give you anything about time. There's like this, you know, all the dynamics and it doesn't give you anything about what's going on in the cell. Most of the interesting stuff is going on inside each cell.
00:16:32
Speaker
you know, how they respond to the intracellular environment, you know, how they respond to neurotransmitters, which is the, you know, the molecules that, that communicate between, uh, different neurons. All that stuff is not given to you at all by the connectome. So you don't get any of that information about the connectome. It does, they have tried to map.
00:16:56
Speaker
you know, the types of, uh, synapses that are there. So like what, uh, you know, and the type of synapse will tell you a lot about what neurotransmitters are being exchanged between different cells. So you can get an idea of dopamine pathways and acetylcholine pathways and things like that. Right. So it does give you a lot of information if you have a model, but it requires other
Connectome's Role in Understanding Diseases
00:17:20
Speaker
bottles. If you want to, if you really want to create, uh,
00:17:24
Speaker
a model that gives you some information about how information is processed, it requires other information. It doesn't, other details, other data that is not just the connectome. You need something about how those neurons are actually interacting.
00:17:38
Speaker
So in other words, we're not yet at the point where if we know someone's full connectome, or even if we get to that point, it's not going to act as a full model of the brain so that we could simulate exactly what the brain's doing and maybe guess at what someone's behavior is.
00:18:01
Speaker
Right. It's a little bit like the genome, right? To if you think about it, because like just knowing what all the genes are is awesome. But if you don't know what the genes do or how they work or like, you know, how they interact physiologically, then you don't know very much at all. But because we already know a ton about what genes do knowing what all the genes are is very helpful.
00:18:22
Speaker
And similarly with the connectome, it's like just knowing the connectome by itself wouldn't be that helpful. But we actually know a lot about the physiology of how neurons work and interact. So knowing how they're connected can be linked to other models and other hypotheses.
00:18:38
Speaker
And one of the things I think that some of the big promise for this, or at least some of the suggested promise for the connectome is that you can help use it to identify neurological diseases. So maybe there's a signature connectome or a signature pattern in connectomes that describes, say, Alzheimer's or Parkinson's or other neurological diseases. So we can get a much better grasp on how they work and what's going wrong.
00:19:09
Speaker
Right. I mean, I guess, you know, one of the things talk about limitations. I mean, we're still very far from a human connectome in the sense of like the orders of magnitude between the fruit fly and the human are are really enormous. I mean, you're talking about going from terabytes of data to petabytes of data.
00:19:27
Speaker
thousands of times more data. And you're already really struggling to even process the information on the fruit fly, which they have done, which is great. I mean, it's amazing if you think about they've actually mapped this entire hemi brain. It's half a brain.
00:19:42
Speaker
as far as they can, they feel complete in terms of their mapping of these. They know they probably missed some cells here and there, but they feel like they've pretty much got most of it. And that's awesome. But it's like way, just way fewer, just number of cells than- Right. Imagine how many fruit fly brains could fit in your head.
00:20:03
Speaker
Yeah, yeah, a lot. Yeah. And the other thing about the fruit fly brain is, you know, fruit flies do have different brains than humans. Like even the neurons, how they work are different. They just like have, there's just structurally different kinds of neurons than we have. So it's interesting. They synapse in a different way too. So there's structurally some, some regularities that are different between brains in that sense. Right. Exactly. I mean, you know, there's some things about the way that neurons work that are highly preserved.
00:20:33
Speaker
you know, across the animal kingdom, but evolutionarily, but you know, there's a lot of things that are very different. And the plants are very, very different than humans.
00:20:43
Speaker
Now, one of the cool things that I'm thinking about possibilities here is I know that there are some people that are working on relating a connectome to behavior. So like I said, with C. elegans, that worm that has its complete connectome mapped. So this is probably the best understood organism there is because it's this model animal that we just know everything about.
00:21:08
Speaker
So even with those 302 neurons, and then in addition to 95 muscle cells that this creature has, it's still not possible to understand how behavior results, or the relation between neurons firing and the resulting
Modeling Neuron Interactions
00:21:26
Speaker
behavior.
00:21:26
Speaker
That's not mapped out yet, and there are people who are working on this. I don't think we're close to it yet, but there's a company, or there's a project called Cybernetic that is attempting to model, it's really a biological model, so it models how membranes and
00:21:47
Speaker
and matter and the environment interact and the idea is you can model these neurons more thoroughly like you say because there's a lot going on in the neuron and then how it interacts with the rest of the body and muscle cells to get a more complete understanding. And of course this is the kind of thing that we would love to have for humans. I mean this is what psychology is all about figuring out exactly what's going on in the brain
00:22:13
Speaker
and then relating it in a really precise manner to the resulting behavior. But if you can't do it in 302 neurons right now, and that's a huge substantial problem, then it's hard to see how that's going to happen with something just hugely more complicated than C. elegans brain.
00:22:35
Speaker
Yeah, so there's a lot of work to be done there. But maybe we could talk a bit about what it would be like to have these models or what the purpose of these big models would be if you could get everything mapped. I mean, I guess that's the goal, would be to have a model of a brain initially with a small animal and then as you get bigger and bigger towards humans,
00:22:57
Speaker
you know, a model that basically represents the operation of a brain in response to stimuli and making decisions and so on and so forth so that you could really see, you know, and test how different changes to the model would affect, you know, responses and behavior to better understand how our brains work. I mean, that's really to me, at least as a psychologist, that's sort of the goal.
00:23:23
Speaker
I agree. And I remember thinking about this years ago that I would love to be at the point where conferences in psychology are about testing one single model organism. So a brain with a connectome fully mapped out that you can use as a test model and convey this to other researchers. I think it would be huge for understanding everything about the brain and behavior. I mean, this would be a great
00:23:51
Speaker
This would be amazing. This is what psychologists and neuroscientists dream of for a research model. So absolutely, that would be, I mean, that would be a huge possibility. What else do you see? I mean, even sort of science fiction level, what else? Well, I guess it kind of gets into that whole thing about like, you know, if you had a model of a brain
00:24:15
Speaker
that was so good, the artificial intelligence of the model was so good, that it mapped every aspect of the way that a human brain worked. What would be the implications? A lot of those details, all the neurotransmitters, everything solved. So if we could figure out for C. elegans and then we move up and solve it for humans too. Right, and that's all in, instead of being in wetware, it's all in hardware, it's all in a computer.
00:24:46
Speaker
What would be the implications of that model? Is that model, if you turn that model on and gave it stimulus and how to make decisions, is it possible that that model could have consciousness, for example? Well, this is where our half hour episode, I don't think can quite
00:25:08
Speaker
I agree. I think that's the most interesting thing to think about is a future in which you can create a full model of your cell for your brain and it can be fully simulated on a computer. Yeah. I mean, that's the singularity right at that point. If you could download
00:25:34
Speaker
You're a brand. That's a little bit different downloading because there's some additional complexity. Well, you have to be comfortable with having your brain sliced up, too. Yeah, there's that issue. That's a topic. That scanning would ever make it possible to get that level of detail, yeah. Right. I mean, I think there's an additional level, well, probably orders of magnitude more complexity of
00:26:01
Speaker
getting a model that is faithfully representing your brain at this moment versus, like, faithfully representing a brain that operates like a human brain. In other words, like, if you think about, like, if you're wired up a brain, like, just, you know, say you had a connectome that was, like, a human, like, taken from an individual and had all the characteristics of physiology that a brain would have,
00:26:32
Speaker
the way that we're developing these things, it wouldn't probably be the case that it would be like that person's physiology, right? It would be a generic physiology. That brain would be essentially starting, but it's interesting. It wouldn't be starting from nothing at all because if you think about the connectome itself or the connections themselves and particularly the synaptic connections and the weights of those, the sizes of those is a function of history and your learning and memory.
00:27:00
Speaker
It's all your developmental history. Right. First, I was thinking, well, you'd just really be having this naive brain. But no, you wouldn't. You wouldn't have a naive brain. Just by the choice of the instance of the model, it has a history. It's not arbitrary.
00:27:18
Speaker
Yeah. And so it may be more likely that it's easier to develop something than it is to copy it fully developed. It may be easier to, I guess. Grow it. It may be easier to model the process of development than it would be to take a full snapshot of your brain as it is now. Yeah. Well, I mean, and then, you know, in terms of
AI, Machine Learning, and Connectomics
00:27:47
Speaker
just, you know, the implications for artificial intelligence in general, you know, the more, you know, as we've said a couple of times before in the show, like there was a moment in history when it felt like we weren't learning that much from the brain about how to build machine learning and artificial intelligence. And more recently we've, you know, as computational power has increased and some new math has been introduced that
00:28:17
Speaker
Actually, you know, the basic idea of having layers of interconnected units, uh, that are either just feed forward or recursive, uh, is actually like a pretty decent framework for, for building learning systems in, in computers. And, uh, as we learn more about how the brains actually connect up and actually put some of that understanding into, into computer code.
00:28:47
Speaker
the implications for machine learning are really, I think, quite profound. Absolutely. Yeah, I think you're right to think of this as a two-way process that machine learning can inform can inform connectomics and vice versa, too. The more we learn about the brain, the more we learn about how well it's designed and why it's designed in the way that it is.
00:29:13
Speaker
Yeah, I mean, that's actually, as you say that now, I'm thinking, really, if you think about how much machine learning was, and machine vision was used to develop the Connectome for the fruit fly, and the fact that we're learning from the Connectomics and the brain, how to build better machine learning, you can start to see how the robots eventually take over, right? Because this- Yep, there's a pretty clear path.
00:29:43
Speaker
machine starts to learn about itself and, you know, from, from loop analysis of the brain and the human and other animals. And, uh, yeah, pretty soon it starts to, starts to take off. Yeah. I think I, I don't even think we need to comment about what could go wrong here. I think we'll leave that up to the imagination of the listener. Absolutely. Absolutely. Well, I think that's probably a pretty good place to stop. What do you think?
00:30:09
Speaker
Great. Yeah. I think hopefully everyone got a little information about the Connectome project and understands how exciting it is and we'll keep tabs on it for the future. Great. Yeah. Well, thank you all for listening. If you have comments, we'd love to get your feedback. I mean, the shorter form episode was something we developed from user feedback, from listener feedback, and we'd love to hear more. So you can contact us on Twitter at
00:30:38
Speaker
at nation cog, you could also contact, uh, raw for me directly on Twitter. Uh, I'm JL Hardy PhD and I am R O F L Nelson at us or, or, you know, DMS at, uh, on Twitter and let us know what you think and any feedback, anything you'd like to hear about. If you'd like to be on the show, uh, any of these things, let us know and, uh, we'll get back to you. Thanks for listening.