What are you guys most excited about with regards to the advancing technologies and methodologies involved in the various connectome projects that have surfaced in the past few years?
For those who aren't aware, a good rundown of the connectome is that it's an effort to take technology that can slice a brain super thin and use electron microscopy to take a snapshot of each layer, and then be able to piece all of these together in a 3D model to model every neuron in a given brain. The hope then, is to be able to compile a database of brains for various species to then hopefully be able to figure out what genetically codes for what patterns, as well as elucidate what circuits could potentially lead to what behaviors/phenotypes.
The best analogue, project-wise would probably be the genome project. I imagine similar analyses would be able to be performed as well- That's what I'm most excited about. Integration of the connectome in bioinformatic analysis, and have a sort of a map of the brain that you can point to regions of, or even individual circuits and divine what they do. The further applications into using these patterns in designing artificial intelligence, to me, seems also particularly interesting.
So what do you guys think about the Connectome- are you involved in neurobio and have your own concerns/hopes/stories about how it's being TOO hyped up?
Assuming we have a fully fledged model of the neurons and connections in the average brain, how will neuroscientists use this when conducting experiments? It seems that, given the insane amount of connections in the brain, one would have great difficulty interpreting the connectome.
Josiah Kelly
That's one of the biggest issues. There are a lot of similarities with this and the genome- an insane amount of information that has to be interpreted to figure out what functional parts are what. The major difference, as is evident, is that with DNA we already had the 'key' to which regions are promoters, and the key for which codons are for what amino acids/where the start sequences are. Given that information, it was relatively easy to piece together genes, and work from there to continue to analyze what happens when there are mutations in genes.
However, if we look at how we studied genes before we could piece together a genome, we might find some clues as to how we can study the connectome The typical way we used to study genes is as so:
>Take variants of flies/mice with genetic markers that tell you where crossing over in meiosis may have occurred >Induce Mutagenesis with transposons in order to create a single generation that creates a high proportion of mutations in their sex cells >screen mutants for the mutation you're looking for >use SNPchips or another assay method to figure out the exact location of the gene by crossover events, develop primers to then PCR the gene, analyze the product, then you know say that X gene controls for the ability to smell, as an example.
While the genetic methods aren't applicable, the dogma behind using some kind of method to break a neuronal circuit, and then seeing what changes between a subject that has a near-identical brain. Problems of course are that we'll need to use clones in almost exactly identical life situations to create brains that are as close to identical as possible, and trying to modify single neurons is very difficult.
However the other thought, and one that's exciting for its implications, is being able to take the connectome and adapt it into a computer modeled logic circuit. Whether this constitutes an AI or could think is to be seen- it could be used to test things, however.
Lucas Wright
Huge scam. C. Elegans connectome has been around for decades and neuroscientists still don't understand it completely. That has 107 neurons. Imagine trying to understand billions.
Henry Jenkins
>107 Sorry, *302.
Kevin Ross
What excites me the most isn't exactly the premise alone- it's the higher resolution technology that allows for more complex organisms, as well as the computational power and methods made available thanks to advances in bioinformatics.
However I do understand the point- while we had the human genome almost 15 years ago now, we didn't understand (and still don't) understand everything about it. Still, it's much more useful than what a connectome might do for us now.
Jaxon Young
>take the connectome and adapt it into a computer modeled logic circuit Totally preposterous. There's more to the function of the nervous system than structure alone. A map of neurons won't tell you if a particular synaptic connection is excitatory or inhibitory, and it also won't account for cell-types and other intra/intercellular phenomenon that we don't understand or are even aware of yet.
I mean, look at this paper published just this month cell.com/cell/fulltext/S0092-8674(17)31504-0 A whole mechanism for intercellular information transfer between neurons that was completely unknown to scientists.
Hudson Peterson
Oh yeah I don't mean to say that we can 1:1 adapt neurons to a computer circuit right now, but I think the broad hope of scientists is that by analyzing the structure of synapses, it might be possible to determine what kind of synapse it is- and hopefully the magnitude required fro excitation. We know a lot now about how neurons work- like in the article you linked there's still a lot left, but the idea is to be able to take stuff like the connectome and even use information about patterns of connections and be able to help circumstantially determine the functions of every part.
Dominic Ortiz
Good thread. Bump.
Dylan Richardson
I don't exactly understand the importance of simulating a 3d human brain. There is a lot of excess crap that can be reduced away completely which remains perfectly true without the spatial definition.
There is advantages to it when we talk about building interfaces into the brain, but as far as machine learning goes? Nothing of real value to be found here.
Landon Hill
I think the use more comes from adapting individual patterns and circuits that might be commonly seen in the brain to their logic circuit counterparts. You're right that a lot of it isn't useful- but through the model of "break it to see what changes", there might be a way to learn how some circuits work more mechanistically.
Basically, if we learn how the brain learns, and the kinds of patterns seen in neurons that facilitate this phenomenon, then we should be able to adapt the mechanisms to some kind of digital format. Though I'd definitely see the argument that cellular and neuronal development has a lot more to assist in this kind of study than studying how the brain is set up itself.
Jackson Gonzalez
What makes you think the brain can be compared to logic circuits. I dont think you can infer function from patterns apriori. Its too complicated and brain function is too fluid. I think actually that from connectomic perspective the idea of parts having partocular functions goes straight out the window.
Not necessarily true as functional networks in the brain work on a fine scale. Think an idea is that greater connectomic accuracy means better simulations that you can use to develop models of brain function. Can compare these to real data given imaging technology improves.
Isaiah Hernandez
I think you're right that with highly complex brains we won't know for a long time, but with increases in computational power and the overall ability to figure out what might be changed in a deletion experiment by comparing connectomes could yield incredibly valuable. The brain, in essence, operates much like a logic circuit- the issue of gradients that arises from the molecular dynamics of how neurons fire can't be ignored, but the storage of information ultimately is based on ON/OFF signals without too much (as far as we know) in the way of degrees- an action potential can be fired, or not fired, never any state in between is the essential dogma- the only place this disagrees with how we traditionally look at a logic circuit is that various inputs and types of neurotransmitters could lead to reaching action potential based differently on what is received, but on a large scale this can be computed by what we know of the cell type, and morphological classification can go a long way in trying to determine what neurotransmitter and in what quantity a neurotransmitter will be released from a neuron when an action potential has fired.
I think you're right that there's a bottleneck ahead of any analysis due to the complexity of the human brain. I think part of the dogma is that while looking at individual neuronal circuits is fruitless, and will be for a long time, that certain macrostructures or repeated circuits or patterns might be apparent and able to be studied further, but the two big applications on my mind for the immediate future are:
1) Cloned mice connectomes taken at different stages in development of the brain, given the exact same stimuli, and cloned mice connectomes taken at the same stage of development given different stimuli, correcting for random differentiated patterns via statistical overview of where the differences occur normally, and 2) Modeling simpler brains than mammals, and being able to study individual circuits for
Jack James
An addendum- I'm a geneticist, so my interest in the connectome isn't based on a neurobio perspective. Apologies if I'm mistaken on any particulars, or even wrong in my thinking- I'd love to have a better understanding if I'm wrong about anything in particular or if there's a better way of looking at it.
Daniel Wilson
I'm pretty excited actually. I want to study the brain in more detail and figure out how various systems work, especially those related to sensory inputs.
Ethan Thompson
We cduse deletion in connectome in the contexts of model tasks/environments to predict specific behaviours or cognitive phenomena though i can imagine. Yes i agree but just saying dont think can say a particular area has an objective underlying function.
I think certain cognitive phenomena maybe describable with logic gates but not sure the brain works like these necessarily. Ultimately information isnt actually carried in single spikes. Its carried in more complex relations among groups of neurons. Also the fact that neurons carry multiple inputs and outputs i think stops this. A neuron may fire on or off but its function is dependent on its outputs and if it has many outputs to different areas not sure the idea of a logic gate carries.
Yeah true about microstructures. Whats your idea about the spikes.
Brandon Rodriguez
For instance a logic gate cdnt explain how neurons code for stimuli with continuous tuning curves.
Eli James
This seems to be focused on the central nervous system. What about peripheral?
Back when I took anatomy we used to do exercises like "what would happen if nerve X got sectioned at Y point?" and I always wondered if this reasoning could be automated. I don't know about you guys but the peripheral nervous system has always looked like a directed graph to me. I tried to compile a list of nerves and their relations into a graph but it was a huge undertaking and the idea kind of died out after I finished my class. Do you guys know if this exists? Something like a nerve database.
Jayden Campbell
I see what you mean. I think it's more like we might find novel patterns that have some kind of common 'function' as far as circuits are concerned.
Let's go to logic, and simplify the model such that each neuronal synapse has some positive (excitatory) or negative (inhibitory) signal value based on the action potential sent. Then, you can say that each neuron consists of "If (excitatory+inhibatory)>[Threshold], then fire AP", where the threshold is the AP threshold which of course depends somewhat on the morphology of the neuron, and we ignore some finer points like the drop in AP signal over distance, or interference, then it's (theoretically) possible to follow each neuron attached, the idea is that to create certain phenomena like continuous turning curves, there might be some combination of neurons, or the overall circuit which gives rise to it.
However, the model of logic gate falls apart with additional biological factors in play, so I could see where maybe it doesn't necessarily work that way. I would not be surprised however if some elements may be found- repeated developed patterns are so common in biology because they are simple and genetically cheap, consisting of just copypastes and edits.
That is a fantastic question. After digging around, it seems like those kinds of questions are tackled using not a connectome, but what is called a "Neurome", which is the collection of all neurons (maybe not including the brain?) in the body. larrywswanson.com/?page_id=1148 for a rat neurome project, and ieeexplore.ieee.org/document/6811580/?reload=true describes the dogma a bit and outlays an experiment on flies.
That reasoning is exactly what makes systems biology not only so potent, but interesting, and promising as hell considering how the descriptive power directly scales with automation and computation power.
Jeremiah Brown
I think the neurons themselves act as logic gates but i dont like using logic gate to describe how the brain works. Think it doesnt get to the heart of what the brain does. I doubt the functional divisions of the brain act like logic gates. And i guess also you get different types of neurons with different types of connections (e.g. driving, modulatory) which might be underdescribed by just looking at the brain as logic gates.
Aaron Gray
There's no equivalent project for humans?
I remember reading something years ago about some how university modeled the human body as linked data structures that computers could use as data but I can't find it. I'm not sure if it contains the nerves, I'm guessing it does since neuroanatomy is part of medical school. Even without detailed neurome information it would help a lot.
