Whole human brain mapped in 3D

Roughly 9.09495TB for anybody interested

Yeah but Ten Trillion Bytes sounds a lot bigger than 9 TB.

Apologies (one of Nature's editors here). Believe it or not it was in terabytes when we sent the copy to our sub-editors (ie copy editors) and they changed it because they thought our readers wouldn't get it. We're as bemused as you.

>what is the possibility that something like this could be used to generate an artificial neural network which imitates, even crudely, the functions of the human brain?

If possible, this would open a huge ethical can of worms: how can we tell such a simulation is not conscious? Would deleting its runtime data after however many simulation cycles be tantamount to murder?

Edit: I am not saying that the above concerns should stop us from working towards developing human brain simulations since the potential benefits of those are just too great. Rather, it is something we have to have in mind as they get more complex and closer to the real thing.

Not a neuroscientist but final year med with a strong interest in neuro-

This will certainly be useful for attempts to recreate the functionality of a brain in silico - big research efforts such as the blue brain project [1] have probably been using even early forms of this data.

Having said that, in some ways it won't provide the useful information that will enable the functionality to be restored to the connections.

heres why.

Current efforts involving simulated brains rely both on older, less precise versions of what this nature article talks about - let's call that the roadmap, as it is a map of where all the neurons sit and where they connect to. This project drastically increases the resolution of this roadmap.

However the additional information needed is what traffic is carried by each road. The brain as we understand it today consists not just of the connections between the neurons but also that each neuron, and each different network of the brain, secretes it's own neurotransmitter. Because some of these neurotransmitters are depolarising and some are inhibitory, and they connect widely across the brain and are much more closely related to the actual information processing, we need to have this information before we will be able to better approximate a brain.

In all likelihood this is information gathering is probably underway right now. From what I understand the 'BigBrain' project aims to do for the human brain what the Allen Brain atlas did for mouse brains, that is not only provide a map of the connections but also stain for gene expression and (from memory) neurotransmitter presence.

With this added information we will be closer to our goal of getting skynet sentient.

[1] http://bluebrain.epfl.ch/

Depends. First, a biological network is not just connections of neurons, it's weighted connections of neurons. And even saying that is a huge approximation.

It's not just about mapping the topology of the brain, just having a neural network of 1e10 vertices and 1e12 links will not give you an artificial intelligence. It's just not that simple. The connectome paradigm is not sufficient. It is hip because of computer scientists using artificial neuron networks to achieve awesome stuff. But that doesn't mean that the complexity of the human brain only resides in the number of neurons and interactions. It's not guaranteed that we are getting closer of the goal just by adding billions of neurons and connections.

I find this wannabe holistic approach a bit confusing as it is horizontal whereas I am convinced that what we need is a vertical approach : up from graph topology down to an accurate biophysical understanding of synapses and information transmission, all across multiple scales, from the angström up to microns.

That's the key : multiscale modeling, not single-huge-scale modeling.

Not a neuroscientist, but used to work in the field, and my first idea was "well by cutting it up, they also cut all connections". And those connections are far more important for understanding how the brain works than seeing where an individual neuron is physically located. Unless off course they can somehow get the connections out of it.

Systems Neuroscientist here. I've only skimmed this paper so far, but it seems as though this dataset is limited to studying cytoarchitecture (what types of neurons are in what regions of the brain) as well as some tracing of major fiber pathways. This does not represent the successful conclusion to connectomics efforts to map every neuron and every connection in the brain (see Sebastian Seung's work and the eyewire project [1]). It's hard to tell exactly how useful this dataset and model are.

There are other exciting developments that map not just neuroanatomical structure but also the distribution of proteins and gene expression [2],[3]

[1] http://eyewire.org/ [2] http://med.stanford.edu/ism/2013/april/clarity.html [3] http://smithlab.stanford.edu/Smithlab/Array_Tomography.html

Another "not a neuroscientist" - but I do have a slightly-more-than-layperson knowledge and interest in the area, and a degree in AI if that matters.

"I'd love for a neuroscientist to weigh in on this - what is the possibility that something like this could be used to generate an artificial neural network which imitates, even crudely, the functions of the human brain?"

Pretty much zero. Because it really just focussed on the organisation of the neurons, and ignores the complexity of the neurons themselves and the chemical/electrical/etc. communication between 'em. Neurons themselves are pretty complex beasts, and we're still learning things about them and the level of "processing" that is happening within neurons, as opposed to between 'em.

For example as recently as ten years ago everybody knew that most of the computation was implicit in the neural connectivity of the synapses. We now know that there is significant computation within individual neurons - in the dendrites of all things (previously thought to be pretty much passive carriers of output from other cells - just wires basically).

(See http://www.annualreviews.org/doi/abs/10.1146/annurev.neuro.2... ).

Don't get me wrong - it's useful work. But nowhere close to the level of detail needed to run people-sims.