The Monumental Challenge of Mapping the Human Brain

What’s keeping us from modeling a human brain, neuron by neuron, on a computer?

There are a lot of projects that are trying to do just this, but you have to realize, the brain is by far the most complex system scientists have ever tried to understand. A human brain is composed of roughly 100 billion neurons, and those form a very intricate but also incredibly dynamic network.

One issue is that modeling these neurons requires very computationally expensive math. Every single neuron has to be defined by a number of different variables, and each variable requires a set of differential equations to run. And don’t forget, we’re talking about having to individually compute all those equations for all these variables for 100 billion neurons all at the same time.

Even if we ignored most of the complexity in the brain—[if we put] aside the fact that we don’t have just one type of neuron, or if we assumed all the neurons are healthy, and so on—there isn’t a supercomputer on earth that could create this type of theoretical model for anything close to even a million neurons over any realistic time period.

So is the barrier just about computing power?

Well, it’s partially a computational barrier. And for this issue, luckily, here we have companies like IBM that are spending millions and millions of dollars just trying to give the world more computing power.

But a big part of the problem is also our biological understanding. In simple terms, we just don’t understand all the variables that affect these neurons, and it’s impossible to accurately model what you simply don’t understand. Right now, with our best computer models, we can successfully model up to (roughly) 10,000 neurons, and that takes a very long time. But when you try to add more and more neurons, the system just breaks down and no longer acts anything like the neurons in the brain. Even though a lot of people try to just tweak these models to more closely match real neurons, the more you tweak, the further you get from any real understanding.

Realistically, we’re a pretty long way away from understanding the biology of how these neurons connect and interact enough for an accurate model. I mean, I could be completely wrong, and tomorrow someone could find some super important breakthrough, but if you look at the average rate of progress in this field, we’re still going to have plenty to learn over the next 20 or 30 years at least.

You’re working with a technology that might help us circumvent this whole issue: microelectrode array dishes. What are they?

An MEA dish is actually quite simple. It’s essentially a cell-culture dish with a grid of electrodes on the bottom. You can plate these dishes with, for example, clumps of 100,000 living neurons—which can live for months—where the dish’s electrodes are in direct contact with the bottom layer of neurons.

With those electrodes, you can record and listen to the electrical activity of those neurons while they’re growing. Or, by prodding them with electrical stimulation, you can record how they react and evolve and change over time. Our ultimate goal with this technology is to use data to understand how these networks of neurons work, and combine blocks of them together to make a realistic brain model.

What’s important here is that . . . we can treat our clump of 100,000 neurons like a black box. We can ignore all the biology of the single neurons that we still don’t understand, because it almost doesn’t matter—we’re just focusing on how these larger networks function. And in this way, MEA dishes will not only give us a more realistic approach to building brain models but—because we’re not relying on assumptions or approximations about individual neurons—also a more biologically oriented one.

Image may be NSFW.
Clik here to view.

But MEA dishes aren’t without their own limitations, right?

Oh, of course. For one, there’s the issue of resolution. The number of electrodes on each MEA dish varies between 60 and 250, and each of those electrodes picks up activity from many neurons. And when we’re dealing with something like 100,000 neurons, you can see that the number of sensors is much, much smaller than the number of active neurons. This means that, of course, this technology isn’t able to record and identify the activity of single neurons. But keep in mind, with these dishes we’re not concerned with what the individual neurons are doing, we’re looking at the network level.

With these dishes we’re only really investigating the neuron network activity in a 2D fashion. And in the brain, our neurons form intricate 3D connections and structures, which is very important. This is admittedly a big issue, but MEA dishes with 3D capabilities will soon be on the market.

You’ve said before that the approach of these MEA dishes is important for more than just computer models. Why is that?

It’s very true. In neuroscience—make no mistake—we’re still trying to understand very, very basic questions about how the brain works. And to this end, much of the way we’ve been studying the brain has really focused on either a top-down or bottom-up approach: either trying to study the brain by looking at it as a single entity, or trying to understand the neurons and then combining them. But both approaches have severe limitations. The brain is just too big and complex to understand without looking at the pieces that make it up, and as we’ve already discussed, modeling the brain based on neurons is still far beyond our grasp.

While the large majority of research is focused on studying the individual neurons, the MEA dish represents a kind of compromise between the two approaches. And the technology offers researchers an almost unprecedented ability to study these midlevel neuron networks in confined and highly controlled experimental conditions. For example, you can apply medicines to your cluster of neurons, or change any number of variables like temperature or disease, and see how that affects the overall network connections.

Don’t get me wrong: MEA dishes is not the only technology available for researchers to study this midlevel in neuroscience. There are techniques like optogenetics that are even more widespread. But this midlevel is not easy to work in; you really have to combine computer techniques with an understanding of the biology, and that takes a lot of specialized expertise. That’s part of the reason there’s so few people involved in this type of research.

I’m sure that if we’re ever going to fully understand the brain, eventually we’ll need to converge what we know about the brain as a whole, and what we know about neurons. When those two sides of the spectrum meet, they’ll do so in this middle ground.