September 11, 2013 | 5
As Albert Einstein famously said, “No problem can be solved from the same level of consciousness that created it.”
The history of science is littered with so-called “intractable” problems that researchers later cracked wide open using techniques their ancestors could hardly imagine. Biologists in the 1950s looked at the staggeringly complex (and beautiful) three-dimensional shapes into which proteins fold and declared that a reliably predictive mathematical model of these convolutions might be unachievable in our lifetimes. But over the past few years, folks with home computers have joined forces to crack many longstanding protein-folding problems using the online game FoldIt.
Instead of relying on the number-crunching power of a single supercomputer or network, crowdsourced games like FoldIt translate vast and complex data sets into simple online interfaces that anyone can learn to operate. The crowdsourced astronomy game Galaxy Zoo also depends on an army of “citizen scientists” for classification of stars hundreds of light years away; while Google built its image search technology on an image-labeling game. In fact, every time you “verify your humanity” on a web form by typing out nonsensical reCAPTCHA text, you’re actually helping Google transcribe books from the world’s libraries into a digital format.
And now, a worldwide team of neuroscience researchers have begun using this crowdsource approach to crack open one of the greatest problems in any scientific field: The construction of a complete wiring diagram for a mammalian brain.
Complexity and simplification
Neuroscientists often compare the task of mapping the brain’s wiring to that of untangling all the cords piled up beneath your desk. It’s just that a human brain contains upward of 84 billion “cords” – nerve cells known as neurons – every one of them sporting multiple “plugs” – connections known as synapses – adding up to as many as 100 trillion (that’s “trillion” with a “t”) interconnections in a single brain.
Throw in the fact that many of these connections are shifting and changing every day, and you can begin to see why some scientists claim that a complete model of a human connectome – a functional map of every neural connection in a human nervous system – is little more than a sci-fi dream right now.
Still, a brain is a finite physical object. Its complexity is enormous, but not ineffable. It can be subdivided into lobes, areas and layers. And just as each level of the ocean and forest is home to its own unique range of life, each layer of the cortex is home mainly to a limited range of specific neuron types, each of which cooperates and competes with others in specific ways.
“If I just told you, ‘There are a lot of trees in a jungle,’ that’d be true, but it’d be a very crude description, because trees come in many different species,” says Sebastian Seung, a neuroscientist at Massachusetts Institute of Technology. “In the same way, neurons come in many cell types, and one of the important tasks in neuroscience today is to identify and enumerate all these different types of neurons in the brain. Nobody knows how many there are.”
This neuron jungle might sound like yet another daunting problem – but in fact, Seung thinks it may provide a simple yet highly accurate way of modeling brain function down at the cellular level. It’s a bit like saying, “We’ll never be able to model the workings of every individual tree in the whole forest – but we don’t have to. All we’ve got to do is understand the behavior of each species.”
Even this is no small task; but Seung and his team have found an innovative way to tackle it.
Solving the unsolvable
Mapping a human connectome demands some conceptual steps up from traditional brain-scanning techniques like fMRI and EEG. So this year, Seung and his colleagues at MIT, along with Moritz Helmstaedter and Winfried Denk at Germany’s Max-Planck Institute for Medical Research, set to work on the problem from a new angle.
The researchers chose a very small chunk of nerve tissue – a tiny slice of retina, to be exact – and looked for an efficient way to tease apart its wiring while preserving its structure. “If we take human colorings or tracings of neurons, and we train the computer to emulate them, the computer still makes mistakes,” Seung explains. “So somehow we have to correct the mistakes of the computer – we have to combine human intelligence and artificial intelligence in order to solve this problem.”
The new approach works like this: First, a human operator works through digital slice after slice of brain matter, using computerized tools to draw a “skeleton” of each neuron he or she notices in each slice – similarly to the way computer animators start by drawing a stick figure of an animated character.
Once humans have drawn in these neuronal skeletons, an automated computer algorithm builds out a 3D model of each neuron’s three-dimensional shape. “If people had to color in the full three-dimensional shape of a neuron, instead of just drawing the skeleton, each neuron would take ten to 100 times longer, and the cost of our study could’ve been has high as $10 million,” Seung says. But using this new technique, the international team was able to complete the project at a much lower budget, in a matter of mere months.
In a recent paper published in the journal Nature, Seung and his colleagues at the Max-Planck Institute unveiled their first major success using this approach: A connectomic map of every neuron in a tiny patch of mouse retina. Along the way, the team discovered several new types of neurons, and generated a neuronal wiring map of unprecedented scale and complexity.
Exciting as this retina-mapping project has been, Seung says, it’s only the beginning of his team’s quest for the ultimate goal: A complete map of a human connectome. And to get there, they’re going to need your help.
Crowdsourcing a revolution
Last year, Seung’s lab rolled out a game called EyeWire. The principle is simple: After a few practice rounds, anyone with a computer and an Internet connection can help researchers map the shapes of actual neurons. In the game, these neurons come with their “skeletons” pre-drawn, and players take on the task of coloring in the neurons’ 3D shapes.
Computers and neuroscience experts then compare the results; and if all goes well, the result is a map of a synapse that’s never been mapped before – a brand-new brain discovery made by people just like you. “That makes it a multiplayer game,” Seung says. “People are rewarded for giving the same answers as others, and that’s how they learn. That’s how they’re incentivized to be accurate – and it also makes the game inherently social.”
Seung’s lab is currently putting together the first journal paper documenting brand-new discoveries made with EyeWire. The researchers aim to use this wealth of data to crack another puzzle: How groups of neurons detect motion. “People want us neuroscientists to explain consciousness,” Seung says; “they want us to cure autism. But in fact, for more than 60 years, neuroscientists have been unable to explain how it is that neurons in the retina can detect motion.”
It’s a sobering fact, but one with a silver lining: Thanks to citizen-scientists working through EyeWire puzzles, Seung thinks he and his team have hit on at least one testable explanation of this longstanding neuroscience mystery. To find out what they’ve discovered, though, we’ll have to wait for the upcoming paper.
EyeWire players have already mapped a large number of neurons – and mapping new ones every day – but it’s still clear that long years of work lie between us and our goal of a complete human connectome. The data gathered so far may seem like drop of water in the ocean, but projects like EyeWire are actually proofs-of-concept for a technique of extraordinary power. “It’s clear that is now possible – not easy, but entirely possible – to start mapping mammalian connectomes at the cellular level,” Seung says. “The technology wasn’t there just a few years ago, but technology is advancing rapidly.”
Images: Sebastian Seung
12 Digital Issues + 4 Years of Archive Access just $19.99X