There’s a video that’s shown in almost every introductory neuroscience course. It doesn’t look like much—a bar of light shifting and rotating across a black screen while the background audio pops and crackles like the sound of a faraway fireworks show. Dry stuff, until you learn that the pops represent the firing of a single neuron in the brain of a cat, who is watching the bar move on the screen. When the bar reaches a specific location and lies at a particular angle, the popping explodes in a grand finale of frantic activity. The message is clear: This neuron really, really cares about that bar.
The experiment shown in the video was performed by David Hubel and Torsten Wiesel in the 1960s and helped scientists infer basic principles about how the visual system works. For decades since, neuroscientists have stuck thin, metal electrodes into the brains of mice, finches, and monkeys to spy on individual neurons and figure out what sets them off. There are neurons that respond to specific colors or shapes; or to particular locations in space or the direction of one’s head; or to whole faces or individual features.
As powerful an engine as single-cell analysis has proven, “Everybody always wanted more neurons,” says Anne Churchland, professor of neurobiology at the University of California, Los Angeles. Part of the reason was simple statistics: More observations are always better, no matter the experiment. But scientists also ran up against analytical walls when they looked at neurons one-by-one. In the prefrontal cortex, the region at the front of the brain that plays major roles in planning, decision-making, and social behavior, neurons respond to such a diversity of things—visual features, tasks, decisions—that researchers have been unable to assign them any particular role, at least individually. Even in the primary visual cortex, the area far to the back of the brain where Hubel and Wiesel made their recordings, only a fraction of neurons actually fire when the animal looks at oriented bars.
With Hubel and Wiesel’s techniques, looking at more than a handful of neurons at once was impossible. But engineers have pushed and pushed that capacity, culminating in the development of Neuropixels probes in 2017. One centimeter long and made of silicon, a single probe can listen to hundreds of neurons at once and is small enough that neuroscientists can stick several of them into an animal’s brain. At the Allen Institute, a nonprofit research institute started by Microsoft cofounder Paul Allen, they used six Neuropixels probes to record simultaneously from eight different regions of the mouse visual system. In August, the institute released data from 81 mice—comprising the activity of around 300,000 neurons. The data is freely available to any researchers who might want to use it.
As the largest data set of this kind ever collected—three times as big as the previous record holder—the release lets researchers observe enormous groups of neurons acting in concert. That unprecedented scale may unlock opportunities to understand parts of cognition that have previously evaded the scientific community’s grasp. “We want to understand how we think and see and make decisions,” says Shawn Olsen, an investigator at the Allen Institute who played a central role in the project. “And it just does not happen at the level of single neurons.”
The challenge now is figuring out just how to parse all that data. Gargantuan data sets aren’t easy to handle; even sharing and downloading them can be difficult. But as tricky as the analysis may prove, working with such data sets is eminently worth it to many researchers, because it lets them study the brain on its own terms.
Most PopularBusinessThe End of Airbnb in New York
Amanda Hoover
BusinessThis Is the True Scale of New York’s Airbnb Apocalypse
Amanda Hoover
CultureStarfield Will Be the Meme Game for Decades to Come
Will Bedingfield
GearThe 15 Best Electric Bikes for Every Kind of Ride
Adrienne So
To Hubel and Wiesel, the brain looked like an assembly line: groups of neurons, each specialized for a specific role, dividing and conquering each task. Show someone a red balloon, and neurons sensitive to red and circles will respond independently. But that approach never really suited how the brain actually functions—it is so densely wired up that no neuron is ever acting in isolation. “The brain is not looking at one neuron at a time,” says Stefano Fusi, professor of neuroscience at Columbia University. “Neurons, they’re looking at thousands of other neurons. So we should take the same perspective.”
Regions like the prefrontal cortex, where every neuron responds to a whole host of things, seem to operate much more like a workshop, in which each artisan has expertise in a wide range of tasks. Some might have particular talents for throwing raw clay, and others might be especially skilled at applying glaze—and when they work together, they can craft a variety of objects. This diversity is an advantage, and it’s likely essential to the complex problem-solving and reasoning skills at which humans so excel. (In a study of the prefrontal cortex, Fusi demonstrated that, when neuronal populations show a rich diversity of responses to different situations, monkeys tend to perform better on a memory task.) Neuron populations that are highly specialized, on the other hand, are inflexible, much like an assembly line: They can only accomplish so many different things.
Assembly lines, though, are extremely easy to understand. Each step in the process can be examined independently to figure out precisely how it contributes to the overall product. But the artisans in a highly interactive workshop can’t be viewed in isolation, and neither can neurons in regions like the prefrontal cortex. And these collective activity patterns are too complicated for humans to grasp without the aid of mathematical tools. “It’s not something that you can visualize,” Fusi says.
Most PopularBusinessThe End of Airbnb in New York
Amanda Hoover
BusinessThis Is the True Scale of New York’s Airbnb Apocalypse
Amanda Hoover
CultureStarfield Will Be the Meme Game for Decades to Come
Will Bedingfield
GearThe 15 Best Electric Bikes for Every Kind of Ride
Adrienne So
So neuroscientists use an approach called “dimensionality reduction” to make such visualization possible—they take data from thousands of neurons and, by applying clever techniques from linear algebra, describe their activities using just a few variables. This is just what psychologists did in the 1990s to define their five major domains of human personality: openness, agreeableness, conscientiousness, extroversion, and neuroticism. Just by knowing how an individual scored on those five traits, they found, they could effectively predict how that person would answer hundreds of questions on a personality test.
But the variables extracted from neural data can’t be expressed in a single word like “openness.” They are more like motifs, patterns of activity that span whole neural populations. A few of these motifs can define the axes of a plot, wherein every point represents a different combination of those motifs—its own unique activity profile.
There are downsides to reducing data from thousands of neurons down to just a few variables. Just like taking a 2D image of a 3D cityscape renders some buildings totally invisible, cramming a complex set of neuronal data down into only a few dimensions eliminates a great deal of detail. But working in a few dimensions is much more manageable than examining thousands of individual neurons at once. Scientists can plot evolving activity patterns on the axes defined by the motifs to watch how the neurons’ behavior changes over time. This approach has proven especially fruitful in the motor cortex, a region where confusing, unpredictable single-neuron responses had long flummoxed researchers. Viewed collectively, however, the neurons trace regular, often circular trajectories. Features of these trajectories correlate with particular aspects of movement—their location, for example, is related to speed.
Olsen says he expects that scientists will use dimensionality reduction to extract interpretable patterns from the complex data. “We can’t go neuron by neuron,” he says. “We need statistical tools, machine learning tools, that can help us find structure in big data.”
But this vein of research is still in its early days, and scientists struggle to agree on what the patterns and trajectories mean. “People fight all the time about whether these things are factual,” says John Krakauer, professor of neurology and neuroscience at Johns Hopkins University. “Are they real? Can they be interpreted as easily [as single-neuron responses]? They don’t feel as grounded and concrete.”
Bringing these trajectories down to earth will require developing new analytical tools, says Churchland—a task that will surely be facilitated by the availability of large-scale data sets like the Allen Institute’s. And the unique capacities of the institute, with its deep pockets and huge research staff, will enable it to produce greater masses of data to test those tools. The institute, Olsen says, functions like an astronomical observatory—no single lab could pay for its technologies, but the entire scientific community benefits from, and contributes to, its experimental capabilities.
Currently, he says, the Allen Institute is working on piloting a system where scientists from across the research community can suggest what sorts of stimuli animals should be shown, and what sorts of tasks they should be doing, while thousands of their neurons are being recorded. As recording capacities continue to increase, researchers are working to devise richer and more realistic experimental paradigms, to observe how neurons respond to the sorts of real-world, challenging tasks that push their collective capabilities. “If we really want to understand the brain, we cannot keep just showing oriented bars to the cortex,” Fusi says. “We really need to move on.”