Unprecedented Simultaneous Recording of the Activity of One Million Neurons Answers Fundamental Question of Neuroscience

[Introversion]

Loading…

scitechdaily.com

The mammalian brain is a web of densely interconnected neurons, yet one of the mysteries in neuroscience is how tools that capture relatively few components of brain activity have allowed scientists to predict behavior in mice. It is hard to believe that much of the brain’s complexity is irrelevant background noise.

“We wondered why such a redundant and metabolically costly scheme would have evolved,” says Rockefeller’s Alipasha Vaziri.

Now, a new study in Neuron—which presents an unprecedented simultaneous recording of the activity of one million neurons in mice—offers a surprising answer to this fundamental question: technological limitations have misled us, and there’s far more to the brain than once thought.

“Previous assumptions about the true dimensionality of the brain dynamics might have been due to the lack of ability to record from a sufficiently large number of neurons,” Vaziri says.

Using a custom technique developed in the Vaziri lab, the researchers discovered that more than 90 percent of the dimensions they observed in neural activity (independent components that one needs in order to describe the observed neuronal dynamics that contain signals that are different from noise) were not connected to any spontaneous movements or sensory inputs in the mice studied. Thousands of these dimensions, containing more than half of the cumulative neural activity of the mice, were spread across the brain in space and time, without forming distinct clusters in any one region and ranging in time from minutes to less than seconds.

The mouse was clearly using this thrum of pervasive, continuous activity for some purpose. But for what? “We still don’t know, but it’s definitely a signal that is distinct from noise,” Vaziri says. “It could offer a window into to a variety of complex internal states or neurocomputation.”

…

 

[pdblake]

There's a Douglas Adams joke in there somewhere. Probably 42 of them.

 

[Introversion]

There's a Douglas Adams joke in there somewhere. Probably 42 of them.

“Brain the size of a planet, most of which apparently actually does something.”

 

[Woollybear]

I literally said something about Life, the Universe, and Everything before popping onto this thread.

It's all connected, man.

 

[Meg]

And over here in this corner, the cognitive scientists are laughing in crazed frustration. Don't get me wrong, this technique is amazing and will probably lead to many important advances in the coming years. But:

more than 90 percent of the dimensions they observed [. . .] were not connected to any spontaneous movements or sensory inputs

Their big breakthrough is that not everything is stimulus and response? Dudes, the 1960's called to tell you the behaviorists lost the fight.

 

[dickson]

Loading…

scitechdaily.com

The article gives only the vaguest description of the analysis that classified the variance in neuronal activity. The language used makes it sound like a form of principal components, but I’m sure it’s something more up-to-date, given the size if the data set.

I tried looking at the article, for which an extended abstract is publicly accessible. Neuronal activity in the mice was recorded during a variety of activities. Half the total variance is contained in 16 components, but the remainder is “distributed in a cortical and granular fashion” (my paraphrase from memory). That's the part that has the researchers puzzled and excited. Me too.

It’s been years since I studied neurobiology, but I felt brain function need not be localized in the sense of contents of a workshop parts cabinet, as “function x resides in Broca’s region” when the chief evidence for localization came from studies of brain damage or surgical intervention, or fMRI in recent decades.

If you take out a router, that destroys the operation of an LAN. But the operation of the LAN doesn’t reside in the router. It passes through it. Not for nothing is a big chunk of a human brain’s mass (white matter, I’m looking at you!) made of myelinated axons ferrying action potential hither to thither.

I’m sure this idea isn’t original to me, but I don’t recall seeing it discussed in quite those terms. Mind, it’s been a couple of decades. I sought a way to bring differential game theory to the table, specifically parabolic DGT. Technical difficulties arose. My DGT studies mutated to the more tractable problem of Nash equillibria in satellite warfare, which at least had the virtue of being marginally relevant to my then-employment.

 

[Meg]

Yeaaaaaah, I kinda picked on their "sensory input & behavior" thing because it was low-hanging fruit for snark. But I got me some complaints about their other points: that information processing is widely distributed, and operates over a wide range of time scales.

brain function localization need not be localized in the sense of contents of a workshop parts cabinet, as “function x resides in Broca’s region” when the chief evidence for localization came from studies of brain damage or surgical intervention.

Indeed, and people who study functional brain activity have been wrestling with this issue at least since I was in grad school in the '90's. Cortex-wide involvement in cognitive activities can be seen from ERP recording in humans, and is even implicit in the methodology of functional brain imaging. (Which looks at two nearly identical tasks and subtracts the activity of one from the other to see where the differences are. Otherwise, all you see is the entire brain being active for every task.)

Their finding that these neuronal populations operate on a wide range of time scales is also something that has long been known. (See the large literature on endogenous oscillators in the brain.)

To be fair, the authors themselves are less breathless about it than the popular report. The last line of their abstract is:

The activity patterns underlying these higher dimensions are fine grained and cortex wide, highlighting that large-scale, cellular-resolution recording is required to uncover the full substrates of neuronal computations.

But they are still presenting ideas as new that aren't, and their (definitely new and cool) method of showing these (already widely accepted) principles doesn't yet yield new insights into how all that activity is structured, what it means. It's unclear what those "full substrates of neuronal computations" can tell us, beyond "cognitive activity involves an awful lot of neurons, and it's complicated."

It's quite possible this will turn out to be a case of using the wrong level of analysis, like using massive big-data description of all a frog's cells to try to understand frog anatomy, or looking at individual pixels in a picture to try to understand what it's a picture of. OR, this method could turn out to be useful for identifying functional patterns of processing . . . but it hasn't done that yet.

I don't actually blame them. This kind of over-promising and over-hyping is pretty much a requirement to have a career in the cognitive sciences these days. (Yes I am old and bitter and glad I'm out. )

 

[pdblake]

“Brain the size of a planet, most of which apparently actually does something.”

I was thinking more along the lines of the mice experimenting on the scientists, not the other way around.

We're the AI

 

[dickson]

But they are still presenting ideas as new that aren't, and their (definitely new and cool) method of showing these (already widely accepted) principles doesn't yet yield new insights into how all that activity is structured, what it means. It's unclear what those "full substrates of neuronal computations" can tell us, beyond "cognitive activity involves an awful lot of neurons, and it's complicated."

The reigning paradigm since Sherrington, if not earlier.