When light enters the fly’s compound eye, hexagonally arranged photoreceptors send electrical signals through a complex neural network that allows the fly to sense movement. Siwanowicz, I. & Loesche, F./HHMI Janelia Research Campus, Lappalainen, JK / University of Tübingen Hide caption
Toggle caption
Siwanowicz, I. & Loesch, F./HHMI Janelia Research Campus, Lappalainen, J.K./University of Tübingen
Scientists have created a virtual brain network that can predict the behavior of individual neurons in a living brain.
The model is based on the fruit fly visual system, allowing scientists to quickly test ideas on the computer before spending weeks or months conducting experiments with real flies or other lab animals.
“Now we can start to infer how the fly brain works before anyone has done any experimental measurements,” says group leader Srinivasan Turaga of the Janelia Research Campus, part of the Howard Hughes Medical Institute (HHMI).
The approach, described in the journal Nature, also suggests that power-hungry artificial intelligence systems like ChatGPT could require far less energy if they used some of the computational strategies found in living brains.
Drosophila brains are “small and energy-efficient,” says Jakob Macke, a professor at the University of Tübingen and author of the study, “and they can do a lot of calculations: fly, walk, detect predators, mate, and survive with just 100,000 neurons.”
In contrast, AI systems typically require computers containing tens of billions of transistors. Worldwide, these systems would consume as much power as a small country.
“When we think about AI today, making these systems more power efficient is a top priority,” said Ben Crowley, a computational neuroscientist at Cold Spring Harbor Laboratory who was not involved in the research.
Borrowing strategies from fruit fly brains may be one way to achieve this, he says.
Biologically Based Models
This virtual brain network was made possible through more than a decade of intensive research into the organization and structure of the Drosophila brain.
Much of this research has been conducted or funded by HHMI, which now has a map of every neuron and every connection in the insect’s brain.
Turaga, Macke and doctoral student Janne Lappalainen were part of a team that thought they could use these maps to create a computer model that works a lot like the fruit fly’s visual system, which makes up most of the animal’s brain.
The team started with the fly’s connectome, a detailed map of the connections between neurons.
“It tells us how information flows from A to B,” Macke says, “but it doesn’t tell us what route the system actually takes.”
Although scientists have been able to get a glimpse of processes inside the brains of live fruit flies, they have had no way to capture the activity of thousands of neurons responding to a signal in real time.
“The brain is so complex that I think the only way we can understand it is to build accurate models,” Macke says.
In other words, simulating the brain, or parts of the brain, on a computer.
So the team decided to create a model of the brain circuitry that enables fruit flies to sense motion, such as the approach of a fast-moving hand or a fly swatter.
“Our goal wasn’t to build the world’s best motion detector, but to find one that works like a fly.”
The team started with a virtual version of 64 different neurons connected in the same way as a fly’s visual system. The network then “watched” video clips depicting different types of movement.
Finally, they asked the artificial intelligence system to study neuronal activity while playing the video clips.
Ultimately, this approach produced a model that could predict how each neuron in the artificial network would respond to a particular video. Remarkably, the model also predicted the responses of neurons in real fruit flies that had watched the same videos in a previous study.
Tools for Neuroscience and AI
Although the paper describing the model was only recently published, the model itself has been around for over a year, and has been drawing the attention of brain scientists.
“I use this model in my research,” says Cowley, whose lab studies how the brain responds to external stimuli. He says the model helps him determine whether ideas are worth testing in animals.
It is expected that future versions of the model will be extended beyond the visual system to include tasks beyond motion detection.
“We now have a plan for how to build a whole-brain model of the brain that does the interesting calculations,” Macke says.
