Brain research: New study uses AI to predict activities of individual neurons

Contents

For decades, neuroscientists have spent countless hours in the lab, painstakingly measuring the activity of neurons in living animals to figure out how the brain enables certain behaviors. These experiments have provided groundbreaking insights into how the brain works, but they have only scratched the surface and left many parts of the brain unexplored. That is now set to change. Once again, thanks to AI.

In a new study, which is freely downloadable under Open Access, researchers used artificial intelligence and a so-called connectome. It maps the entirety of all neurons in a living being and their connections. The model predicts the role of neurons in a living brain. The researchers used the eyes of a fruit fly and the corresponding connections as a basis. Using information about the connections of their neuronal circuitry from the fruit fly's connectome and a guess about what the circuit should do, researchers created an AI simulation. It predicts which neuron stands for which activity in the circuit.

Predictions about neuronal activity possible

Janelia research group leader Srini Turaga says: "We now have a computational method to turn measurements of the connectome into predictions of neuronal activity and brain function without first having to figure out what each neuron is doing through hard-to-obtain measurements of neuronal activity."

The model developed by the team of scientists from the Janelia Research Campus of the Howard Hughes Medical Institute (HHMI) and the University of Tübingen is very detailed: It used the connectome to create a detailed network simulation of the fly's visual system. Each neuron and each synapse in the model corresponds to a real neuron and a real synapse in the brain.

Although the behavior and responsibilities of the individual neurons and synapses were not known, the researchers were able to derive previously unknown parameters using the data from the connectome and deep learning methods. They combined this information with assumptions about the target of the circuit: they assumed that the eyes should detect whether something is moving in the field of vision.

Model confirms previous studies

"At this point, everything fell into place, and we were able to find out whether this model gives us a good model of the brain", says Janne Lappalainen, a PhD student at the University of Tübingen who led the research. This can be seen as the new model predicted the neuronal activity of 64 neuron types in the visual system of the fruit fly in response to visual input.

The model also reproduces the results of two dozen experimental studies. These studies were conducted over the past two decades.

Study could change the way we look at brain function

What is special about this study is that it can predict the activity of individual neurons using only the connectome. This approach has the potential to change the way we look at how the brain works. Theoretically, scientists can now use such a model to simulate any experiment and generate detailed predictions that can then be verified in the laboratory.

The research results list more than 450 pages of predictions from the new model, including those for identifying cells that were not previously known to be involved in motion recognition. They can now also be verified retrospectively on living flies.

To categorize the research results, Jakob Macke, a senior author of the paper and professor at the University of Tübingen, explains: "There is a big gap between the static snapshot of the connectome and the dynamics of real-time computation in the living brain. The question was whether we could bridge this gap in a model. The study shows a strategy for doing so."

(mil)