AcademicsInternational NewsUniversity News

Tübingen and Virginia Researchers Utilize AI to Decode Fruit Fly Brain Calculations

By iednewsdesk On Oct 10, 2024

Information in the brain is transmitted via electrical signals between specialised cells, the neurons. Large networks of such neurons control perception, behaviour and cognition. Science has long searched for ways to simulate neural networks in the brain with computers in order to understand how they work. With new findings about the neural circuitry in the brain of the fruit fly and methods of artificial intelligence, researchers have now succeeded in creating a neural network that does something previously unimaginable: It predicts the activity of individual neurons without having to take measurements on a living brain. The study by Professor Jakob Macke and Janne Lappalainen from the University of Tübingen and Dr. Srinivas Turaga and colleagues from the Janelia Research Campus of the Howard Hughes Medical Institute in Ashburn, Virginia (USA), was published on Wednesday in the journal Nature .

For decades, neuroscientists have been measuring neural activity in living animals to better understand the connections between brain activity and behavior. These experiments have yielded groundbreaking insights into how the brain works – but much of the brain remains unexplored.

The teams from Tübingen and Virginia have now used artificial intelligence and the connectome, a map of neurons and their connections in brain tissue, to predict the activity of neurons in the living brain. By using only information from the connectome of the fruit fly’s visual system and making assumptions about the functions of the circuit, the researchers have created an AI simulation that can predict the activity of each neuron in the circuit. “We now have a computational method that allows us to translate measurements of the connectome into predictions of neuronal activity and brain function without first having to carry out complex measurements on the living neuron,” says Srinivas Turaga, head of the Janelia research group and one of the lead authors of the new study.

The research team used the connectome to create a detailed mechanistic network simulation of the fly’s visual system, where each modeled neuron corresponds to a real neuron and each modeled synapse corresponds to a real synapse in the brain. Although they did not know the dynamics of the neurons in the real tissue, the team was able to predict these unknown parameters using deep learning methods. To do this, they combined the information from the connectome with their knowledge of the circuit’s function: detecting movement. “With this combination, we were able to test whether our connectome-based approach can provide a good model of the brain,” says Janne Lappalainen, a doctoral student at the University of Tübingen and lead author of the study.

The new model predicts the activity of 64 different neuron types in the fruit fly’s visual system and reproduces the results of over two dozen experimental studies over the past two decades. According to the authors, the new work has the potential to fundamentally change brain research by allowing predictions about the activity of individual neurons to be derived directly from the connectome. In principle, the model can be used to simulate any experiment, so that predictions derived from it can then be tested in the laboratory.

The new research paper contains over 450 pages of predictions derived from the model, including the identification of cells that were previously unknown to be involved in motion detection and that can now be studied in fruit flies. “Until now, there was a large gap between the static connectome and the dynamics of computations in the living brain. The question was whether we could close this gap with a model. We have now succeeded in doing this using the specific example of the fruit fly,” says Jakob Macke, one of the lead authors of the study. With this approach, it is possible to create artificial neural networks that are similar to the fruit fly brain and can be used for a variety of studies in the future: For example, they could be used to investigate why biological neural networks are orders of magnitude more efficient than artificial neural networks.

Janne Lappalainen is a PhD student at the University of Tübingen and the International Max Planck Research School “Intelligent Systems” and a visiting scientist at the Janelia Research Campus of the HHMI. Professor Jakob Macke leads the group “Machine Learning in Science”, as part of the Tübingen Cluster of Excellence “Machine Learning: New Perspectives for Science” and is a researcher at the Tübingen AI Center and the ” Bernstein Center for Computational Neuroscience Tübingen”. The project was partly funded by the ERC Consolidator Grant DeepCoMechTome .

University of Tübingen