AI Silences Fruit Fly Courtship Song & Controls Neurons in Real-Time

Researchers have developed an artificial intelligence system capable of identifying and interrupting a fruit fly’s courtship ritual in real-time, offering a novel approach to dissecting the neural basis of complex social behaviors. The system, dubbed YORU (Your Optimal Recognition Utility), not only detects the initiation of the courtship “song” – a wing vibration – but also selectively silences the neurons responsible for producing it, allowing scientists to directly test the causal link between brain activity and behavior.

AI Detects Behavior with Speed and Accuracy

Traditional methods of tracking animal behavior often rely on painstakingly analyzing individual body parts frame by frame, a process that becomes exceedingly difficult when multiple animals are interacting or overlapping. YORU circumvents this challenge by recognizing entire postures as single behaviors within a single video frame. After being trained on labeled examples, the AI can identify and categorize actions with an accuracy ranging from 90% to 98% across a variety of species, including flies, ants, and zebrafish.

The speed of this detection is critical. The system operates on a remarkably fast timescale, completing the full loop from camera frame to neural inhibition in an average of 31 milliseconds. This rapid response time, approximately 30% faster than conventional pose trackers (which average a 47-millisecond delay), allows for intervention *during* the behavior, rather than after it has concluded.

Targeting Neurons with Optogenetics

The ability to selectively silence neurons is achieved through a technique called optogenetics. Researchers genetically engineered the fruit flies so that specific brain cells would respond to green light. When YORU detects the beginning of the wing extension associated with courtship, it triggers a precisely aimed light pulse that inhibits the activity of these targeted neurons. This allows researchers to observe the immediate consequences of disrupting the courtship behavior.

Previous brain control systems often illuminated entire arenas, affecting all animals simultaneously. YORU’s precision allows for the targeting of individual flies within a group, minimizing disruption to surrounding individuals. During testing, the light remained focused on the intended target for 89.5% of the stimulation time.

Linking Brain Activity to Behavior

Beyond simply controlling behavior, the system also facilitates the study of the relationship between brain activity and observable actions. By combining YORU with calcium imaging – a technique that tracks neuron activity through glowing signals – researchers can correlate specific neural patterns with behaviors like running and grooming in mice. The resulting maps generated by YORU align with those created through traditional human scoring, validating the AI’s reliability as a readout of behavioral states.

Real-Time Control and Reduced Mating Success

In experiments, the researchers demonstrated the system’s effectiveness by silencing the courtship neurons of male fruit flies during their attempts to woo females. The interruption of the “song” significantly reduced the male’s mating success, providing direct evidence of the role these neurons play in courtship behavior. Professor Azusa Kamikouchi of Nagoya University, the study’s senior author, stated, “We can silence fly courtship neurons the instant YORU detects wing extension.”

Addressing Limitations and Future Directions

The current system has some limitations. Because it analyzes single frames, it may struggle to accurately detect behaviors that unfold over multiple frames. It doesn’t inherently track individual animal identities, meaning it can identify a behavior but may not always be able to follow which individual continues to perform it. Hardware limitations, such as delays introduced by projectors and controllers, can also occasionally allow a fast-moving animal to escape illumination.

Researchers are actively working to address these challenges. Future development will focus on capturing more complex and extended behaviors, as well as reducing hardware-related delays. A key aspect of the project’s success is its usability. The developers have created a graphical interface that allows researchers without extensive coding experience to train new behavior detectors and run experiments. This accessibility is expected to accelerate research into the neural basis of social behavior across a wider range of species.

The study, published in Science Advances, represents a significant step forward in the field of neuroethology, offering a powerful new tool for understanding the intricate relationship between brain activity and behavior. By enabling real-time manipulation and observation, this system promises to unlock new insights into the neural mechanisms underlying social interactions in animals.