Artificial intelligence has revolutionized the field of animal behavior research, offering a solution to conflicts that have impeded progress in the past. Developed by researchers in Seattle, this innovative software facilitates the rapid analysis of animal behavior, allowing for a more precise correlation between behaviors and the activity of individual brain circuits and neurons.
This program not only accelerates research into the neurobiology of behavior but also enables the comparison and reconciliation of results obtained from different studies. According to Sam Golden, an assistant professor at the University of Washington School of Medicine, the AI system allows laboratories to develop behavioral procedures according to their preferences while still facilitating general comparisons across studies that implement varying behavioral approaches.
Published in the prestigious journal Nature Neuroscience, the paper detailing this program was led by senior authors Golden and Simon Nilsson, with Nastacia Goodwin as the first author. The development of this AI software is a significant step forward in understanding the neural mechanisms underlying animal behavior, which in turn can contribute to advancements in treating human disorders like addiction, anxiety, and depression.
Typically, researchers observe and record animal behaviors in the lab, noting physical responses to different stimuli and linking these behaviors to changes in brain activity. For example, to study aggression, researchers might place two mice in an enclosed space and observe their proximity, posture, and physical displays like tail rattling. However, this manual annotation and classification process can be time-consuming and subject to human error.
To automate this task, the researchers created an AI program called SimBA (Simple Behavioral Analysis) that tracks and classifies animal behavior rapidly and accurately. Despite the widespread adoption of such programs, discrepancies in results between different laboratories arose due to subjective definitions of behaviors and the opacity of AI algorithms.
To address this issue, Goodwin and Nilsson incorporated a machine-learning explainability approach into SimBA, providing transparency on how the AI system arrived at its classifications. By utilizing the Shapely Additive exPlanations (SHAP) score, the researchers could objectively determine the predictive strength of individual features used in the algorithm, thereby standardizing behavioral descriptions and facilitating cross-lab comparisons. This approach enhances reproducibility and interpretability in behavioral research, paving the way for more robust scientific conclusions.
With the adoption of AI software like SimBA and the implementation of explainability tools like SHAP, the field of animal behavior research is poised to make great strides in understanding the intricacies of behavior and its neural underpinnings. By harmonizing methodologies and promoting transparency, scientists can unlock new insights into the complex workings of the brain and behavior, ultimately benefiting both human and animal health outcomes.
