Model Overview
A research team at the Korea Advanced Institute of Science and Technology (KAIST) has developed an AI model named BehavERT that interprets animal movement patterns similarly to how language models analyze words. Unlike conventional behavioral analysis methods, this model requires no prior biological training to autonomously detect social behavior deficits in autism model mice. The findings were reported via The Korea Herald.
Technical Principle and Performance
BehavERT is built on the classic BERT architecture. It converts skeletal coordinates of mouse body parts—including nose, ears, spine, limbs, and tail—into tokens for sequence analysis. In experiments, the model successfully distinguished Shank3B knockout autism model mice from control groups by observing mouth-to-mouth contact behavior. Compared with existing methods, BehavERT outperformed in all five international benchmarks. The team emphasized that the model can not only classify behaviors but also comprehend their behavioral significance, offering a novel perspective for subsequent research.
Application Prospects
The research team suggests that BehavERT could play a crucial role in drug development, psychiatric research, and behavioral genetics. For example, in drug screening, the model can rapidly assess the impact of candidate compounds on animal social behavior. In psychiatry, it can help establish more precise animal models to simulate human mental disorders. In behavioral genetics, it enables efficient analysis of associations between genes and specific behaviors.