PoseR: a deep learning toolbox for classifying animal behaviour.

Area of Science:

• Ethology and Behavioral Neuroscience
• Computational Biology and Machine Learning
• Animal Science

Background:

• Animal behavior analysis is crucial for understanding cognition and relies on interpreting movement patterns.
• Current methods often require extensive, species-specific feature engineering from pose estimation data.
• A need exists for generalized, standardized tools for efficient and reproducible behavioral classification.

Purpose of the Study:

• To develop a generalized behavioral classifier using deep learning and pose estimation data.
• To create a versatile and scalable tool applicable across multiple species and contexts.
• To simplify and standardize the animal behavior analysis workflow.

Main Methods:

• Utilized spatio-temporal graph convolutional networks for behavior classification.
• Developed PoseRecognition (PoseR), a tool transforming pose estimation coordinates into semantic labels.
• Validated the approach using diverse model organisms: zebrafish larvae, fruit flies, mice, and rats.

Main Results:

• PoseRecognition (PoseR) accurately and rapidly classifies animal behavior from pose estimations.
• The tool demonstrated versatility across different species and experimental contexts.
• Achieved efficient behavioral analysis by automating the transformation of pose data into meaningful labels.

Conclusions:

• PoseRecognition (PoseR) provides a foundational, standardized approach to animal behavior modeling.
• The tool enhances the efficiency, reproducibility, and scalability of behavioral analysis.
• Facilitates cross-species and cross-contextual behavioral studies, advancing ethological research.