Automatic mapping of multiplexed social receptive fields by deep learning and GPU-accelerated 3D videography

Area of Science:

• Neuroscience
• Behavioral Science
• Biophysics

Background:

• Social interactions significantly influence brain and body functions.
• Current methods for analyzing social behavior are often manual and lack scalability.
• Objective, high-resolution tracking is crucial for understanding the neural basis of social behavior.

Purpose of the Study:

• To develop a scalable and objective system for tracking multiple interacting mice.
• To enable high-resolution analysis of social behavior and its neural correlates.
• To investigate the neural circuits underlying complex social encounters.

Main Methods:

• A novel hardware/software system combining 3D videography, deep learning, and physical modeling.
• GPU-accelerated robust optimization for automatic multi-animal tracking in various conditions, including darkness.
• Simultaneous electrophysiological recordings and posture dynamics capture at high spatiotemporal precision (~2 mm, 60 frames/s).

Main Results:

• Fully automatic, low-error tracking of multiple unmarked mice during spontaneous social encounters.
• High-precision capture of posture dynamics in complex social interactions.
• Identification of multiplexed 'social receptive fields' in barrel cortex neurons through a behavior-neural activity statistical model.

Conclusions:

• The developed 3D tracking system provides a powerful tool for neurobehavioral studies.
• This approach facilitates objective and scalable analysis of social behavior and neural activity.
• The system is broadly applicable to multi-animal interactions in challenging low-light environments.