Fleeing is believing: adaptive behavior under social threat as an inference process

Area of Science:

• Computational neuroscience
• Animal behavior modeling

Background:

• Adverse social experiences significantly impact animal behavior.
• Mechanisms for selecting context-appropriate behaviors after social interactions are poorly understood.
• Existing models lack mechanistic frameworks for individual variability and outcome prediction.

Purpose of the Study:

• To develop a mechanistic computational model of social defeat in mice.
• To explain individual behavioral variability and predict outcomes following social stress.
• To provide a framework for understanding the computational basis of behavior modification by social experiences.

Main Methods:

• Modeled social defeat as a partially observable Markov decision process (POMDP).
• Implemented a heterarchical agent architecture with interacting modules.
• Used the model to fit observed behavioral motifs and mouse phenotypes.

Main Results:

• The model successfully reconstructed behavioral motifs like investigation, hesitation, and flight.
• It captured the impact of social defeat as a parameter shift in the internal generative model.
• Behavioral motifs emerged from model inference dynamics without trajectory-level fitting.
• The model reproduced effects of interventions and generated testable predictions.

Conclusions:

• The developed POMDP model provides a mechanistic framework for understanding behavior after social defeat.
• The heterarchical architecture offers a foundation for modeling diverse behavioral domains.
• This computational approach has potential clinical relevance for human trauma and anxiety disorders.