BioLLMAgent: A Hybrid Framework with Enhanced Structural Interpretability for Simulating Human Decision-Making in Computational Psychiatry

Researchers introduce BioLLMAgent, a hybrid framework combining reinforcement learning models with large language models to simulate human decision-making in computational psychiatry. The framework demonstrates strong interpretability while accurately reproducing human behavioral patterns and successfully simulating cognitive behavioral therapy principles.

• →BioLLMAgent combines traditional RL models with LLM capabilities to balance interpretability and behavioral realism in psychiatric research.
• →The framework achieved excellent parameter identifiability with correlations above 0.67 across six clinical and healthy datasets.
• →Multi-agent dynamics simulations suggest community-wide educational interventions may outperform individual treatments.
• →The system successfully simulates cognitive behavioral therapy principles and provides a computational sandbox for testing psychiatric interventions.
• →Framework validation spans multiple tasks including Iowa Gambling Task, reward-punishment learning, and temporal discounting.