Opponent learning with different representations in the cortico-basal ganglia pathways can develop...

Area of Science:

• Neuroscience
• Computational Psychiatry
• Cognitive Science

Background:

• Obsessive-compulsive disorder (OCD) is linked to impaired model-based control.
• Previous research suggests OCD involves differential memory traces for positive and negative prediction errors (PEs).

Purpose of the Study:

• To computationally explore the relationship between impaired model-based control and altered prediction error memory in OCD.
• To model how cortico-basal ganglia pathways might contribute to OCD pathophysiology.

Main Methods:

• Developed a dual-system computational agent combining successor representation (SR)-based (model-based) and individual representation (IR)-based (model-free) systems.
• Simulated agent behavior in an environment modeling the obsession-compulsion cycle.
• Tested the agent in a two-stage decision task, comparing SR-only and SR+IR agents.

Main Results:

• A dual-system agent with differential learning rates for positive (SR) and negative (IR) PEs developed an obsession-compulsion cycle.
• The SR+IR agent exhibited reduced model-based control weights compared to an SR-only agent.
• This model reconciles impaired model-based control and memory trace imbalance theories in OCD.

Conclusions:

• Opponent learning between model-based (SR) and model-free (IR) systems may underlie obsession-compulsion.
• This framework offers a novel explanation for OCD pathophysiology.
• Future work could explore opponent learning in threat/aversiveness circuits for a fuller model of OCD.