Individual differences in policy precision: Links to suicidal risk and network dynamics.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Cognitive Science

Background:

• Behavioural modelling advances understanding of psychiatric conditions.
• Existing models often lack biological plausibility.
• Reinforcement learning (RL) models are widely used but may not fully capture neural mechanisms.

Purpose of the Study:

• Develop and evaluate a novel active inference model for decision-making.
• Assess the biological plausibility and explanatory power of the active inference model compared to RL.
• Investigate the relationship between model parameters, brain network activity, and psychiatric conditions.

Main Methods:

• Probabilistic two-armed bandit task.
• Active inference framework.
• Comparison with established reinforcement learning (RL) models.
• Analysis of large-scale brain network activity and inter-subject variability.

Main Results:

• The active inference model outperformed conventional RL models in explaining choice behaviour variability.
• Policy precision optimization, balancing model predictions and observations, was key.
• Temporal dynamics of policy precision improved explanations of brain activity and inter-subject variability.
• Policy precision correlated with default mode, dorsal attention, and frontoparietal network activity.
• Disrupted coordination was observed with prolonged ventral attention network dominance.
• Heightened sensitivity to negative outcomes linked to suicidal risk in major depressive disorder.

Conclusions:

• Active inference provides a biologically plausible framework for decision-making modelling.
• Policy precision is a crucial parameter linking behaviour, brain networks, and psychiatric conditions.
• This model offers insights into neural mechanisms underlying decision-making impairments and suicidal risk.