Neural mechanisms for integrating prior knowledge and likelihood in value-based probabilistic inference.

Area of Science:

• Cognitive Neuroscience
• Decision Theory
• Neuroimaging

Background:

• Bayesian decision theory models how prior knowledge and current sensory information are combined to form beliefs about outcomes.
• Understanding the neural basis of integrating prior and likelihood information is crucial for explaining adaptive decision-making.
• Previous research has implicated various brain regions in decision-making, but the specific mechanisms of Bayesian integration remain under investigation.

Purpose of the Study:

• To investigate the neural mechanisms underlying Bayesian integration of prior knowledge and likelihood information in humans.
• To identify brain regions involved in combining past experiences (prior) with current sensory data (likelihood) for reward probability estimation.
• To examine how neural representations in these regions reflect behavioral weighting of prior versus likelihood information.

Main Methods:

• Developed a novel lottery decision task manipulating prior knowledge and likelihood of reward probability on a trial-by-trial basis.
• Utilized functional magnetic resonance imaging (fMRI) in human participants to measure brain activity during the task.
• Analyzed behavioral data to assess how participants weighted prior and likelihood information as a function of sample size.

Main Results:

• Behaviorally, participants increasingly weighted likelihood information over prior information as sample size increased, consistent with Bayesian principles.
• The medial prefrontal cortex (mPFC) showed activity correlating with the posterior distribution's mean, reflecting integrated reward probability.
• Neural representations of both prior and likelihood information were found within the mPFC, dynamically adjusting to behavioral weighting shifts.

Conclusions:

• The medial prefrontal cortex (mPFC) plays a critical role in the neural integration of prior knowledge and likelihood information.
• The mPFC represents distinct neural codes for prior and likelihood, dynamically integrating them to form a posterior estimate.
• These findings provide direct neural evidence for Bayesian computation in the human brain during decision-making under uncertainty.