A prospective code for value in the serotonin system

Area of Science:

• Neuroscience
• Computational Neuroscience
• Behavioral Neuroscience

Background:

• Dorsal raphe nucleus serotonin neuron responses to stimuli are not fully explained by existing theories like reward, surprise, salience, or uncertainty.
• Previous models fail to reconcile observations of serotonin neuron activity, such as differential responses to surprising rewards versus punishments.

Purpose of the Study:

• To develop a unifying computational model for dorsal raphe nucleus serotonin neuron activity.
• To integrate reinforcement learning theory with the filtering properties of the dorsal raphe nucleus.

Main Methods:

• Developed a novel computational model based on a prospective code for value.
• Integrated concepts from reinforcement learning theory and recent findings on dorsal raphe nucleus filtering properties.
• Quantitatively predicted in vivo population activity of serotonin neurons.

Main Results:

• The proposed prospective code for value successfully explains serotonin neuron activation by both rewards and punishments.
• The model accounts for the stronger activation by surprising rewards compared to punishments, a discrepancy unresolved by prior theories.
• The model demonstrates superior quantitative prediction of in vivo population activity compared to existing theories.

Conclusions:

• The prospective code for value offers a unifying framework for understanding serotonin neuron function in response to salient stimuli.
• This work bridges computational neuroscience and serotonin system research, advancing the understanding of serotonin's role in learning and behavior.