Synapses, predictions, and prediction errors: A neocortical computational study of MDD using the temporal memory...

Area of Science:

• Computational neuroscience
• Neurobiology of mental health
• Machine learning applications in psychiatry

Background:

• Synapses and dendritic spines are implicated in major depressive disorder (MDD) pathophysiology.
• Ketamine and psilocybin show rapid antidepressant effects, suggesting a role for synaptic plasticity.
• MDD can be conceptualized as persistent negative affective states due to impaired interoception and prediction errors.

Purpose of the Study:

• To investigate the relationship between synaptic changes and the predictive functions of the neocortex in the context of MDD.
• To model how synaptic loss affects the brain's ability to predict future states, potentially explaining MDD symptoms.

Main Methods:

• Utilized the temporal memory (TM) algorithm, an unsupervised machine learning model simulating a neocortical layer.
• Trained the TM model on sequences of affective states (represented by letters).
• Simulated MDD by progressively reducing the number of synapses in the TM model and assessing its predictive capacity.

Main Results:

• Non-linear decrease in prediction accuracy as synapses were destroyed.
• A 25% reduction in synapses significantly decreased prediction confidence, even when predictions remained accurate.
• Further synapse reduction led to a marked drop in the number of predictions.

Conclusions:

• Findings suggest that synaptic loss can lead to persistent negative affective states with high prediction errors, aligning with the Bayesian brain hypothesis of MDD.
• Synaptic growth, as potentially facilitated by treatments like ketamine and psilocybin, may enhance prediction confidence and represent more complex future states.
• This study pioneers the use of the TM model to link synaptic-level changes to psychiatric symptomatology, offering insights into treatment mechanisms for MDD.