Simulated synapse loss induces depression-like behaviors in deep reinforcement learning

Area of Science:

• Neuroscience
• Artificial Intelligence
• Computational Psychiatry

Background:

• Major Depressive Disorder (MDD) is a complex mood disorder with unclear underlying mechanisms.
• Current models often focus on neurotransmitter imbalances, but connectivity deficits are also implicated.
• Deep Reinforcement Learning (DRL) offers a computational framework to model brain functions.

Purpose of the Study:

• To investigate the role of reduced neural connectivity in simulating MDD-like behaviors using DRL.
• To compare the effects of simulated spine loss with other MDD theories.
• To explore potential therapeutic avenues by modulating neural plasticity.

Main Methods:

• Modified a Deep Reinforcement Learning agent by simulating dendritic spine loss, mimicking a key feature of MDD.
• Assessed the agent's behavior for MDD-like symptoms such as anhedonia and altered reward processing.
• Compared the outcomes with simulations based on alternative MDD hypotheses (e.g., dopamine system dysfunction).

Main Results:

• Simulated dendritic spine loss induced a range of MDD-like behaviors, including anhedonia, increased temporal discounting, and altered exploration/exploitation.
• Alternative computational models of MDD did not replicate the same breadth of symptoms.
• The model supports a connectivity-centric view of MDD, offering insights into dopamine system dysfunction.

Conclusions:

• Reduced brain connectivity, simulated by dendritic spine loss, is a sufficient computational mechanism to produce MDD-like behaviors.
• This challenges purely monoamine-based theories and highlights the importance of information processing capacity.
• Reversing simulated spine loss rescued behavior, supporting treatments that promote neural plasticity and synaptogenesis.