Reinforcement learning deficits exhibited by postnatal PCP-treated rats enable deep neural network classification

Area of Science:

• Neuroscience
• Behavioral Science
• Computational Psychiatry

Background:

• Flexible decision-making is crucial for cognitive function and is impaired in schizophrenia.
• The probabilistic reversal learning (PRL) task assesses decision-making flexibility and has been adapted for rodents.
• Disrupting glutamate neurotransmission during early development can cause schizophrenia-relevant abnormalities.

Purpose of the Study:

• To investigate if early postnatal disruption of glutamatergic transmission using phencyclidine (PCP) impairs decision-making in rats using the PRL task.
• To determine if computational analysis and deep neural networks (DNNs) can identify treatment-related behavioral changes.

Main Methods:

• Rats were treated with PCP during early postnatal development to disrupt glutamatergic transmission.
• Behavioral performance on the PRL task was assessed, measuring reversals, win-stay, and lose-shift responses.
• Computational analysis and DNNs were employed to analyze behavioral data and predict treatment groups.

Main Results:

• Postnatal PCP-treated rats showed impaired decision-making, completing fewer reversals and exhibiting altered reward/punishment sensitivity.
• Computational analysis revealed a significant impairment in the learning rate in PCP-treated rats.
• A DNN successfully predicted the treatment group based on behavioral data with high accuracy.

Conclusions:

• Disrupting early postnatal glutamatergic neurotransmission impairs flexible decision-making, providing a relevant animal model for schizophrenia.
• DNNs show promise as a tool for analyzing behavioral data and potentially aiding in the diagnosis of psychiatric disorders like schizophrenia.