An in-silico integration of neurodevelopmental and dopaminergic views of schizophrenia

Abstract: Deep reinforcement learning (DRL) algorithms have the potential to provide new insights into psychiatric disorders. Here we create a DRL model of schizophrenia: a complex psychotic disorder characterized by anhedonia, avoidance, temporal discounting, catatonia, and hallucinations. Schizophrenia's causes are not well understood: dopaminergic theories emphasize dopamine system dysfunction, while neurodevelopmental theories emphasize abnormal connectivity, including excitation/inhibition (E/I) imbalance in the brain. In this study, we suppressed positive (excitatory) connections within an artificial neural network to simulate E/I imbalance. Interestingly, this is insufficient to create behavioral changes; the network simply compensates for the imbalance. But in doing so it becomes more sensitive to noise. Injecting noise into the network then creates a range of schizophrenic-like behaviours. These findings point to an interesting potential pathology of schizophrenia: E/I imbalance leads to a compensatory response by the network to increase the excitability of neurons, which increases susceptibility to noise. This suggests that the combination of E/I imbalance and neural noise may be key in the emergence of schizophrenic symptoms. We further notice altered response to reward prediction error in our model, and thus propose that E/I imbalance plus noise can account for both schizophrenia symptoms and dopamine system dysfunction: potentially unifying dopaminergic and neurodevelopmental theories of schizophrenia pathology.