A computational model of levodopa pharmacodynamics in Parkinson's disease.

A series of computational models were developed to better understand basal ganglia functions and the effects of levodopa pharmacodynamics in Parkinson's disease. The models employed a relatively new computational approach known as a neural network, which is a small number of simple processing units interconnected with designated constraints. A key difference from traditional computational modeling is that the networks are "trained" rather than programmed with experimental input and output data. After training, only a limited number of these models, could explain the pharmacodynamic data observed by Mouradian et al. in different groups of Parkinsonian patients. These successful models strongly argue for at least two pharmacologic mechanisms to explain the antiparkinsonian effect and dyskinesia tendency for the different classes of Parkinson's patients: never-treated, stable, wearing-off, and on-off. They suggest different roles for the striatal units by examining predictions of motor and dyskinesia tendency through theoretical blockade of each kind of unit. The models show that the antiparkinsonian effect in Parkinson's disease cannot be explained by the action of dopaminergic neurons on striatal neurons alone. Although the models necessarily oversimplify basal ganglia function, they provide a useful quantitative insight into how motor and dyskinesia behaviors may develop in different Parkinsonian subgroups.