The paper presents a two-layer radial-basis-function neural network trained via Bayesian regularization to obtain high-accuracy numerical solutions of a delay differential Parkinson's disease model with five state variables (healthy/infected neurons, activated microglia, extracellular α-synuclein,…

Although the computational approach yields precise simulations useful for studying disease dynamics, it provides little in the way of actionable therapeutic mechanisms, biomarkers, or translational insights for Parkinson's drug discovery, but may be a helpful tool for hypothesis testing and model…

The purpose of this study is to find the numerical solutions of the delay Parkinson's disease model by employing a computing neural network framework. The disease model is divided into five components: healthy brain neurons, infected brain neurons, activated microglia cells, extracellular α-synuclein, and the activated T-cell population. A two-layered neural network scheme using radial basis functions in both hidden layers, and twelve and twenty neurons in layer-1 and layer-2 for solving the Parkinson's disease model. A dataset is obtained using the implicit Runge-Kutta method, which is trained by the Bayesian regularization taking reasonable percentages of training, testing and validation. The objective of this research is to minimize the mean square error using the proposed two-layered neural network structure. The absolute error values ranged from 10-07 to 10-09 confirming the accuracy of the designed technique. In addition, the optimal training between 10-11 to 10-13, along with error histograms, regression analysis, and state transition plots, validates the accuracy of the proposed scheme.