Reverse Engineering Algorithm for Neural Networks Applied to the Subthalamopallidal Network of the Basal Ganglia

The ordinary differential equations approach to neutral networks modeling is one of the most sensible approach but also very difficult. We proposed a reverse engineering algorithm for neural networks based on linear genetic programming. This algorithm allows the automatic discovery of the structure, estimation of the parameters, and even identification of the biophysical mechanisms involved. It starts either from experimental time-series data or from simulated data. We tested the algorithm on simulation data of a model for the subthalamopallidal network of the basal ganglia, and the results are very good. The high accuracy and short CPU time are mainly due to domain knowledge use, like the Hodgkin-Huxley formalism, and to reducing the problem of reversing a system of differential equations to one of reversing individual algebraic equations. To our knowledge, this is the first realistic reverse engineering algorithm, based on linear genetic programming, applied to neural networks.