Robust Parameter Selection for Compartmental Models of Neurons Using Evolutionary Algorithms

Abstract

Many modeling efforts have met skepticism among experimental neuroscientists largely because of a perception that system parameters are selected arbitrarily. In particular, a lack of high resolution data describing the spatial distribution of voltage-gated and/or calcium-dependent channels has rendered compartmental models suspect of representing singular solutions. On the other hand, a robust manifold of solutions may exist in nature and account for the behavioral repertoire of the biological system. We demonstrate the use of Evolutionary Algorithms (EAs) for the robust fitting of nonlinear channel distributions to experimental observations using the model and kinetics of Traub et al (1991). These results represent a successful method for optimizing parameters in high dimensional (> 100) nonlinear systems.