Optimization of the spiking neural network parameters using genetic algorithm in a computational model of schizophrenia

Abstract

Background and Aim: Computational models of spiking neural networks, employed to understand the dynamical mechanisms of impaired gamma frequency oscillations in schizophrenia, usually are complex nonlinear systems and have a large number of free parameters. Whole parameter space exploration and estimation of the optimal parameter sets in complex biological models is a difficult task. The aim of this study was to assess the performance of the genetic algorithm in a parameter identification problem. Methods: Computational model of a spiking neural network is composed of 800 pyramidal neurons (PCs), 150 regular-spiking interneurons (RSIs) and 50 fast-spiking interneurons (FSIs) described by the integrate-and-fire models (K.Spencer, 2009). All cells are randomly interconnected and have recurrent connections between each other. The background activity in the cortex is represented by a Poisson noise input to the network cells (PCs, RSIs and FSIs) at 100 Hz. In addition, network receives 20 Hz and 40 Hz drive excitatory stimulation. Genetic algorithm is applied to estimate the optimal synaptic weights to PCs, RSIs, FSIs and GABA receptor-gated channel time constant. Results: The optimal parameter set of synaptic weights to PCs, RSIs, FSIs and GABA receptor-gated channel time constant is obtained and allows reproducing synchronous network oscillations at 20 Hz and 40 Hz, observed experimentally in healthy and pathological conditions. Conclusions: Genetic algorithm is an effective tool for large-scale nonlinear optimization problems and can be applied to estimate the optimal parameter set in a computational model of schizophrenia-affected neural network.