Nonparametric modeling of single neuron

Abstract

Nonlinear dynamic models were built with Volterra Lagurre kernel method to characterize the input-output properties of single hippocampal CA1 pyramidal neurons. Broadband Poisson random impulse trains with a 2 Hz mean frequency, which include the majorities of the spike patterns in behaving rats, were used to stimulate the Schaffer collaterals. Corresponding random-interval post-synaptic potential (PSP) and spike train data were recorded from the cell bodies using whole-cell recording technique and then analyzed with the nonlinear dynamic model. The model consists of two major components, i.e., a feedforward three order Volterra kernel model characterizing the transformation of presynaptic stimulations to pre-threshold PSPs, and a feedback one order Volterra kernel model capturing the spike-triggered after-potential. Results showed that the model could predict 1) the sub-threshold PSPs trace with a normalized mean square error around 10% and 2) the spikes with accuracy higher than 80%.