Spike train analysis for single trial data

Traditional methods in neural data analysis are not appropriate for analyzing the spike train of a single experimental trial. We show that, by constructing a model of firing statistics, a more accurate estimate of the firing rate for a single spike train can be obtained. The model is based on the assumption that the neuron's spikes are generated by a non-homogeneous Poisson process which follows Markovian dynamics. We test the method by reconstructing the input stimulus based on the neurons’ responses either on the raw spike data or the firing rate estimate. The spike data were recorded from macaque V1 neurons in response to a sinewave grating undergoing pseudo-random walk. For a large percentage of the cells studied, the reconstruction is significantly improved by using the estimated firing rate over the raw spikes, suggesting that estimated rate reflects more accurately the underlying state of the neurons.