Information transmission and recovery in neural communication channels revisited

Abstract

Nerve cells in the brain generate all-or-none electric events-spikes-that are transmitted to other nerve cells via chemical synapses. An important issue in neuroscience is how neurons encode and transmit information using spike trains. Recently, signal transduction through two neurons connected by an excitatory chemical synapse was studied by Eguia et al. [Phys. Rev. E 62, 7111 (2000)]. They reported an apparent violation of the data processing inequality: The mutual information between the input signal and the output of the first neuron can be lower than the mutual information between the input signal and the output of the second neuron, that only receives input from the first neuron. We investigate whether it is possible, using a different method, to retrieve, from the first neuron's spike train, all the information about the input that is present in the second neuron's output. We find that single interspike intervals (ISI's) from the first neuron, at a resolution of 0.5 time units, contain more information about the input signal than those of the second neuron. Using a classification procedure based on the ISI return map, we recover 71% of the input entropy using the first neuron's spike train, and only 42% using the second neuron's spike train. Hence for these spike-train observables the data processing inequality is not violated.