Laboratory of Neuropsychology, National Institute of Health, CNRS and University of Paris VI
A few simple, traditional measurements completely specify the structure of spike trains
Over the past decade there has been an intense debate about the dimensionality of the neural code. It is obvious that the strength (over some relatively broad time span) of a neuronal response codes information about experimental conditions. However, the possibility that temporal response features carry other information about experimental or cognitive conditions tantalizes both experimentalists and theoreticians. If the timing of spikes carries such additional information, then neurons process more, and perhaps other, information than the response strength itself allows. Thus, it is important to identify which response features code information, and how precisely they must be specified. Ideally we want to identify the smallest number of response parameters that are needed to specify all of the the response features that we measure. Our recent work has focused on identifying the response features needed to specify the neuronal responses, including precisely timed spike patterns. By specifying only the spike count distribution (not Poisson, approximately truncated Gaussian), the perievent spike density (thus incorporating the latency and having bandwidth < 20 Hz), and the spike interval distribution (almost not needed), we have formed a simple stochastic model that generates spike trains that are indistinguishable from the experimentally recorded ones for neurons throughout the visual system and in the motor system. The same stochastic model provides the basis for predicting exactly the number of synchronous spikes and other exactly timed spike patterns seen across motor cortical neuronal pairs during motor preparation and movement. Thus, it appears that precisely timed response structures are directly related to few relatively simple, traditionally-measured parameters of the responses.