Can We Predict Every Spike?

Abstract

Is it possible to predict the spike times of a neuron with millisecond precision? In the classical picture of rate coding (Adrian, 1928), single spikes do not play a role, and the question would have to be answered negatively. For rate coding in a singleneuron, the relevant quantity to encode a stimulus such as pressure onto a touch sensor in the skin (Adrian, 1928) or presence of a light bar in the receptive field of a visual neuron (Hubel and Wiesel, 1959) is the number of spikes a neuron emits in a short time window of, e.g., 100ms. The timing of the spikes is considered as irrelevant. However, over the last 20 years many researchers have shown that it is not only the temporally averaged firing rate that carries information about the stimulus, but also the exact timing of spikes. For example, spike timing has shown to be relevant to encode force amplitude and direction in touch sensors of the skin (Johansson and Birznieks, 2004) as well as the whole-field visual movements (Bialek et al., 1991) or object movement (Gollisch and Meister, 2008) in visual neurons.If spike timing is important, a whole series of questions arises: What is the precision of spike timing if the same stimulus is repeated several times? Do spikes always appear at the same time? What would be a sensible measure of spike timing precision and reliability? Can a neuron model match the spike timing precision of a real neuron? Does it matter which neuron or what stimulus we take? If so, what would be a useful stimulus?To answer these related questions, let us think of the following experimental protocol. An experimentalist injects a time-dependent input of, say, 20 second duration into a single neuron. The neuron responds with spikes. The experimentalist now repeats the same stimulus sequence several times. At each repetition, the neuron responds with a spike train that may or may not look similar to the previous one: some spikes appear at exactly the same time during the stimulus sequence, some are missing, some are shifted by a few millisecond or appear at a completely different time. The information derived from this type of experiment which dates back to Bryant and Segundo (1976) and has been popularized by should be sufficient to answer questions regarding precision and reliability of spike timing.

Spike-timing prediction in cortical neurons with active dendrites

“…The predicted spike trains achieved an averaged coincidence rate of 50%. The scaled coincidence rate obtained by dividing by the intrinsic reliability (Jolivet et al, 2008a ; Naud and Gerstner, 2012 ) was 72%, which is comparable to the state-of-the performance for purely somatic current injection which reaches up to 76% (Naud et al, 2009 ). Comparing with a passive model for dendritic current integration, we found that the predictive power decreased to a scaled coincidence rate of 53%.…”

The tuning of tuning: how adaptation influences single cell information transfer

“…Finally, we find that both the intrinsic (membrane conductance) and response (STA, FI) properties of excitatory neurons are more heterogeneous, compared to inhibitory neurons. [22,24,[35][36][37][38][39][40][41][42][43][44][45] and the heterogeneity in such cell properties has been investigated in excitatory (but not inhibitory) cells [46]. With this manuscript, we add a functional dimension to these basic properties of cortical spike initiation: We show how different mechanistic features of cortical cells can influence their information transfer and binary classification.…”

The tuning of tuning: How adaptation influences single cell information transfer

“…It has been shown repeatedly, that the spiking behaviour of cortical neurons can be fitted relatively well with a simple threshold model with an extra feedback variable [22,24,[67][68][69][70][71][72][73][74][75][76][77].…”