Figure 1.

Spike trains can be characterized in terms of events. (A) A rastergram with spike trains obtained from an example model neuron in response to the same fluctuating current waveform presented multiple times (Tiesinga & Toups, 2005). The spike trains vary from trial to trial because of an independent noise current (Tiesinga & Toups, 2005). In a rastergram, each spike is represented by a tick, with its x-ordinate being the spike time and the y-ordinate being the trial number. In the rastergram, vertical bands correspond to spike alignments, each of which is an event. (B) The spike time histogram is the number of spikes that fall in a bin, normalized by the number of trials and the bin width so that the result is a firing rate expressed in Hz. Spike alignments (events) lead to peaks in the histogram. The spike time jitter of an event is the width of the corresponding peak, which can be estimated as the standard deviation of the spike times in the event. Sometimes we will use precision instead, which is numerically equal to the inverse of the jitter. The area under the peak is proportional to the reliability, the fraction of trials on which there was a spike during the event. When peaks are isolated, it is easy to assign each spike to an event. However, if there are overlapping peaks, such as the peak indicated by the double arrow, event assignment is much more difficult. In this article, we introduce a technique, based on detecting spike patterns, which can efficiently separate events. (C) When the same spike trains as in panel A are sorted using the clustering algorithm (see section 3.3), spike patterns emerge. The horizontal line separates two spike patterns, and the gray curves are the corresponding histograms. The aggregate histogram in panel B is a weighted sum of the two histograms shown separately in panel C.