Fig. 5: Variability and clustering of spike auto-correlograms in the AT.

a Auto-correlograms of an HD (1), non-bursty (2) and bursty (3) neuron during wake, REM sleep and NREM sleep epochs. b Clustering and t-SNE embedding of auto-correlograms (each point represents the stacked auto-correlograms from wake, REM sleep and NREM sleep of a neuron). HD neurons are marked with a white dot. The three example neurons from c are circled. K-means clustering of the auto-correlograms result in two clusters (see text). Cluster #2 is color-coded relative to burst index during NREM sleep. c Left, 15 superimposed auto-correlograms of randomly selected cells from cluster #1 (normalized by the baseline between 50 and 100 ms), during REM (top) and NREM (bottom) sleep. Color (white to dark red) indicates average firing rates (from low to high). Note the high similarity of auto-correlograms during REM sleep, independent of basal firing rates, and the mild variability during NREM sleep. d Left, density of cluster #1 (i.e. mostly HD neurons) and, right, mean burst index of cluster #2 over the anatomical schematics, for the same example mouse. e Distribution of Pearson’s correlation coefficients between pairs of neurons recorded on the same shank showing the dynamical homogeneity of HD neurons compared to non-HD neurons. f Classification score of HD versus non-HD neurons (based on auto-correlograms) for increasing duration of auto-correlation. Score is relative to classifiers trained with shuffled data (0, chance level; 1, perfect classification).