Fig. 6: Decoding of touch using high-frequency AP burst events.

a Example trace of same L5tt neuron of Fig. 5a–e with 100 ms sliding window (+10 ms steps) to obtain spiking characteristics and associated behavior. High-frequency bursts are highlighted in red, accompanied by red asterisks. Decoding was analyzed by a 100 ms sliding window (10 ms steps) to sample the rate of bursts. b Receiver operating characteristic curve (ROC-curve), decoding performance was estimated as the area under the ROC curve (AUROC). Bracket line indicates chance level. c Histogram of AUROC values after shuffling the data for 5000 repetitions. Note that the true AUROC value (blue line) is beyond the 99th percentile of the shuffled distribution, indicating significant decoding power. Example matches Fig. 5, 5tt, #1). d, e Analogous to b, c but for example L5tt, #2, matching Fig. 5, 5tt, #2). Note that the true AUROC value exceeds the 99th percentile of the shuffled distribution, indicating significant decoding power. f ROCs for the population of L5tts. g ROCs for individual L2/3 (in red), L4 (black), L5st (green), and L6 (gray) recordings. Bracket line indicates chance level. h AUROC values as a function of cell type. i Cumulative distribution of absolute AUROC value as a function of cell-type (Kruskal–Wallis, p < 0.01, post hoc test L5tt vs L2/3 < 0.01, L5tt vs L6 p < 0.05, Kolmogorov–Smirnov test for comparison of two distributions, L5tt vs rest, p < 0.01). j AUROC values as a function of cell type, computed on decoding value of spike rate. k Cumulative distribution of absolute AUROC value as a function of cell-type, computed on decoding value of spike rate (Kruskal–Wallis, p < 0.01, post hoc test L5tt vs L2/3 < 0.01, L5tt vs L6 p < 0.05, Kolmogorov–Smirnov test for comparison of two distributions, L5tt vs rest, p < 0.01). l Cartoon illustrating Jensen–Shannon metric (JS distance, see Methods). m JS distance including 95% confidence intervals (error bars) for individual identified neurons, based on spike count and bursting. n Cartoon illustrating JS metric for “burst only”. o JS distance for “burst only” (correcting √JSD “spike count + bursting” with √JSD for “spike only”) is cell-type specific (Kruskal–Wallis, p < 0.001) and significantly higher for L5tt compared with L2/3, L4, and L6 (Dunn’s Multiple comparisons test, p < 0.05). Blue markers represent neurons with significant information in bursts not present in spike counts (p < 0.01 after correction for false discovery rate, Benjamini–Hochberg procedure, see Methods).