Extended Data Fig. 1: Human sleep scoring, polysomnography, example traces.
From: A nonoscillatory, millisecond-scale embedding of brain state provides insight into behavior
A, 8-seconds of exemplary polysomnography data from an animal during wake. From top-to-bottom: the low-passed LFP trace, the LFP spectrogram, the EMG, and the MinMax normalized change in position tracked by DeepLabCut (DLC) are shown (Mathis et al.75). To form EEG-like traces, raw data were low-pass filtered at 125 Hz, downsampled to 500 Hz, and 8 non-adjacent channels were averaged. B, 8-seconds of exemplary data from NREM, C, 8 seconds of exemplary data from REM. D, An hour of data sleep-scored for Wake, NREM and REM with polysomnography. Delta band power (0.1-4 Hz) is highly enriched in NREM (slow wave) sleep, theta band power (6-8 Hz) is enriched in REM sleep, and high resolution motor output disambiguates periods of waking, including microarousals (Watson et al.57). E, Top- Confusion matrices for average individual scorer performance using EMG or DLC for motion (rows) to consensus scoring among three expert scorers using polysomnography, including DLC and EMG (columns). Bottom- Confusion matrices for microarousal detection in NREM for average individual scorer performance using polysomnography (including DLC and EMG) and polysomnography (with DLC, but not EMG). Ground truth is consensus of three expert scorers using polysomnography, including DLC and EMG (columns). F, Exemplary hour of data sleep-scored by three experts to produce a consensus score, which is used as training data for a CNN (predictions shown). Blue movement trace is based on DeepLabCut. G, Box plots of normalized optical flow (see methods section “Flickers During Waking Behavior”) for active wake (active motion state), quiet wake (inactive motion state), NREM, and REM. Box plots box the intermediate quartiles, and whiskers extend 1.5 IQR beyond this box. Highly significant differences in means were observed for all comparisons (p < 0.0001 for all comparisons, ANOVA with post-hoc EMMeans with Tukey correction, n = 1 animal). ns p > 0.05, * p < =0.05, ** p < 0.01, *** p < 0.001.
