Fig. 3: Utilizing large-scale time-series datasets and multivariate analysis allows for the comprehensive depiction of the sleep landscape.

a Large datasets (the UK Biobank, approximately 100,000 participants) were used to identify general sleep patterns. b Tri-axial acceleration data were processed using ACCEL to generate sleep-wake time series, from which 17 common sleep indices and 4 rhythm-related indices were calculated. c Clustering analysis was performed using statistical methods (UMAP + DBSCAN). Color-coded clouds in the cube represent identified clusters. d Each cluster exhibits a different index pattern. e Representative sleep-wake patterns of individuals from the clusters. Blue boxes indicate sleep. Clusters identified include long sleepers (upper middle left), fragmented short sleepers (upper middle right), irregular sleep patterns (upper right), and clusters containing various types of insomnia.