Skip to main content


Fig. 3 | Genome Medicine

Fig. 3

From: Machine learning identifies a compact gene set for monitoring the circadian clock in human blood

Fig. 3

Applying ZeitZeiger to gene expression from perturbations of the sleep–wake and light–dark cycles. For each of the three datasets, a predictor was trained on control samples from the other two datasets, then tested on all samples from the dataset of interest. a Violin plots of error and (b) boxplots of absolute error for each condition in each dataset. Numbers above the plots indicate P values < 0.1 (circular ANOVA for error and Wilcoxon rank-sum test for absolute error). For ease of visualization, boxplots of absolute error do not show three outliers in GSE39445, one in GSE48113, and three in GSE56931. The left-most condition in each dataset is the control. In GSE39445, “restriction” refers to seven nights of 6-h sleep opportunity. In GSE48113, “out of phase” refers to the forced-desynchrony protocol in which sleep is out of phase with melatonin and the central clock. In GSE56931, “deprivation” refers to one full day of sleep deprivation and “recovery” refers to a normal sleep opportunity the next night

Back to article page