Fig. 5.

TimeCycle Biological Data Results. (A) TOP: Upset plot showing overlap between significant cycling genes sets detected by TimeCycle at an FDR < 0.05 across 3 distinct mouse liver time-series datasets as described in (Ness-Cohn et al., 2020).The Zhang and Hughes datasets were sampled every 2-h for 48-h. The Hogenesch study, sampled every 1-h for 48-h, was downsampled into two datasets sampled every 2-h for 48-h (Hogenesch 2A and Hogenesch 2B). BOTTOM: Distribution of FDR-adjusted P-values of the known circadian genes (orange) versus all genes (gray). Circadian genes were extracted from the CGDB database (Li et al., 2017). Only circadian genes that were experimentally validated in mouse liver tissue through low-throughput methods were included in the analysis. (B) Percentage of genes concordantly called cycling or non-cycling across studies at varying FDR thresholds in each pair of studies; comparisons to other methods are also shown. (C) LogFC amplitude, period and phase scatterplot comparison of genes identified as cycling in both the Zhang and Hughes datasets with an FDR < 0.05. (D) Heatmap of the cycling genes detected in each dataset with an FDR < 0.05 ordered by phase. Orange and gray represents gene expression above and below diurnal mean, respectively. (E) Rank correlation CDFs of genes identified by each method with an FDR < 0.05 and LogFC < 2 in one dataset compared to their pairwise expression in the other datasets. The null distribution represents the rank correlation of a random sampling of all 12 868 genes with replacement in the pairwise comparisons of all datasets. All gene correlation represents the rank correlation on a per gene basis for all 12 868 genes in the pairwise comparisons of all datasets.