Fig. 1: Discovering circRNAs from large-scale full-length single-cell RNA-seq datasets.

a The total number of cells and circRNAs detected in the collected scRNA-seq datasets. b Workflow of scRNA-seq data integration and circRNA detection (see Methods). c Overlap of circRNAs detected in the scRNA-seq datasets, circAtlas database, and the integration of other 10 bulk RNA-seq based databases. d The average expression levels (counts per million, CPM) of circRNAs in the circAtlas database. Colors represent circRNAs that were uniquely detected in circAtlas (purple, n = 730,657) or simultaneously detected in circAtlas and the scRNA-seq cohort (blue, n = 112,075). e The length of fully assembled sequence of 619,060 circAtlas-specific and 103,758 circRNAs shared between circAtlas and the scRNA-seq cohort. f The number of species that circRNAs were conservatively expressed in. g Log-scaled mean expression level and the number of expressing cells for mouse circRNAs. Sizes of points indicate the number of expressing cells. Filled colors represent the mean CPM of circRNAs. h, i The number of circRNA host cells (h) and mean BSJ reads (i) of circRNAs (n = 239,471) that were uniquely detected in the scRNA-seq data or validated circRNAs (n = 114,919) which were also observed in bulk RNA-seq circRNA databases. j Cumulative distribution of mouse circRNAs ordered by the number of expressing cells. The scRNA-seq specific and validated circRNAs are colored in red and grey, respectively. k Expression of scRNA-seq specific circRNAs in cells that have more than 10 BSJ reads in total. The x- and y-axis represent the proportion of scRNA-seq specificcircRNA number and BSJ reads in these cells, respectively. Colors indicate the density of circRNAs at each point. All center lines in the box plots indicate the median values, and box limits indicate the upper and lower quartiles of plotted values. The upper and lower whiskers indicate the largest and smallest values within the range of 1.5x interquantile range (IQR) distance from the box limits. ****P < 0.0001, Wilcoxon rank-sum test (two-sided). Source data are provided as a Source Data file.