Skip to main content
Fig. 1 | Genome Biology

Fig. 1

From: ReadZS detects cell type-specific and developmentally regulated RNA processing programs in single-cell RNA-seq

Fig. 1

Overview of the ReadZS. A Read positions are ranked in equal-sized genomic bins, separated by read strand. Within each genomic window, the read distribution for each cell is summarized by a weighted, normalized function of read positions (“Methods”). With metadata, cell type-specific RNAP can be detected by finding windows with significantly different median ReadZS by cell type. Continuous metadata such as pseudotime enables discovery of multivariate relationships between ReadZS and metadata. GMM-based peak detection is used to compare read distributions with annotated 3′ UTRs. B Read distributions in the genomic window with largest effect size in HLCA P3 when requiring minimum 10 counts in 20 cells, which overlaps the genes CORO1B and PTPRCAP. Peaks in significant windows called by the GMM (see text) are starred. C CALM1 is called in both P3 and P2 as having cell type-specific differences in RNAP. Peaks in significant windows called by the GMM are starred; peaks are called across all cell types. In CALM1, the peaks are 254 and 285 bp from the closest downstream 3' UTR. D KLF6 is called in both P3 and P2 as having cell type-specific differences in RNAP. The relative rank of each cell type (ranked by highest to lowest median ReadZS) is shown for each participant. Peaks in significant windows called by the GMM are starred; peaks are called across all cell types. The ReadZS is technically reproducible across cell types in the 3 HLCA participants; p-values, computed via simulation (see “Methods”), show strong ReadZS concordance in all pairs. F Histogram and CDFs of the distribution of distances from GMM-called peaks to closest downstream annotated 3′ UTR in HLCA P3; lines denote the 25th, 50th, and 75th quantile, respectively. Distance distributions are compatible with expectation from 10x library construction. G Above: read distributions in the genomic window with largest effect size in HLCA P3 when requiring minimum 5 counts in 10 cells, which overlaps the genes CATIP1, respectively. Below: ReadZS distribution for the genomic window overlapping CATIP in four cell types from HLCA P3

Back to article page