Fig. 1

Developing SCAPTURE to identify cleavage and polyadenylation sites (PASs) from 3′ tag-based scRNA-seq. a Comparison of human PBMC transcriptome profiling with different deep sequencing datasets. Wiggle tracks show an enrichment of 10x Chromium reads (rose) at the 3′ end of the GAPDH gene locus, close to its known PAS (GENCODE), while reads of TruSeq RNA-seq (gray) and Smart-seq2 (dark blue) cover the whole gene body. Data were retrieved from published PBMC TruSeq RNA-seq, Smart-seq2, and 10x Chromium [17]. b Distribution of deep sequencing reads on mRNA genes. Pileup of deep sequencing reads from the same published datasets (a) indicates enrichment of 10x Chromium reads (rose) at 3′ ends of genes, compared to coverage of gene bodies by TruSeq RNA-seq (gray) and Smart-seq2 (dark blue) data. The distribution of PASs on mRNA genes were annotated in GENCODE. c Schematic of a stepwise SCAPTURE pipeline for single-cell PAS calling, filtering, transcript calculating, and APA analyzing. Top left, calling peaks from 3′ tag-based scRNA-seq (the first step). Top right, identifying high-confidence PASs with an embedded deep learning neural network DeepPASS (the second step). Bottom right, quantifying PASs to represent transcript expression at a single-cell resolution (the third step). Bottom left, applying SCAPTURE to APA analysis at single-cell level (the fourth step). See “Methods” section for details