Identifying genes and cell types with dynamic alternative polyadenylation in multi-cluster single-cell transcriptomics data

Alternative polyadenylation (APA) causes shortening or lengthening of the 3-untranslated region (3-UTR) of genes across multiple cell types. Bioinformatic tools have been developed to identify genes that are affected by APA (APA genes) in single-cell RNA-Seq (scRNA-Seq) data. However, they suffer from low power, and they cannot identify APA genes specific to each cell type (cell-type-specific APA) when multiple cell types are analyzed. To address these limitations, we developed scMAPA that systematically integrates two novel steps. First, scMAPA quantifies 3-UTR long and short isoforms without requiring assumptions on the read density shape of input data. Second, scMAPA estimates the significance of the APA genes for each cell type while controlling confounders. In the analyses on our novel simulation data and human peripheral blood mono cellular data, scMAPA showed enhanced power in identifying APA genes. Further, in mouse brain data, scMAPA identifies cell-type-specific APA genes, improving interpretability for the cell-type-specific function of APA. We further showed that this improved interpretability helps to understand a novel role of APA on the interaction between neurons and blood vessels, which is critical to maintaining the operational condition of brains. With high sensitivity and interpretability, scMAPA shed novel insights into the function of dynamic APA in complex tissues. Key PointsO_LIWe developed a bioinformatic tool, scMAPA, that identifies dynamic APA across multiple cell types and a novel simulation pipeline to assess performance of such tools in APA calling. C_LIO_LIIn simulation data of various scenarios from our novel simulation pipeline, scMAPA achieves sensitivity with a minimal loss of specificity. C_LIO_LIIn human peripheral blood monocellular data, scMAPA identifies APA genes accurately and robustly, finding unique associations of APA with hematological processes. C_LIO_LIscMAPA identifies APA genes specific to each cell type in mouse brain data while controlling confounders that sheds novel insights into the complex molecular processes. C_LI