Benchmarking computational methods for identifying and quantifying polyadenylation sites from 3' tag-based single-cell RNA-seq data.

Alternative polyadenylation (APA) is a widespread post-transcriptional regulatory mechanism in eukaryotes, which contributes greatly in shaping transcriptome complexity and proteome diversity. The advancement of 3' tag-based single-cell RNA sequencing (scRNA-seq) technology has facilitated the emergence of various computational methods for identifying and quantifying polyadenylation sites (pAs) at the single-cell level. However, the lack of benchmarking complicates the choice of a suitable method. Here, we systematically benchmarked 10 methods using 9 simulated datasets and 25 real-world scRNA-seq datasets covering four sequencing protocols and three species of animals and plants. First, we proposed strategies based on prior pA annotations and base compositions around pAs to evaluate the sensitivity and accuracy of different methods on identifying pAs. Particularly, we evaluated the consistency of pA identification results across different methods, as well as the quality of unique pAs identified by each method. Furthermore, we assessed the performance of pA quantification of different methods using strategies based on correlation coefficients, cell type clustering, and differential APA detection. Finally, we evaluated computational resource consumption for each method. Practical guidelines were provided for choosing suitable methods, particularly with respect to factors such as sensitivity and accuracy in pA identification and quantification, the availability of prior pA annotations, scRNA-seq protocols, and species.