Accurate profiling of single-cell alternative transcript start sites by correcting RNA degradation.

The generation of transcript variants via alternative utilisation of transcription start sites (TSSs) is a pivotal regulatory mechanism in physiological and pathological states. Recent advancements in 5' single-cell RNA sequencing (scRNA-seq) have enabled TSS analysis at the single-cell level. However, RNA degradation leads to non-uniform read coverage, posing a critical challenge that significantly compromises accurate TSS quantification of scRNA-seq data. To address RNA degradation and improve TSS quantification, we develop scATS (single-cell alternative transcription start site) to estimate RNA degradation at both isoform and sample levels, and provide TSS quantification with or without degradation correction. Application of scATS reveals dynamic and context-dependent regulation of TSSs in haematopoiesis and disease, providing additional information on TSS isoforms that aids cell clustering at a finer resolution. Furthermore, we establish a machine-learning pipeline, lung cancer relevance score (LRS), to identify TSSs associated with lung cancer. We analyse TSS isoforms of CCR6, CCR2 and RTKN2 in lung cancer cell lines and confirm that isoforms highly transcribed in lung cancer promote cell proliferation and migration. Combined, we present a robust tool to accurately quantify TSS by accounting for RNA degradation, a common issue that confounds transcript quantification, and experimentally demonstrate the important roles of TSS-mediated gene regulation in tumourigenesis.