Integrating Mutation and Stop Signals for Improved RNA Structure Analysis and Insight Discovery

Background: Understanding RNA structures is essential for exploring its diverse cellular roles. Chemical modification-based RNA structure probing remains a key approach to revealing RNA structures in complex in vivo conditions. Practically, chemical modifications generate both mutation and stop signals during reverse transcription within a single experiment. However, traditional analysis pipelines often rely on only one of the two signals without fully leveraging both. Results: To address this, we developed a machine learning-based approach, STONE, that integrates both signals from a single experiment. STONE improved RNA structure analysis across multiple independent validation regions, notably 28S rRNA, viral RNA and regulatory RNAs. In genome-wide datasets, especially single-cell data, STONE significantly increased nucleotide coverage per transcript and improved gene detection reliability. In genome-wide datasets, especially single-cell data, STONE substantially enhanced structural information coverage at both transcript and nucleotide levels. This maximized signal utilization, yielding RNA structures in single-cell data comparable to those from bulk datasets. Furthermore, STONE-derived structural scores allow direct identification of RNA-binding protein binding sites directly from a single probing experiment, with results on RNAs such as U1 snRNA, RNase P RNA, and XIST lncRNA closely matching established techniques like CLIP-seq and RNP-MaP. Conclusions: By integrating mutation and stop signals from one single experiment, STONE advances RNA structure analysis accuracy, extends nucleotide coverage, and facilitates complex RNA-protein interaction studies, broadening the method's applications in RNA-based research.