NanoRAPID: A Deep Learning-based Framework for Single-molecule RNA Structure Analysis Using Nanopore Direct RNA Sequencing

RNA structure is fundamental to its diverse biological functions. Current chemical probing methods, coupled with next-generation sequencing, offer insights into RNA secondary structure but are limited by indirect readouts and averaging of multiple molecules. Nanopore direct RNA sequencing (DRS) enables direct detection of modifications on long RNA reads, while accurately identifying probe-modified sites from DRS data remains challenging. Here, we present NanoRAPID (Nanopore RNA Structural Probe IDentification), a convolutional neural network (CNN)-based framework for analyzing RNA secondary structures using DRS data. By directly analyzing raw current signals, NanoRAPID achieves improved accuracy in probe-site identification. Validation with DRS datasets, including NAI-N3 [2-(azidomethyl)nicotinic acid imidazolide] and diethyl pyrocarbonate (DEPC) treatments, demonstrates the robustness and transferability of NanoRAPID. Transcriptome-wide analysis reveals distinct structural features across RNA categories, with mRNAs exhibiting variable structures in gene bodies and untranslated regions (UTRs), in contrast to the compact and uniform architecture of rRNAs. Notably, NanoRAPID identifies isoforms with distinct structural conformations, suggesting a dynamic equilibrium between conformational states. Furthermore, we observe a positive correlation between N6-methyladenosine (m6A) modification levels and 3' UTR accessibility, suggesting that RNA structural context may influence m6A deposition. NanoRAPID is a precise and versatile tool for detecting RNA structural probing signals from DRS data and is freely available on GitHub: https://github.com/luolab-sysu/NanoRAPID and BioCode: https://ngdc.cncb.ac.cn/biocode/tools/BT008087.