Comprehensive mapping of RNA modification dynamics and crosstalk via deep learning and nanopore direct RNA-sequencing

Despite the extensive studies of individual RNA modifications, the lack of methods to detect multiple modification types simultaneously has left the global epitranscriptomic landscape and its underlying crosstalk largely unexplored. Here, we present ORCA (Omni-RNA modification Characterization and Annotation), a deep learning framework that enables comprehensive mapping of RNA modification landscape using nanopore direct RNA sequencing. ORCA employs domain adversarial learning to detect and quantify a wide range of modifications by leveraging mixed stoichiometry-driven signal and sequence variability between modified and unmodified nucleotides. It also incorporates a transfer learning module for accurate annotation of modification types with minimal prior knowledge. Applying ORCA to multiple human cell lines reveals widespread, isoform-specific modification patterns, as well as intricate cooperative and competitive interactions among neighboring modification sites. This approach substantially expands the repertoire of known RNA modification sites and elucidates their spatial organization, revealing the emerging roles of RNA modifications in splicing regulation. ORCA thus provides an unbiased and generalizable framework for decoding RNA modification dynamics and their regulatory complexity across diverse biological contexts. RNA modifications influence gene regulation, but global mapping was limited. Here, the authors introduce ORCA, a deep learning framework using nanopore RNA sequencing to detect multiple modification types, revealing isoform-specific patterns and regulatory interactions.