A universal framework for RNA modification characterization and annotation using deep neural networks and nanopore direct RNA-sequencing (PRJCA040561)

We aim to profile transcriptome-wide RNA modifications using nanopore direct RNA sequencing. In this project, we developed a deep learning framework, ORCA, which enables unbiased detection and annotation of diverse RNA modification types. ORCA integrates domain adversarial learning and transfer learning to capture stoichiometry-driven signal patterns and accurately predict modification types with minimal prior knowledge. Using this method, we generated high-resolution RNA modification maps across multiple human cell lines. The data deposited here support epitranscriptomic research by providing isoform-resolved modification profiles.