RNA modifications are post-transcriptional events in which RNA-binding proteins alter a particular nucleotide in a transcribed RNA strand, affecting its activity, location, and stability1. Up-to-date, more than 100 types of RNA modifications have been identified, with some even linked to the development of cancers, cardiovascular disorders, and other diseases2-4. There have been huge technological advances in recent years that have radically expanded our ability to detect these modifications5, however most analysis pipelines are inherently restricted to known modification motifs. In this study, we develop a deep learning framework for the accurate identification of RNA sites likely to undergo any of seven different modification types, including N6-methyladenosine (m6A), Pseudouridine (ψ), 1-Methyladenosine (m1A), 2’-O-methyladenosine (Am), 2’-O-methylcytidine (Cm), 2'-O-methylguanosine (Gm), and 2’-O-methyluridine (Um). Data were curated from publicly available experimental datasets6, and represented using one-hot encoding of sequences, chemical and conservation properties7, and optimised versions of transformer-based machine learning techniques developed for natural language processing (RNABERT)8. Across cross-validation and independent blind tests the model performed strongly, providing a powerful base for better understanding RNA modification sites, and for use in genome wide predictive mapping.