Deep Learning for RNA–RNA Interaction Prediction

RNA–RNA interactions (RRIs) play a central role in post-transcriptional gene regulation, influencing processes such as translation, splicing, RNA stability, and ribonucleoprotein complex assembly. While accurate computational prediction of RRIs could pave the way for RNA-targeted therapies, it remains a major challenge due to the intricate and dynamic behavior of long RNA molecules in vivo. To address these challenges, we introduce RIME, a deep learning framework that predicts RRIs using only sequence information. RIME leverages embeddings from the Nucleotide Transformer language model, which capture complex biological patterns beyond conventional thermodynamics-based features. Across multiple datasets, RIME consistently outperforms existing tools and successfully highlights key sequence determinants of RNA interactions, such as low-complexity repeats, as confirmed by enrichment analyses. Notably, the model excels in predicting high-confidence and functionally validated interactions, demonstrating its ability to extract meaningful signals from the complex sequence landscape of long RNAs. The code implementing RIME is freely available at: https://github.com/giorgiobini/ RIME. A web server is also accessible at: https://tools.tartaglialab.com/rna_rna.