NucleicBERT interprets RNA sequence space through self-supervised language modelling

NucleicBERT is a large language model trained on 30 million noncoding RNA sequences via masked language modeling, enabling it to predict RNA structure and function directly from single sequences without relying on evolutionary data. This approach addresses the scarcity of RNA structural data and computational challenges of existing methods, offering a scalable alternative for RNA structure-function prediction and aiding RNA-targeted drug discovery.