NucleicBERT: Deciphering the language of nucleic acids by a large-language model

The vast majority of the human genome comprises non-protein-coding regions whose structural and functional roles remain poorly understood. Many of these regions function through RNA, yet progress in deep learning for RNA has lagged behind proteins because most methods rely on abundant structural labels or evolutionary alignments, both sparse for RNA. To address these challenges, we developed NucleicBERT, a self-supervised masked-language model that learns contextual representations capturing local and distal dependencies without requiring alignments or evolutionary information. Explainable AI analysis reveals that the model clusters RNA types in latent space and attends to structural properties like secondary structure and tertiary contacts, effectively “rediscovering” RNA biology from sequence correlations alone. When fine-tuned for downstream structural and functional tasks, NucleicBERT requires only single sequences, yet surpasses current state-of-the-art RNA models. This alignment-free framework addresses the scarcity of annotated 3D RNA data while providing a rapid, computational complement to experimental techniques. By bridging abundant unlabeled primary sequence corpora with more scarce structural annotations, NucleicBERT advances RNA structure prediction and provides insights into the working of LLMs. NucleicBERT is available at https://github.com/KIT-MBS/NucleicBERT .