USTC Releases Open-Source RNA Sequence Foundation Model structRFM, Published in Nature

Researchers at the University of Science and Technology of China (USTC) have released structRFM, an open-source RNA sequence foundation model that integrates structural knowledge into large-scale sequence learning. Published in Nature, the model significantly improves performance across various RNA structure prediction and functional inference tasks. This breakthrough advances AI-driven biological research by providing a robust tool to understand the complex relationship between RNA sequences, structures, and functions. By releasing the complete paired data, model weights, and source code, the team provides valuable open resources for drug discovery and biological mechanism studies. Trained on approximately 21 million computationally annotated sequence-secondary structure paired data, structRFM uses a structure-guided pre-training framework. It can extract sequence-level, nucleotide-level, and nucleotide-pair-level features from just an input sequence to excel in tasks such as homology classification, secondary/tertiary structure prediction, and splicing site identification.