Ustc releases open-source rna foundation model structRFM, published in nature

Researchers from the Medical Imaging and Robotics Laboratory at the University of Science and Technology of China (USTC) have unveiled structRFM, an open-source foundation model for RNA sequences. This model enhances performance across various downstream tasks related to RNA structure prediction and functional inference, potentially aiding AI-driven scientific research and drug discovery. This achievement marks the 16th publication by USTC with the first author and corresponding affiliation in Nature, Science, or Cell since the beginning of 2026. The team, led by Professor Zhou Shaohua and Associate Researcher Xiong Peng, developed structRFM by training it on approximately 21 million annotated sequence-secondary structure pairs, integrating structural knowledge into large-scale sequence learning. StructRFM allows for feature extraction at sequence, nucleotide, and dinucleotide levels using only the input sequence. Experimental results show the model performs well in zero-shot homology classification and improves three-dimensional structure prediction when its features are incorporated. Furthermore, it supports functional inference tasks such as splice site prediction and non-coding RNA classification. The complete data, model weights, and source code have been made publicly available for reproducibility.