Researchers release RNA model trained on 21 million sequence-structure pairs

Researchers published structRFM, an RNA foundation model trained on 21 million sequence-structure pairs, and made its dataset, model weights and source code openly available. The Nature study reports improved RNA structure and function predictions, including adaptation to newly released RNA families. The model learns from both RNA sequences and the interactions that pair their bases. Its Zfold adaptation produces three-dimensional predictions with performance comparable to AlphaFold 3 across CASP15-natural, CASP16 and RNA-Puzzles. Those comparisons cover 8, 13 and 20 RNAs, respectively.