USTC team publishes structRFM in Nature, an open RNA model that learns how RNA folds, with weights and data released
A University of Science and Technology of China team, with a collaborator from Zhejiang University, published structRFM in Nature on November 4. What an RNA molecule does depends heavily on how it folds, meaning which of its bases pair up, yet most RNA language models learn from sequences alone. structRFM builds base-pairing information into pre-training, using 21 million RNA sequence–structure pairs. The paper reports that Zfold, a version adapted for 3D structure prediction, performs comparably to AlphaFold3 on the CASP15-natural, CASP16 and RNA-Puzzles test sets. The model also helps predict RNA function and carries over to newly released RNA families and long non-coding RNAs. The training data, model weights and code are all public.
Why it matters
This is AI research in Nature's main journal: its 3D structure predictions are on par with AlphaFold3, and the training data, weights and code are all public, so RNA researchers can use them directly.
Sources1
Written by dalao.ai: the title, summary and why it matters are our own words. The original reports are the record; their publishers keep authorship.