structRFM model trains on 21 million RNA pairs
structRFM is a new structure‑guided RNA foundation model trained on 21 million RNA sequence‑structure pairs. The authors add a multi‑source ensemble, MUSES, to reduce predictor bias and use a dynamic masking ratio that balances structure‑guided and nucleotide‑level masking. The model learns joint sequence and structure representations at the sequence‑, nucleotide‑ and pairwise levels.
Key points
- structRFM trained on 21 million RNA sequence‑structure pairs
- MUSES ensemble reduces predictor bias and dynamic masking balances structure and nucleotide masking
- Zfold matches AlphaFold3 performance on CASP15‑natural, CASP16 and RNA‑Puzzles
In zero‑shot homology classification, structRFM ranks highly and sets new benchmarks for secondary‑structure prediction. Its tertiary‑structure adaptation, Zfold, improves three‑dimensional predictions and extracted secondary structures, achieving performance comparable to AlphaFold3 across CASP15‑natural, CASP16 and RNA‑Puzzles. The model also boosts function‑prediction tasks and generalises to newly released RNA families and long non‑coding RNAs.
The 21‑million‑pair dataset, model weights and source code are openly available. The pre‑training dataset can be downloaded from Zenodo (https://doi.org/10.5281/zenodo.18272040) and the model checkpoints from GitHub (https://github.com/heqin-zhu/structRFM).
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Ricursive Intelligence releases Ripple-65B Preview, a 65B model with 1M context · 1 src
- Pangram launches AI image and video detectors with 99.96% accuracy · 2 src
- Kimi-K3.1-Pro is now open-weights · 1 src
- Cognition says SWE-2 Flash runs three times faster and costs one-fifth · 1 src
- A new way of measuring AI models suggests China's threat to OpenAI and Anthropic may have been overblown · 4 src
Comments
via GitHub Discussions