Key Findings
The structRFM model establishes new benchmarks for RNA secondary-structure prediction and ranks highly in zero-shot homology classification. A key achievement is the development of Zfold, a tertiary-structure adaptation that consistently improves three-dimensional predictions and their extracted secondary structures. Performance metrics for Zfold are comparable to AlphaFold3 across the CASP15-natural, CASP16, and RNA-Puzzles datasets. Additionally, the model demonstrates improved performance in function-prediction tasks and generalizes to long non-coding RNAs and newly released RNA families.
Technical Details
structRFM was trained on 21 million RNA sequence–structure pairs. The architecture integrates base-pairing interactions into masked language modelling using a pair-matching operation. To enhance the model, the researchers introduced two specific mechanisms:
- MUSES: A multi-source ensemble of secondary structures designed to mitigate predictor bias.
- Dynamic Masking Ratio: A system that balances nucleotide-level masking with structure-guided masking.
The model generates joint sequence and structure representations as pairwise, nucleotide-level, and sequence-level features to allow for broad downstream adaptation.
Background & Context
While RNA language models have utilized large-scale sequence data, RNA function is fundamentally shaped by hierarchical structure. Previous approaches often relied on task-specific objectives or noisy annotations, which limited their generalizability. To address this, structRFM utilizes a massive dataset sourced from RNAcentral v24.0, Rfam v14.10 and v15.0, PDB, and the NCBI nucleotide (nt) database, among others.
Significance & Outlook
The results highlight the effectiveness of multimodal, structure-guided pre-training for robust RNA inference. To support further development of RNA foundation models, the complete 21-million-pair dataset, model weights, and source code have been made openly available via Zenodo and GitHub.
Source: https://www.nature.com/articles/s41586-026-11210-2
Get our weekly technology intelligence — free
Receive an infographic that lets you judge at a glance whether each field’s analysis report is worth reading.
Subscribe Free — Weekly Tech Intelligence
By subscribing, you’ll receive Troy-Technical’s weekly technology intelligence newsletter.
- Your email and selected fields are used only to deliver the newsletter.
- We never share your information with third parties.
- You can unsubscribe anytime via the link in each email.
See our Privacy Policy for details.
Takes about a minute · Unsubscribe anytime
