RNA Models
RNA structure, function, and expression modeling
73 models in this category
What RNA foundation models do
RNA foundation models learn the sequence, structural, and functional diversity of the transcriptome, tackling tasks that range from secondary structure prediction and splicing code modeling to non-coding RNA classification and mRNA design optimization. Unlike DNA, RNA folds into complex functional structures — hairpins, pseudoknots, ribozyme active sites — meaning that structure-aware representations are often essential rather than optional. Models like RNA-FM are pretrained on tens of millions of RNA sequences spanning multiple RNA families, giving them broad coverage of the transcriptome's functional landscape.
Applications: splicing, structure, and therapeutic design
Splicing prediction is one of the highest-demand applications: models trained on pre-mRNA sequences can score cryptic splice site activation by variants, a key step in interpreting pathogenic mutations. Secondary structure prediction benchmarks like bpRNA and ArchiveII are standard evaluation sets for RNA structure models. On the therapeutic side, RNA foundation models are increasingly applied to mRNA vaccine design — optimizing codon usage, secondary structure, and untranslated region elements to improve stability and expression.
Notable Models
Top-rated rna models from our evaluations
Hybrid framework that predicts ribosome location profiles from mRNA sequence alone, pairing a structure-aware TASEP simulation with a Mamba polisher.
Mamba-based mature RNA foundation model, contrastively trained on splice isoforms and 400+ mammalian species orthologs for mRNA property prediction.
RNA inverse-folding model that generates sequences predicted to fold into a target 3D backbone, capturing non-canonical pairs and tertiary motifs.
RNA foundation model pretrained on 23.7 million non-coding RNA sequences, producing embeddings for structure prediction, annotation, and RNA design.
RNA language model that learns base-level embeddings capturing sequence context and secondary structure, enabling fast structural alignment.
RNA foundation model pretrained jointly on sequences and secondary structures for structure prediction, homology and splice site classification.
Frequently asked questions
What is an RNA foundation model?
An RNA foundation model is a large neural network pretrained on collections of RNA sequences — coding and non-coding — to learn representations of RNA sequence, structure, and function. These representations support downstream tasks including secondary structure prediction, splicing classification, and functional annotation of non-coding RNAs. RNA-FM is a well-known example trained on diverse RNA families.
How do RNA foundation models handle secondary structure?
Some RNA models, like those benchmarked on bpRNA, are trained with explicit base-pair annotations and learn to predict contact maps or dot-bracket structures directly. Others learn structural priors implicitly through sequence co-evolution patterns, similar to how protein language models capture contact information. The best approaches for therapeutic design often combine a pretrained sequence encoder with a structure-prediction decoder fine-tuned on experimentally determined structures.
Can RNA foundation models help with mRNA vaccine design?
Yes, and this has become a commercially relevant application. Optimizing codon usage, 5' and 3' UTR sequences, and mRNA secondary structure for stability and translation efficiency are all tasks where foundation model embeddings and generative models have shown promise. Several groups have reported that model-guided design outperforms simple codon optimization heuristics on in vitro stability benchmarks, though large-scale clinical validation is still limited.
What distinguishes RNA models from DNA models?
RNA models must account for the fact that RNA folds into functional three-dimensional structures with strong sequence-structure coupling — a constraint less central to most DNA modeling tasks. RNA models also often span diverse functional classes (mRNA, tRNA, rRNA, lncRNA, miRNA) that have very different sequence statistics, making broad pretraining corpora and multi-task objectives more important than in narrower DNA regulatory sequence models.