| name | nucleotide-transformer |
| description | InstaDeep Nucleotide Transformer family — DNA foundation models from NT/NT-v2 through NTv3 (multi-species, long-context seq–function and generation). Use when: (1) DNA embeddings or MLM scores on contigs/MAGs, (2) Functional-track / genome-annotation style prediction with NTv3, (3) Contrasting representation gLMs vs generative Evo 2 or metagenomic gLM2. Upstream: https://github.com/instadeepai/nucleotide-transformer. HF: InstaDeepAI. Route via genome-language-model. Check CC BY-NC-SA license for your use case.
|
| license | MIT |
| category | analysis-tools |
| tags | ["Nucleotide-Transformer","NTv3","gLM","DNA","InstaDeep","HuggingFace"] |
| upstream | https://github.com/instadeepai/nucleotide-transformer |
| stage | function |
Nucleotide Transformer (NT / NTv3)
Upstream: instadeepai/nucleotide-transformer ·
HF: https://huggingface.co/InstaDeepAI
Citation
Dalla-Torre, H. et al. Nucleotide Transformer: building and evaluating robust
foundation models for human genomics. Nat. Methods 22, 287–297 (2025).
https://doi.org/10.1038/s41592-024-02523-z
NTv3: A foundational model for joint sequence-function multi-species modeling at
scale for long-range genomic prediction. bioRxiv
https://doi.org/10.64898/2025.12.22.695963 (2025).
See also docs/references.md.
Analytical thinking
NT / NT-v2: representation DNA LMs (embeddings, phenotype-related
fine-tunes). NTv3: unifies representation learning, functional-track /
annotation prediction, and controllable generation; U-Net-like, single-base
tokens, contexts up to ~1 Mb; pretrained on OpenGenome2 then post-trained on
multi-species tracks.
| vs | Prefer NT family when |
|---|
evo2 | Embeddings / tracks / annotation heads (not primarily AR design) |
omg / gLM2 | Multi-species DNA tracks; not specifically OMG mixed AA+DNA corpus |
mmseqs2 | Learned DNA features vs classical homology |
License is often CC BY-NC-SA — confirm before commercial use. Prefer
multi-species / OpenGenome-pretrained checkpoints for microbes over human-only
variants.
How to run
Pin exact checkpoint (e.g. NT-v2 vs NTv3 size) and max context in Methods.
Decision tree
DNA foundation model need?
├─ Embeddings / NTv3 tracks & annotation → nucleotide-transformer
├─ Efficient multi-species / GUE → dnabert2
├─ Long-range RC-aware → caduceus
├─ Long-context generative design (all domains) → evo2
├─ Metagenomic mixed-modality gLM2 → omg
└─ Protein homology → homology-search / plmsearch
Related skills
genome-language-model · dnabert2 · caduceus · evo2 · omg ·
metagenomics-llm · homology-search · tool-selection
