MarinDNA

Open development of genomic language models — data, modeling, and evaluation.

_{Inspired by Marin.}

News

• 2026-08-03 — Blog post — A 1B standard Transformer rivals Evo 2 40B on variant effect prediction.
• 2026-05-26 — Poster — Data curation strategies for genomic language models.

Research

These documents synthesize MarinDNA's current answers and help organize future experiments.

Current priorities

• Complex-trait VEP
• Genomic anchor selection and projection
• RAG

Other active questions

• Bidirectionality
• Context size
• Data mixing
• Evolutionary timescales
• Latent biological features
• Sequence-to-function modeling
• Species conditioning
• Tokenization
• Training regions

Resources

• Models and datasets on Hugging Face
• Variant effect prediction leaderboard
• Interactive sequence explorer
• Model inference and BRCA1 variant effect prediction notebook

Community

Join the Marin Discord; MarinDNA discussion happens in the #dna channel.

Citation

If you find datasets, models, or experiments from this repo useful, please cite:

MarinDNA: open development of genomic language models. Open Athena, 2026. https://github.com/Open-Athena/marin-dna

BibTeX:

@misc{marin-dna,
  title  = {MarinDNA: open development of genomic language models},
  author = {{Open Athena}},
  year   = {2026},
  url    = {https://github.com/Open-Athena/marin-dna},
}