Hugging Face Releases New Open Source Model Enabling DNA Analysis

Hugging Face released an open-source artificial intelligence model named Carbon-X that combines genomic sequences with natural language, allowing researchers to chat with DNA and analyze genetic variations.

The company developed Carbon-X in collaboration with research labs in China and Italy by training an existing language model on genomic sequences and text from humans, other mammals, plants, and fungi.

The tool arrives as the scientific community navigates a landscape where advanced artificial intelligence capabilities remain concentrated within a small group of heavily funded private technology laboratories.

“Carbon-X allows researchers to ‘chat with DNA’ for the first time: They can provide a sequence and ask questions about it in natural language,” said Lewis Tunstall, a senior research scientist at Hugging Face.

The 14 billion parameter model is designed to run locally on a user's own hardware rather than on remote cloud servers, which helps protect sensitive or unpublished patient data.

“Researchers need models they can study, adapt and run with sensitive or unpublished sequences on their own hardware,” said Tunstall.

Viruses were intentionally excluded from the training data to prevent the model from being misused to design biological weapons.

Business Insider reported that the model can evaluate whether specific mutations increase a person's risk of disease and can generate new DNA sequences to aid in developing gene therapies.

Citing company blog posts, CNET reported that Carbon-X can produce hallucinations and make errors through miscounts, while output quality degrades when users request excessively long DNA sequences.

The release follows an agreement in September in which Nvidia announced plans to acquire Hugging Face for nearly $13 billion.

Developers caution that the model has not yet been validated through actual wet lab experiments, leaving its real-world prediction accuracy unconfirmed.