Hugging Face’s New AI Biology Model Lets You Chat With DNA

Hugging Face and researchers in China and Italy have developed Carbon-X, an AI model that lets people ask questions about DNA in ordinary language. The team sees it as a possible aid to research into drugs and genetic therapies, and perhaps to patients interpreting genomic information, though its predictions have not yet been validated in laboratory experiments.

Carbon-X was trained on 300 billion tokens combining DNA sequences and text, and can process sequences of about 50,000 DNA letters at a time. Its 14-billion-parameter size is intended to let it run locally, so users’ data is not sent to a company; it was not trained on viruses, and full human genomes cannot be uploaded. Unlike tools that return numerical predictions or generate DNA, Carbon-X combines DNA and natural language in a conversational interface.