Hugging Face Debuts Carbon-X for Conversational DNA Analysis
AI-generated image · US National Wire
The new open model allows researchers to query genomic sequences using natural language, though developers note a lack of wet lab validation.
Hugging Face has released Carbon-X, an open AI model designed to allow users to "chat with DNA" to understand how genetic changes impact health and gene activity, according to reporting from Wired.
Developed alongside researchers from the University of Naples Federico II in Italy and the Zhongguancun Academy in China, the model integrates genomic sequences with natural language. This allows nontechnical users to ask questions about DNA through a conversational interface, a capability Lewis Tunstall, a senior research scientist at Hugging Face, says could assist in the design of genetic therapies and new drugs.
To build Carbon-X, the team expanded an existing language model's vocabulary to read DNA in six-letter groups, training it on 300 billion tokens. This dataset included English text, biological descriptions, and DNA sequences from protozoa, fungi, plants, invertebrates, and vertebrates, including humans. Tunstall noted that viruses were excluded from training to prevent dangerous applications.
Unlike Google DeepMind's AlphaGenome, which provides numeric predictions, or the Arc Institute and Nvidia's Evo 2, which can generate genomes, Carbon-X focuses on accessibility. The model features 14 billion parameters, allowing it to run locally on a laptop to ensure sensitive patient data is not sent to the cloud. Users are limited to querying DNA sequences of approximately 50,000 base pairs.
Despite the utility, Wired reports that the developers have not yet validated the model through wet lab experiments, meaning the accuracy of its predictions remains unverified.