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

Hugging Face, a platform for AI tools, has unveiled a new AI biology model called Carbon-X that enables users to chat with DNA sequences. Collaborating with researchers from China and Italy, the model is trained on genomic sequences and natural language, allowing people to ask questions about DNA and receive conversational answers.

Lewis Tunstall, a senior research scientist at Hugging Face, explains that the tool is designed to assist researchers in designing new drugs and genetic therapies, as well as help patients interpret their genomes.

Carbon-X stands out from other AI models as it can understand the language of DNA, which consists of the letters A, C, G, and T instead of the binary code 0s and 1s. The human genome contains 6 billion of these letters, and while scientists have sequenced the entire genome, the understanding of how most of it works remains limited.

AI models like AlphaGenome and Evo 2 are helping researchers decipher some of these mysteries, but Carbon-X takes it a step further by integrating DNA and natural language in a conversational format.

The model was trained on 300 billion data points, including DNA sequences, English text, and examples linking DNA to biological descriptions. The DNA sequences came from various organisms, including mammals, other vertebrates, invertebrates, plants, fungi, and protozoa. However, the model was not trained on viruses to prevent potential misuse.

Users will not be able to upload full human genomes as the model can only process DNA sequences of approximately 50,000 base pairs at a time. Instead, researchers can use Carbon-X to determine whether a genetic mutation is harmless or disease-causing, which is crucial for developing targeted drugs or therapies.

To ensure user privacy and security, the developers designed a smaller 14-billion-parameter model that can run locally on the user's hardware without sending any data to Hugging Face or other companies. Nonetheless, the team acknowledges that they have not yet validated the model's accuracy through wet lab experiments.