Hugging Face releases Carbon-X, an open-source, 14B-parameter multimodal AI model to let users query DNA sequences via natural language, trained on 300B tokens

You can now query DNA sequences using natural language, which was previously impossible without specialized bioinformatics tools.

Hugging Face, in collaboration with researchers from China and Italy, has released Carbon-X, an open-source, 14-billion-parameter multimodal AI model. Trained on 300 billion tokens of genomic sequences and natural language, Carbon-X enables users to query DNA sequences using natural language. This allows for conversational answers regarding the effects of DNA changes on gene activity and health. Lewis Tunstall, a senior research scientist at Hugging Face, stated that the tool aims to assist researchers in developing new drugs and genetic therapies, and potentially help patients interpret their own genomes. Unlike previous models like Google DeepMind's AlphaGenome, which outputs numeric data, Carbon-X offers a conversational interface. The model can process DNA sequences of up to approximately 50,000 base pairs and was intentionally designed to run locally on user hardware to ensure data privacy, avoiding the need to upload sensitive patient data to cloud-based models.

The release of Carbon-X signifies a growing trend towards making complex biological data accessible through intuitive, conversational AI interfaces. By integrating genomic data with large language models, Hugging Face is democratizing access to genetic information, potentially accelerating drug discovery and personalized medicine research. This approach lowers the barrier to entry for researchers and clinicians who may not have deep expertise in bioinformatics, enabling broader participation in genomic analysis.

This development impacts how researchers interact with genomic data, moving from numerical outputs and specialized software to natural language queries. The model's local execution capability addresses privacy concerns associated with sensitive patient data, a critical factor for widespread adoption in clinical settings. Future developments may see even more sophisticated models that can handle larger genomic segments or integrate additional biological data types, further blurring the lines between AI and biological research.