A new artificial intelligence model called Evo 2 can analyze and generate DNA sequences across bacteria, archaea, plants, animals, and other forms of life, marking a major step toward a unified AI system for biology. Developed by researchers at the Arc Institute and Stanford University, Evo 2 was trained on over 9.3 trillion nucleotides from 128,000 genomes, including 80,000 bacterial and 1,000 archaeal genomes, as well as data from plants and animals.
Evo 2 uses a deep learning architecture similar to large language models, but instead of processing text, it learns the patterns and grammar of DNA. The model can predict the effects of genetic mutations, design new DNA sequences, and even generate entire genomes from scratch. In tests, Evo 2 successfully predicted the impact of mutations in the BRCA1 gene, which is linked to breast cancer, and designed functional CRISPR-Cas enzymes.
The model is available as an open-source tool, allowing researchers worldwide to use and build upon it. The team behind Evo 2 hopes it will accelerate discoveries in genomics, medicine, and synthetic biology. However, they also acknowledge ethical concerns, such as the potential misuse of AI-generated DNA sequences, and have implemented safeguards to prevent harmful applications.
Evo 2 represents a significant advance over previous models, which were limited to specific organisms or types of data. By covering the entire tree of life, Evo 2 could help scientists understand evolutionary relationships, develop new antibiotics, and engineer organisms for sustainable bioproduction. The research was published in the journal Science on February 19, 2025.
