EVO2 is a Large Language Model trained using 9 trillion DNA base pairs from all domains of life. It can, their Nature paper says, predict everything from noncoding pathogenic mutations to clinically significant BRCA1 variants, and they just proved it. By creating 16 viruses that don't exist in nature utilizing natural phage ΦX174.
In new work, they generated genome combinations designed to infect E. coli bacteria, then they created 300, and 16 were actually viable. Their AI designed the entire genome.
There will always be concern about science, because most people, and certainly activists, don't have any expertise, but the viruses were created to target bacteria, not humans. There weren't even any human pathogens for people in the training data, and these aren't Chinese scientists selling stuff in a wet market, so this could end up really helping the next time we need a therapy against rapidly mutating pathogens - or even antibiotic-resistant ones.
Sequences generated and computationally evaluated with design criteria inspired from the phage ΦX174 and its host E. coli C. The most promising designs are chemically synthesized and tested in host cells, yielding viable generated phage genomes divergent from known natural phages, while retaining host specificity.
Citation: Samuel H. King, Claudia L. Driscoll , David B. Li, Daniel Guo, Aditi T. Merchant, Garyk Brixi, Max E. Wilkinson, and Brian L. Hie,Generative design of bacteriophages with genome language models.Science393,eaec2657(2026).DOI:10.1126/science.aec2657
