Overview
- A Stanford team published a Science paper Thursday reporting that genome‑language models called Evo 1 and Evo 2 generated thousands of complete viral genome candidates and that 16 of 302 synthesized sequences produced bacteriophages that killed Escherichia coli in high‑security lab tests.
- The models were trained on millions of natural sequences from viruses, bacteria, plants and humans and used the well‑studied phage ΦX174 as a template to guide generation of new genome architectures.
- The experimental success rate was low: researchers synthesized 302 designs from thousands generated and saw only 16 functional phages, showing the approach can produce viable genomes but does not yet deliver routine or predictable designs.
- The team limited the work to bacteriophages, used high‑security facilities and excluded sequences for viruses that infect complex organisms, but the release of Evo 2 as open‑source has prompted experts to call for stronger oversight and built‑in risk controls.
- Researchers say the method could speed development of tailored phage therapies for antibiotic‑resistant infections, but it will require deeper biological understanding, stricter governance and new safety standards before clinical use or wider adoption is possible.