Symbolic imageAI models design working bacteriophage genomes
Researchers at Stanford University tasked two AI genome models with designing the genetic code of a new virus. Sixteen of the resulting bacteriophages proved viable in laboratory tests.
What happened
- FAZ reported on Friday, 7 August, that a team around Samuel King published the work in the journal "Science".
- The models Evo 1 and Evo 2 were trained on genetic code, not text, and used the phage ΦX174 as a template.
- The researchers picked 300 of several thousand generated genomes and rebuilt them in the lab from synthetic nucleic acids.
- The viable phages infected Escherichia coli bacteria and inhibited their growth.
- All designs differed from natural phages, yet matched the template by more than 90 percent.
What could happen next
- Open: whether genome language models can design viruses beyond bacteriophages, which infect only bacteria.
- FAZ files the work under biosecurity and calls the result impressive and frightening, naming no concrete safeguards.
- Earlier de novo attempts failed: per King's team, a single mutation can render even simple genomes unviable.
- Unclear how far the designs are genuinely new, given the high overlap with the natural template.