AI creates 16 new bacteriophages that infect bacteria

from scratch

An artificial-intelligence system generated functional bacteriophages—viruses that infect only bacteria—that had never existed in nature. Designing such viruses from the ground up could open new avenues for combating antibiotic-resistant bacteria, while also raising concerns about misuse for biological weapons.

Previously, scientists could synthesize viruses from scratch mainly by copying known pathogens or their variants for drug and vaccine development. In contrast, the new study, conducted by researchers at Stanford University and the Arc Institute, succeeded in getting a language model to design completely novel, functional viruses based on genetic information from millions of animals, plants and microorganisms.

models and methodology

The work relied on Evo 1 and Evo 2, foundational models built for computational biology. Both models were trained on millions of genomes across the tree of life to detect complex evolutionary patterns, including gene organization and biological constraints that keep an organism functional. The researchers chose bacteriophages because of their relatively small genomes, which are easier to synthesize and manipulate in the lab.

The reference virus for the experiment was the bacteriophage Phi X-174, which infects the bacterium *E. coli*. The goal was not to replicate Phi X-174 but to use it as a template so that the algorithms could generate thousands of entirely new genomes whose genetic architecture would support *E. coli* infection. The resulting phage genomes retained the functional organization required for host recognition, DNA injection, replication and assembly of new viral particles, while the DNA sequences themselves differed markedly from those of natural bacteriophages.

from screen to lab

The team screened the model-generated genomes and selected 300 sequences that appeared most likely to be functional. They synthesized these genomes molecule by molecule in the laboratory and introduced them into *E. coli* to test whether active phages would be produced. Of the 300 synthesized genomes, only 16 gave rise to fully active bacteriophages with previously unpublished sequences, distinct genes, novel regulatory elements and even altered genome sizes.

The study, published this week in *Science*, also evaluated whether the model-designed phages could overcome bacterial resistance. In an experiment, *E. coli* strains that had already developed resistance to the original Phi X-174 were exposed to a mixture of model-designed phages and a comparable mixture of natural phages similar to Phi X-174. The results showed that the AI-generated phages rapidly overcame bacterial resistance and established infection. Their behavior varied across samples: some infected bacteria more quickly, while others displayed different replication characteristics.