Generative AI Enters Life Sciences, Creating Novel Bacteriophages for the First Time | WPN World People News

Researchers from Stanford University and California-based research institute Arc Institute have published a new study in the journal Science, reporting that they used generative artificial intelligence to design novel bacteriophages that have never existed in nature and successfully killed Escherichia coli (E. coli) in laboratory experiments. The study is regarded as a major breakthrough in the integration of artificial intelligence and synthetic biology, but it has also raised concerns over biological safety.

The research team developed a model called Evo 2, which operates in a manner similar to large language models. The key difference is that instead of learning from text, the model is trained on genetic sequences such as DNA. Researchers first trained the model using large amounts of genetic data from viruses, bacteria, plants and humans, then fine-tuned it to enable the AI to design bacteriophage genomes.

From a large number of candidate sequences generated by the AI, the team selected 302 for synthesis and testing. Ultimately, 16 were able to successfully infect and eliminate E. coli, with some performing even better than the natural bacteriophage ΦX174 (Phi X-174) that served as the basis for the designs. Because these types of bacteriophages infect only bacteria, the researchers said they do not pose a risk of infecting humans.

Brian Hie, who led the study, said that AI has previously been used mainly to develop new drugs or antibiotics. This study, however, represents the first time a fully functional viral genome has been generated from scratch, demonstrating that generative AI is now capable of designing entirely new biological systems.

As antibiotic resistance continues to worsen, phage therapy has once again attracted growing attention in recent years. Because bacteriophages can precisely target specific bacteria, they are considered a potential alternative for combating multidrug-resistant pathogens and could eventually have applications in treating infections and other areas of biomedical research.

However, the findings have also raised concerns among biosafety experts. Scholars at the Johns Hopkins Center for Health Security have warned that global governance mechanisms for AI-designed genomes remain inadequate. If the technology is misused, it could increase the risk of developing dangerous pathogens or even biological weapons.

The research team said it excluded viruses capable of infecting humans, animals or plants from the training data, and that all experiments were conducted in laboratories complying with established safety regulations. Nevertheless, because Evo 2 has been released as an open-source model, experts argue that alongside continued efforts to promote medical innovation, greater attention should also be paid to DNA synthesis screening, research ethics and biosafety oversight to ensure that technological progress is balanced with public safety.