Stanford Researchers Use Generative AI to Create Synthetic Viruses

Researchers at Stanford University have used a generative artificial intelligence program called Evo 2 to design and produce 16 synthetic phages, which are small viruses that infect bacteria. The study, published in the journal Science, demonstrates for the first time that generative design can create complete functional genomes. In laboratory tests, these synthetic phages were more effective at eliminating the common microorganism E. coli than the natural phage phiX174 used as a model for the AI.

The Evo 2 model operates similarly to large language models by predicting sequences, but it applies this approach to DNA. While the created viruses are closely related to existing ones, they possess distinct features that would be difficult to evolve naturally. The researchers noted that this approach could eventually be extended to larger organisms, such as bacteria and more complex life forms.

Potential applications include the advancement of phage therapy, which uses specific phages to treat bacterial infections. This is seen as a possible alternative to antibiotics, especially as antibiotic resistant pathogens become more concerning. To ensure safety during the initial study, the team used a non pathogenic strain of E. coli and a model not trained on viruses that infect plants or animals.

The project is open source and available for free download. Brian Hie, the project leader, stated that the potential benefits to humanity outweigh the risks of bioterrorism. However, other experts, including researchers from Johns Hopkins University, warned that current governance structures are insufficient to oversee generative genomics and that the study raises urgent biosafety and biosecurity questions.