Scientists have engineered 16 entirely new, functional viruses using artificial intelligence, a development that could offer new avenues for combating antibiotic-resistant bacteria while simultaneously presenting serious biosecurity questions. The research, published in the journal Science, marks the first instance of artificial intelligence being used to design complete viral genomes from scratch.
A team from Stanford University and the Arc Institute utilized AI models, specifically Evo 1 and Evo 2, trained on a vast dataset of over two million bacteriophage genomes. These models, akin to language models that generate text, were programmed to recognize patterns in DNA structure and subsequently generate novel genetic sequences. The researchers focused on designing bacteriophages, viruses that specifically infect bacteria and do not harm humans or animals. The AI generated thousands of potential viral genomes, from which approximately 300 were selected for laboratory synthesis and testing. Of these, 16 proved to be viable and functional viruses.
These AI-designed viruses demonstrated an ability to infect and kill E. coli bacteria. Notably, a mixture of these synthetic phages was effective against E. coli strains that had evolved resistance to naturally occurring bacteriophages. This capability is seen as a significant step toward developing more resilient and adaptive phage therapies, a promising alternative to traditional antibiotics that are losing efficacy against evolving pathogens. The researchers suggest that such AI-driven design could accelerate the development of treatments for infections caused by drug-resistant bacteria like tuberculosis and MRSA.
However, the creation of novel viruses by AI has also drawn attention to potential biosecurity risks. Experts have warned that the same technology used for therapeutic purposes could, if misused, be employed to design dangerous pathogens. While the researchers intentionally excluded human pathogen data from the AI's training set to mitigate the risk of creating viruses harmful to humans, the broader implications of this technology are a cause for concern.
In an accompanying commentary in Science, biosecurity experts Thomas Inglesby and Moritz Hanke of the Johns Hopkins Center for Health Security highlighted that the ability to compose viral genomes using generative AI now exists, but governance frameworks to manage this capability are lagging. They emphasized that the technology could potentially be used to encode new pathogens that existing countermeasures cannot contain. While the current study focused on relatively simple bacteriophage genomes, the underlying technology could theoretically be applied to more complex and dangerous viral agents.
The researchers involved in the study acknowledged these concerns, urging for careful consideration of biosafety, biocontainment, and biosecurity throughout the research process. They also pointed out that AI-designed viruses could potentially be engineered with built-in safety checks, a level of control not present in naturally evolving pathogens. The AI model, named Evo 2, has been made openly available, allowing other researchers to build upon the work.
The development represents a significant advancement in generative genomics, moving beyond designing individual genes or small genetic systems to engineering entire biological systems at the genome scale. While the immediate applications focus on combating bacterial infections, the successful creation of functional, novel viruses by AI opens a new chapter in synthetic biology, necessitating ongoing dialogue between scientific advancement and regulatory oversight.