Researchers have employed an artificial intelligence system to engineer new viral genomes that differ significantly from any currently identified natural viruses, while specifically targeting designated hosts, as detailed in a recent study. This advancement holds promise for medical innovation but also raises critical concerns about potential misuse of the technology.
Published on Thursday in the journal Science, the study explains that the AI tool, named Evo, utilized genetic sequences from a vast array of sources across "all domains of life" as its foundational language library, which is akin to how models like ChatGPT are trained on extensive text datasets.
The collaboration between scientists from Stanford University and the Arc Institute led to the generation of thousands of genome combinations tailored to infect E. coli bacteria. From these generated genomes, approximately 300 were synthesized and tested in the lab, revealing that 16 could function as viable viruses.
These newly created viruses are classified as bacteriophages, which specifically target bacteria without posing a threat to human health.
The research indicates that the AI's genetic library enabled it to grasp "evolutionary constraints" inherent in natural genomes. However, the scientists identified one virus exhibiting traits that were "evolutionarily distant," implying that the AI had induced changes that might have taken natural evolution millions of years to reach.
Experimental results indicated that a blend of these novel viruses was capable of overcoming antibacterial resistance in various E. coli strains, an achievement that a similarly composed mix of naturally occurring phages failed to accomplish.
The authors highlighted the significance of this research in paving the way for developing adaptive and robust phage therapies against swiftly evolving pathogens, especially pertinent as scientists work to address the growing issue of drug resistance in bacteria like E. coli.
Jordi García Ojalvo, a professor of systems biology at Pompeu Fabra University in Barcelona, characterized this development as a substantial breakthrough in the field.
Nevertheless, the research is not without caution. Experts from the Johns Hopkins Center for Health Security noted in a related article that, while the findings are promising for advancements in life sciences, they also bring urgent biosafety and biosecurity considerations to the forefront.
With the capability to create viral genomes through generative AI now available, the authors pointed out the absence of governance to ensure its safe application. The authors of the study tackled biosecurity issues head-on but acknowledged that their approach may be more thorough compared to other developers of advanced biological AI systems.
While focused on E. coli and involving viruses that do not affect humans, the broader implications for other types of viruses remain uncertain. The Johns Hopkins experts specifically warned against pursuing research on eukaryote-infecting pathogens, which have the potential to cause infections in humans and animals, such as malaria and various yeast infections.
They cautioned that such developments might yield new pathogens capable of infecting humans, animals, or plants in ways that current preventative measures cannot manage.
García Ojalvo assessed the biosafety risks associated with this research as relatively lower compared to other AI applications, given that each genome must undergo individual testing post-design, resulting in a low rate of viable creations—only 16 successful viruses derived from hundreds of thousands generated.
"It’s hard to envision these models spontaneously producing viable genomes without thorough testing,” he remarked.
