Scientists have used artificial intelligence to design complete working genomes of bacteriophages, viruses that infect bacteria, for the first time, marking a major step toward using AI to engineer complex biological systems, while also raising fresh questions about the safe use of AI in biotechnology.
According to a report in the New York Times, the study titled “Generative design of bacteriophages with genome language models” revealed that AI models trained on millions of natural genomes were able to generate entirely new bacteriophage genomes which functioned in laboratory experiments. Researchers used Evo 1 and Evo 2 AI models to generate 700,000 candidate genomes out of which 302 were selected for synthesis and then 285 were successfully built and 16 were made viable to produce viruses capable of infecting Escherichia coli C bacteria
Unlike previous AI based biological design, which has largely focused on individual genes and proteins, the new approach enabled the design of entire viral genomes. The researchers used two genome language models, Evo 1 and Evo 2, developed by computational biologist Brian Hie of Stanford University, which were trained on DNA sequences from millions of organisms to learn the evolutionary patterns which shape genomes.
“Genome language models are artificial intelligence (AI) algorithms that have shown promise in designing biological systems. Much like how other language models are trained on large corpora of text, genome language models are trained on large corpora of DNA comprising millions of genomes from all domains of life,” the study mentioned.
The researchers mentioned that by using the natural phage ΦX174 as a design template, they established a framework for generating and evaluating thousands of AI-generated genomes, nearly 300 of which were chemically synthesized and tested in laboratory conditions, yielding 16 viable phages. The viable generated phages showed strong host specificity and diverse fitness profiles, including competitive infection kinetics.
The researchers from Stanford University and Arc Institute also tested whether the AI designed viruses could help tackle bacterial resistance, a major challenge for phage-based therapies. A cocktail of the AI-generated bacteriophages successfully killed E.coli strains that had evolved resistance to the natural ΦX174 virus. ΦX174 and the E. coli C host are non-pathogenic, with a long history of safe use in molecular biology.
By comparison, a similar mixture of naturally occurring ΦX174 like phages failed to overcome the resistant bacteria.
Viruses that infect eukaryotes — including human, animal and plant viruses — were excluded from Evo 1 and Evo 2’s training data as a deliberate biosafety measure when the models were built.”
The authors argue in their paper that whole-genome design projects should include safety specialists from model-building through to lab testing.
According to the researchers, the findings demonstrate that AI can capture the evolutionary rules of DNA well enough to design complete, functional genomes with specific biological traits. “Our approach expands what synthetic genomics can achieve alongside methods such as directed evolution and rational engineering, lays out a path for generating adaptive and resilient phage therapies against rapidly evolving pathogens, and establishes a foundation for the generative design of larger, more complex genomes,” the researchers highlighted.
The study was conducted by Samuel H. King Claudia L. Driscoll, David B. Li Daniel Guo Aditi T. Merchant Garyk Brixi Max E. Wilkinson, and Brian L. Hie.
Meanwhile in an accompanying article titled “AI-designed viral genomes”, Professor Thomas V. Inglesby and Moritz S. Hanke at the Centre for Health Security, Department of Environmental Health and Engineering, Johns Hopkins University School of Public Health raised concerns saying, “Although this is promising for life sciences applications, it also raises urgent biosafety and biosecurity questions. The ability to compose viral genomes using generative AI now exists; the governance to safely steer it does not.”
Also Read: DeepMind and Isomorphic Labs Pitch “Bioresilience” Plan to Keep AI Out of Bioweapon Hands
