Researchers at Stanford University and the Arc Institute in California have used artificial intelligence to design complete bacteriophage genomes from scratch, producing functional viruses capable of infecting bacteria that had evolved resistance to a naturally occurring bacteriophage.
The work demonstrates how generative AI systems can engineer entire biological systems rather than just individual genes or smaller genetic components. However, the research also highlights significant biosafety and biosecurity concerns surrounding the ability to design and synthesise complete genomes.
Despite advances in DNA sequencing and synthesis making it more feasible to read and write entire genomes, designing a functional genome from scratch remains exceptionally challenging because genes, regulatory sequences and other genetic elements interact in complex ways.
AI-guided genome design
Samuel King, a PhD student in Bioengineering at Stanford University and a member of Brian Hie’s lab at the Arc Institute in Palo Alto developed, with several colleagues, an approach that combines the Evo family of genomic language models, including Evo 1, with computational biology and experimental screening.
The researchers applied the approach to bacteriophages, viruses that infect bacteria and have potential applications as biotechnology tools and treatments for bacterial infections.
Using the well-studied ΦX174 bacteriophage as a model, the team computationally designed hundreds of candidate genomes. These were then tested experimentally to identify genomes capable of producing functional phages.
The researchers identified 16 functional phages. Their genetic sequences and structures differed substantially from one another, showing that functional bacteriophage genomes could be generated through multiple distinct designs.
Some of the engineered phages performed comparably to naturally occurring relatives, with the researchers seeing that combinations of the newly designed phages could overcome resistance in two strains of E. coli that had resisted ΦX174-like phages.
The findings show that AI-guided generative genomics could eventually contribute to the development of more durable phage-based therapies, particularly where bacteria have developed resistance to existing phages.
Biosafety concerns
The ability to design and synthesise functional genomes using AI also creates new challenges for biological safety and security.
King and colleagues emphasise the importance of expert oversight and robust safeguards throughout the genome design process. They argue that existing safety frameworks could be adapted to address generative genomics while additional protections could be built into AI models.
One potential safeguard suggested could involve excluding sensitive viral sequences from training data, providing an additional layer of risk mitigation.
What comes next?
Despite the results looking encouraging for the development of more durable phage-based therapies, particularly as bacteria continue to develop resistance to existing treatments, further research will be needed to determine how reliably AI-designed phages can be developed for therapeutic use and how their potential risks can be managed.
As whole-genome design becomes more accessible, the researchers argue that safety and security measures will need to advance alongside the technology to ensure its medical potential can be realised responsibly.