Using AI, researchers have designed complete, functional bacteriophage genomes from scratch and tested them against bacteria that had evolved resistance to a natural bacteriophage. Their work marks a step toward generative systems capable of engineering entire biological systems rather than individual genes or small genetic systems, while also raising important biosafety and biosecurity concerns. “[The authors] engage with biosafety and biosecurity questions more deliberately than most developers of powerful biological AI models,” write Thomas Inglesby and Moritz Hanke in a related Perspective. Advances in DNA sequencing and synthesis have made it increasingly possible to read and write entire genomes, but designing a functional genome from scratch remains extraordinarily difficult because genes, regulatory sequences, and other elements interact in highly complex ways. Here, Samuel King and colleagues introduce an approach to designing functional genomes that combines the Evo genomic language models they’d previously built (including Evo 1), computational biology, and experimental screening. They used this approach to generate complete bacteriophage genomes; bacteriophages have substantial utility as biotechnologies and have therapeutic relevance as treatments for bacterial infections. Using the well-studied ΦX174 bacteriophage as a model, King et al. computationally designed hundreds of candidate genomes and experimentally identified 16 functional phages, whose genetic sequences and structures differed substantially from one another. Some of the engineered phages performed comparably to naturally occurring relatives, they report, and notably, some combinations of the new phages overcame resistance in two strains of E. coli that resisted ΦX174-like phages. The findings demonstrate that AI-guided generative genomics could eventually enable the design of more durable phage-based therapies.
King et al. emphasize that the ability to design and synthesize functional AI-generated genomes introduces important biosafety and biosecurity concerns and underscores the need for expert oversight and robust safeguards throughout the design process. “Groups conducting future whole-genome design work should consult both safety and security professionals throughout the project lifecycle,” they say. They argue that existing safety frameworks can be adapted to generative genomics, while model-level protections, such as excluding sensitive viral sequences from training data, may provide an additional layer of risk mitigation. “The question is no longer whether generative viral genome design will exist,” say Inglesby and Hanke in the Perspective. “It is whether society can build oversight that allows its benefits to unfold while preventing it from enabling serious harm.”
