Large genome models used to design new viruses

Large genome models, which are AI systems trained on DNA sequences similar to how language models are trained on text, have been used to generate new viral genomes for viruses that infect bacteria. Researchers tested whether these models could output complete virus genomes by prompting them with the ending sequence of an existing bacteriophage and filtering the results through biological constraints. The work demonstrates that genome models can create viable viruses with distinct features that would be difficult to evolve naturally, though the viruses generated remain closely related to existing ones. Researchers deliberately excluded viral sequences that target complex cells from their training data as a precaution, but they note that the potential exists for similar AI systems to eventually design viruses that could target vertebrates, warranting advance consideration of biosecurity measures.