AI Models Trained on Genomes Used to Design New Viruses
Researchers at Stanford University have used large genome models to design new viruses that infect bacteria. The AI-generated viruses are closely related to existing ones but have features that are hard to evolve naturally. The researchers suggest preparing for potential misuse in designing viruses that target vertebrates.
A lot of AI work in biology has focused on designing proteins, but researchers have now used large genome models, which are trained on DNA sequences, to generate genomes of viruses that infect bacteria. These AI-created viruses are closely related to an existing virus but have distinct features that would be challenging to evolve. The work was done at Stanford University, and the researchers suggest we may want to start thinking about preparing for the potential that someone could develop a related AI that can design a virus targeting vertebrates. Large genome models, similar to large language models, are trained on DNA sequences, which use four letters: A, T, C, and G. They recognize biological context, and when prompted with parts of a gene cluster, they can output DNA sequences encoding proteins with related functions, potentially including novel functional proteins.
Source: Ars Technica —
original