Generative design of novel bacteriophages with genome language models

Samuel King (Arc Institute) and colleagues published a new preprint on generative design of phage genomes using language models.

Their AI model, built from Evo1 and Evo2 (millions of genomes across life), fine tuned on 15,000 Microviridae genomes, and prompted with phiX174's genome to get started, successfully generated 16 viable phage genomes (which the team rebooted in E. coli). Compared to wild type phiX174, the phages had substantial evolutionary novelty, enhanced fitness, and the improved ability to overcome bacterial resistance when used as a cocktail.