In a sweeping stress test of biology’s predictive tools, researchers systematically altered nearly every base in the genome of the classic bacteriophage ΦX174 and tracked how the virus fared in E. coli. Roughly half of single-nucleotide changes and 60% of amino-acid substitutions proved harmful—far more than expected—while a few tweaks boosted viral fitness despite decades of lab adaptation. State-of-the-art AI systems that aim to forecast the impact of mutations struggled to anticipate these outcomes, highlighting gaps in current models and the need for richer experimental datasets. The preprint suggests that even in a model organism with a 5,386-base genome and 11 proteins, biology retains surprises—tempering expectations for AI-driven design in genomics and synthetic biology.
Related articles:
– World’s first AI-designed viruses a step towards AI-generated life
– Why phage viruses could be the key to treating deadly infections — if they can be harnessed safely
– AI can write genomes — how long until it creates synthetic life?
– These ‘master’ proteins protect us from deadly mutations — and could inspire new drugs
– These powerful tools reveal the ‘control knobs’ of the genome
