The scientific community has received sensational confirmation: generative artificial intelligence models have, for the first time, managed to design a complete genome of a virus that attacks bacteria — a bacteriophage. This is not just a laboratory curiosity, but a fundamental shift in our ability to construct biological systems.

A group of researchers led by Brian Hie and first author Samuel King presented results that were previously considered unattainable. Until now, generative AI in biology has only worked with individual genes or proteins. Designing an entire genome is a task of colossal complexity: even minimal mutations can destroy an organism's viability. However, the team managed to obtain 16 fully functional viruses, none of which replicate known natural counterparts.

From thousands of variants — to life

As a basis, the researchers took the natural phage ΦX174, which infects the bacterium Escherichia coli C. The Evo 1 and Evo 2 models, trained on millions of genomes from all domains of life, absorbed evolutionary constraints, much like language models learn grammar. The task was not easy: generating thousands of genome variants, of which nearly 300 were chemically synthesized and tested in the laboratory. The result — 16 viable samples.

Comparison with the natural phage revealed striking differences. The AI phages contain new mutations, divergent genes, and regulatory elements, as well as differences in genome length. Moreover, cryo-electron microscopy showed that one of the viruses incorporated into its capsid a protein borrowed from an evolutionarily distant phage. This proves: the model did not simply sift through mutations, but created fundamentally new designs.

A breakthrough in phage therapy

The key result is not the fact of generation itself, but the AI's ability to overcome bacterial resistance. The scientists tested the created viruses on strains of E. coli that had developed protection against ΦX174. A mixture of the designed phages quickly broke through this defense, while a cocktail of natural counterparts failed. This opens a direct path to creating "custom-made" phage preparations — to combat specific, rapidly mutating pathogens, which is especially relevant in the era of the antibiotic crisis.

My view: This work marks a transition from "reading" the genome to actively "writing" it. Investments in generative AI for biotech are no longer futurism, but pragmatic calculation. In the coming years, we will witness the commercialization of such technologies, and the phage therapy market, valued in the billions of dollars, will receive a powerful growth catalyst.