In an experiment with resistant E. coli, a mixture of AI-designed phages overcame resistance, while a mixture of closely related natural phages did not

In an experiment with resistant E. coli, a mixture of AI-designed phages overcame resistance, while a mixture of closely related natural phages did not

On August 6, Science published a paper by a team from Arc Institute and Stanford on bacteriophage genome design. The authors compared a mixture of phages with AI-generated genomes against closely related natural ΦX174-like phages and a single ΦX174 phage. Only the generated mixture overcame resistance in E. coli.

A bacteriophage is a virus that infects bacteria and replicates inside them. Designing a complete virus requires more than selecting individual proteins: the entire sequence must allow the virus to produce viable progeny. In ΦX174, this information fits into 5 386 DNA bases and 11 genes. The genes in this compact genome overlap, so a single substitution can alter two proteins at once.

Samuel King’s team fine-tuned the Evo 1 and Evo 2 genomic models on related phages, provided ΦX174 as a template, and generated complete genome variants. The researchers then filtered the sequences, synthesized and assembled 285 candidates, and tested them in bacterial cultures. Sixteen phages replicated and lysed the cells, which means that they ruptured the cells and released their progeny.

In the earlier experiment, the generated cocktail was compared only with a single ΦX174 phage. It therefore remained uncertain whether its advantage came from the origin of the genomes or simply from combining several phages in one mixture.

The new control addresses this question. The authors tested three conditions: a mixture of generated phages, a mixture of closely related natural phages, and a single ΦX174 phage. In his description of the experiment, King reports that the generated mixture overcame resistance. The natural mixture and the single ΦX174 phage did not produce this result.

The added control separates the effect of using a mixture from the effect associated with the genomes’ origin. In this laboratory system, the two closely related mixtures produced different results.