Phage therapy has long been considered a promising approach to combating bacterial infections, especially in the context of the global spread of antibiotic resistance. However, bacteriophages have an important characteristic: they are typically extremely specific to their bacterial hosts. A phage effective against one bacterial strain may be completely ineffective against another. Therefore, finding the right phage for a specific infection can be complex and time-consuming.

A new study by American scientists demonstrates a fundamentally different approach: not only finding the desired bacteriophage in nature, but also creating its genome using artificial intelligence.

How Artificial Intelligence Learned to "Write" Phages

Researchers from Stanford University and the Arc Institute used the Evo 1 and Evo 2 genomic language models. Similar in principle to how a language model works with text, these systems analyze DNA sequences and learn to recognize patterns underlying genome function.

For the experiment, the scientists chose the well-studied lytic bacteriophage ΦX174, which infects Escherichia coli. Its genome contains less than 6,000 base pairs, making it a relatively simple target for an initial demonstration of the technology's capabilities.

Evo 2 received a small amount of initial information about the ΦX174 genome and began generating new genomic sequences. Importantly, this wasn't just editing one or a few genes of the natural phage. The model generated entire genomes, proposing new combinations of genetic information.

Almost 300 candidates were tested in the laboratory

Computer-generated genomes are just the first step. To test whether the generated sequences are capable of functioning in a real biological environment, the researchers synthesized and tested nearly 300 candidates.

As a result, 16 artificially engineered bacteriophages were viable. They were able to infect and lyse E. coli.

This is a fundamentally important point: the artificial intelligence didn't simply produce attractive genetic sequences. Some of the genomes it created were found to be functional after synthesis and lab testing. The researchers also discovered significant evolutionary novelty in the generated phages.

Some of the new phages showed even higher fitness efficiency than natural ΦX174 in competition and bacterial lysis rate tests.

Separately, the scientists examined the structure of one of the created phages using cryo-electron microscopy. It turned out that it uses an evolutionarily distant variant of a protein involved in packaging DNA into the capsid. This demonstrates that the model is capable of generating not just random mutations, but biologically functional and simultaneously unusual solutions.

Why is this important for phage therapy?

One of the main challenges of phage therapy is phage resistance. Bacteria, as with antibiotics, can evolve and become insensitive to a specific phage.

Therefore, a promising approach is to use phage cocktails – combinations of several phages that attack the bacteria in different ways.

In a new study, a cocktail of engineered phages was able to quickly overcome resistance to natural ΦX174 in three E. coli strains. This is particularly exciting: instead of waiting for a new phage to be discovered in nature, it could potentially be possible to engineer a set of genetically diverse phages to target a specific bacterial challenge.

This approach could potentially be useful for creating phages against more complex pathogens, particularly bacteria, for which antibiotic resistance is a particularly pressing issue. Researchers cite MRSA and Pseudomonas aeruginosa as potential targets. However, this is still a promising approach, not one with proven clinical efficacy.

There's still a long way to go from laboratory phage to medicine.

Despite the headlines about "AI creating viruses," it's important to understand the scale of the result.

The created phages were tested against E. coli in laboratory conditions. This is not yet a drug for human use and is not a clinical trial.

For potential use in phage therapy, it will be necessary to prove the safety, stability, specificity, efficacy, and predictability of such phages in the human body. It is also necessary to determine how the immune system will respond to them and how quickly bacteria can develop resistance.

However, it is precisely the principle demonstrated by this work that may become very important.

Previously, the basic logic was this: find a phage → test it against a bacterium → modify or form a cocktail if necessary.

Now another opportunity appears:

define desired properties → ask the model to design the genome → synthesize candidates → test the most promising variants.

This could significantly change the speed of searching for new bacteriophages.

A New Era of Personalized Phage Therapy?

The most interesting question is whether this technology could one day make phage therapy more personalized.

Let's imagine a patient with an infection caused by a multidrug-resistant bacterium. Instead of spending months searching for a natural phage, it might be possible to use genomic models to create candidates adapted to a specific bacterial strain and then test them in the lab.

This scenario is still a long way off. But new work demonstrates that designing functional bacteriophages at the genome level using AI is no longer just a theoretical idea.

At the same time, the technology also raises the complex issue of biosafety. If artificial intelligence can create functional viral genomes, it's necessary to simultaneously develop control and safety systems to ensure that such tools are used to treat and protect against infections, rather than to create dangerous biological agents.

This study doesn't mean that tomorrow doctors will receive a "patient-specific, AI-designed phage." But it does demonstrate a crucial technological milestone: for the first time, artificial intelligence has helped engineer entire bacteriophage genomes, some of which have been shown to be functional in the lab.

For phage therapy, this may mark a transition from the era of natural phage discovery to the era of rational bacteriophage design.

And perhaps it is here that artificial intelligence will become not a competitor to phage therapy, but one of the most powerful tools for its development.