The computer did not build a virus or handle any cells. Instead it suggested the genetic makeup for one. Scientists at Stanford University and the Arc Institute applied artificial intelligence to create full genomes for bacteriophages that target bacteria. Out of 285 AI-generated sequences that were built and tested, 16 produced working phages. Some of these overcame bacterial defenses that blocked the original virus.
People have synthesized and altered viral genomes for years. The new element is AI taking part in the design process. This development brings medical possibilities while also prompting biosecurity questions.
Bacteriophages have been studied for over a century. In 1977 a small phage became the first virus with a fully sequenced DNA genome. By the early 2000s researchers could synthesize viral DNA from sequence data and produce active viruses. Later work showed that viruses could be intentionally changed, as seen in controversial flu experiments from 2011-2012 that altered transmission traits.
In the recent study the team already knew the genome of phage ΦX174 and how to synthesize it. They used two genome language models called Evo 1 and Evo 2. These systems learn patterns in DNA letters across many genomes and then propose new sequences. The models were trained on related bacteriophage genomes.
The AI did not create an unrelated virus. It produced new versions of ΦX174-like genomes. Researchers chose some sequences, made the DNA, and placed it into E. coli cells. Sixteen designs worked and created new phage particles.
One AI-designed phage combined changes across multiple genes in ways that standard methods had not achieved. This suggests AI could help find sets of genetic alterations that function together.
Phage therapy offers one use. Antibiotic resistance limits standard drugs, and phages can kill bacteria. Their drawback is narrow effectiveness and the risk of resistance. Generative methods may allow design of needed phages rather than only searching for them. The experiment showed AI versions could defeat resistant E. coli strains.
Similar approaches could aid vaccine components, antibodies, and cancer-targeting viruses. The core advance is AI helping shape biological functions.
The risk involves faster development of such capabilities, even though the tested phage is simple.
