AI designed a virus, scientists brought it to life

Researchers from Stanford University and the Arc Institute have utilized artificial intelligence to design the genetic code of bacteriophages, viruses that infect bacteria, and successfully brought some of these designs to life in a laboratory setting. In a study published last week in Science, AI models generated hundreds of new phage designs, 16 of which produced functioning viruses capable of reproducing and attacking E. coli bacteria.

The researchers also discovered that combinations of these newly created phages could attack bacteria that had developed resistance to the original phage, potentially leading to new treatments for antibiotic-resistant bacterial infections. However, this breakthrough also highlights the rapid advancement of AI in designing biological systems, raising concerns about safety, oversight, and the implications of increasingly capable tools.

To grasp how AI can design a virus, one must think of DNA as a language. Researchers developed models trained on vast quantities of genetic information, enabling them to learn patterns and rules governing genome construction and function. These models, named Evo 1 and Evo 2, generate genetic sequences, similar to how conventional language models produce text.

When computers start writing biology, the process is not entirely new. Researchers have previously created simple life forms using DNA designed and synthesized in the laboratory. What differs now is the power of computational tools. The latest study employed technologies akin to those behind language models to learn from extensive genetic data and generate new sequences.

These models, like language models, learned to build bacteriophage viruses capable of attacking bacteria. While AI's potential contributions to understanding biological functions and designing bacteriophages for treating antibiotic-resistant infections are significant, the field is still in its infancy. Questions remain about whether engineered viruses or naturally occurring phages should be the future of phage therapy.

Despite the technology's limitations, researchers will need more data and improved models before their outputs can be consistently relied upon. Like other AI systems, identifying when the machine "lies" or generates dangerous outputs remains a challenge. However, the bacteriophages involved in the current study do not pose a direct threat to humans, as they target bacteria rather than human cells.