Machine, make me a virus: The dream and nightmare of new biology

Researchers from Stanford University have designed and created from scratch viruses that attack resistant bacteria, proving that artificial intelligence tools are already capable of writing functional genetic code. The breakthrough presents immense medical promise, but also comes with fears of a biological arms race.

GlobesAuthor: Gali Weinreb
Source
Machine, make me a virus: The dream and nightmare of new biology
Photo: Globes / הווירוס החדש שיצר כלי בינה מלאכותית / צילום: Samuel H. King et. al

Researchers from Stanford have designed and created from scratch, with the help of artificial intelligence, viruses that attack a resistant bacterium in the laboratory, stirring the scientific world. The potential of these new microorganisms in medicine is immense, but the questions about the future are even greater. Will we soon be able to ask a learning program to design living creatures as we wish, with the traits we desire?

The answer is still no, but it is the most hesitant "no" we have heard to date from scientists on this subject. In principle, they say, the technology exists. "In 50 years, if you want to dive in the sea, I will be able to grow you gills," predicts Dr. Lior Nissim, head of the Synthetic Biology Biomedical Applications Laboratory at the Hebrew University's Faculty of Medicine. "If you had asked me five years ago, I would have said it would happen in 100 years at the very least. Right now I'm betting on 50, maybe 40, and maybe even faster. And maybe a few decades later, your child will be able to draw an invented character, and you will be able to press a button and produce such an animal for him, a real one."

How an AI tool created a virus that does not exist in nature

  1. Learning the genetic database: The AI model learns DNA sequences of existing phages (viruses that attack bacteria).

  2. Defining the task: The researchers set the model to design a phage that will attack an E. coli bacterium.

  3. Creating variations: The model suggests new genetic sequences that do not exist in nature.

  4. Filtering: The researchers choose the most likely candidates and build them.

  5. Activity test: 16 selected phages did indeed infect and kill the defined bacterium.

The software predicted, and was right

The study that sparked scientific and media interest was published earlier this month in the leading journal Science. Researchers from Stanford University and the Arc Institute in Palo Alto, which specializes in artificial intelligence for biology, reported that they designed and then built from scratch short viruses called phages. They were designed to attack an E. coli bacterium, and succeeded in doing so.

"The researchers took a common phage against which the bacterium knows how to develop resistance, and asked dedicated language models for genetic design to create variations of these phages that are unknown to the world, but according to the software will be active and overcome bacterial resistance. The designed virus was a combination of the basic phage and all kinds of proteins known from other phages. The system predicted that they would give the phage certain properties, and was right."

The models used to design the phages are Evo and Evo2, built by the ARC Institute. Dr. Yishai Pinto, a senior lecturer and head of the Computational Virology Research Laboratory at Bar-Ilan University, explains how they work: "These are generative models for genomic design. Just as we can tell a chat bot to write a paragraph for a certain purpose, this machine can write genomes according to goals that we define. This system was told, 'You have seen many phages that kill E. coli. Now design me many more like that.' And then they were filtered and left with those that looked more active. Of them, 16 phages did indeed infect the bacterium and kill it."


Related News