Scientists Create 16 New AI-Designed Viruses That Can Reproduce

For the first time, scientists have created 16 new, previously nonexistent types of viruses based on genetic information designed by artificial intelligence (AI) and confirmed that they can actually reproduce.

This achievement has been hailed as a major milestone in AI-driven biotechnology.

At the same time, however, concerns have been raised that leveraging AI to more easily create unprecedented viruses could lead to the technology being weaponized for biological warfare.

Scientists affiliated with Stanford University and the Arc Institute in California, USA, published a paper titled "Generative design of bacteriophages with genome language models" in the journal Science, published by the American Association for the Advancement of Science (AAAS), on August 6 (local time).

The corresponding author of the paper is Brian Hie, an assistant professor of chemical engineering at Stanford University who also serves as an Innovation Investigator at the Arc Institute.

The research team selected a bacteriophage called ΦX174 as the baseline framework for the genome to be designed by the AI model.

A bacteriophage refers to a virus that infects and uses bacteria as hosts to reproduce.

They are often commonly referred to as phages.

The research team chose ΦX174 as their baseline because it is widely known to exclusively infect and destroy Escherichia coli (E. coli) while remaining harmless to humans, making it exceptionally low in risk.

The genome language models "Evo 1 and 2" were used as the AI models.

After training on millions of natural genomes as well as additional genomic data of ΦX174 and closely related phages, these models generated candidates that had the potential to become reproducible new phage strains.

According to the U.S. daily The New York Times (NYT), the AI generated about 700,000 candidate genomes.

The research team chemically synthesized and tested approximately 300 of these candidates in the laboratory that they judged to have the highest probability of success.

As a result, they confirmed that 16 of these phages were actually capable of reproducing and that their genetic information differed from existing phages known to exist in nature.

In particular, a mixture combining some of these phages demonstrated the ability to rapidly infect and destroy E. coli strains that were resistant to natural ΦX174.

Comparative mixtures of natural ΦX174-like phages lacked this capability.

According to the NYT, the genomic information processed by Evo, spanning animals, plants, microorganisms, and viruses, amounted to approximately 9 trillion base pairs (bp).

According to the BBC, the genetic information of the newly designed phages was about 5,400 base pairs.

Phages and other viruses are not considered "living" organisms in the traditional sense.

The living cell with the smallest amount of genetic information has about 500,000 bp, while the human genome consists of about 3 billion bp.

Regarding the possibility of creating fully-fledged living organisms through AI-driven genome generation, corresponding author Hie told the BBC, "It would probably take a lot of effort, but it's not impossible," adding that the research team is "definitely interested in pushing in that direction."

Professor Patrick Cai, director of the Manchester Institute of Biotechnology at the University of Manchester in the UK, evaluated the study as an "important milestone," noting that "genome language models are beginning to learn the design principles embedded in evolution."

He pointed out that a pathway for designing genomes using AI has now been opened.

In a commentary titled "AI-designed viral genomes" published simultaneously with the Science paper, Dr. Thomas Inglesby and Dr. Moritz Hankel of the Johns Hopkins Center for Health Security assessed that the study "raises urgent biosafety and biosecurity issues."

In their commentary, they pointed out the risks, stating, "The capability to create viral genomes using generative AI now exists. Governance to safely guide this does not."

Observers warn that if such technology continues to develop without restrictions and is misused, it could potentially lead to the creation of biological weapons equipped with lethal toxicity or mass destruction capabilities.

Dr. Hankel told the NYT that in the future, someone might be able to prompt, "Hey, genome language model, design an influenza genome engineered to be more transmissible or more lethal."

(Photo: Science paper capture, Stanford University public photo, Yonhap News)
※ Please note: This article was translated by AI and may contain errors.