Stanford researchers have successfully used AI to design complete viral genomes, cultivating them in the lab, marking a significant advancement in biotechnology.
Researchers from Stanford University have reached a significant milestone by employing artificial intelligence to create complete viral genomes. This innovative achievement involved cultivating AI-generated viruses in a laboratory, as reported in a study published in the journal Science.
The team utilized generative models known as Evo1 and Evo2, which were trained on the genetic information of various organisms, including viruses, bacteria, and humans. These models function similarly to large language models like ChatGPT, but instead of generating text, they predict DNA sequences.
Specifically, the models were enhanced to produce bacteriophages, which are viruses that attack specific bacterial strains.
From thousands of generated variants, the researchers identified 302 promising candidates and successfully synthesized 16, which showed effectiveness against Escherichia coli (E. coli), with some overcoming the bacteria's defenses.
Notably, the sequences produced displayed unique patterns not found in naturally occurring organisms.
Graduate student Samuel King recounted the moment when the team detected viral activity in the lab, noting clear spots on Petri dishes with bacteria—indicative of viral replication. Stanford associate professor Brian Hays remarked that the moment was met with applause from the team when the results were shared.
Concerns and Implications for Biosecurity
The creation of new bacteriophages could significantly impact the treatment of antibiotic-resistant infections, but it has also raised biosecurity concerns. In a commentary accompanying the study, experts from the Johns Hopkins Center for Health Security highlighted the potential risks, questioning the implications of designing viruses with enhanced capabilities.
“You could say, ‘Hey, genomic language model, make me a flu genome modified to be more contagious or more lethal,’” cautioned Moritz Hanke.
Experts have warned against pursuing the creation of viruses that could cause human diseases. The researchers took measures to mitigate risks by excluding viruses that infect complex organisms from their training data and conducting their work in secure laboratory conditions. Hays assured that existing safety protocols are in place to ensure the technology is used responsibly.
However, some experts, like Tom Ellis from Imperial College London, argue that the AI-generated viruses are relatively simple and pose minimal risk, emphasizing that creating a dangerous pathogen would require a far more complex approach. Mark Guell from Pompeu Fabra University in Spain called the study a pivotal development, stating that it marks the beginning of using computers to design biological systems.
Hays further noted that while viruses are not classified as living organisms, creating living cells using AI would require a significant technological advancement, as the genome of a phage consists of approximately 5,400 base pairs, while living cells have genomes with around 500,000 base pairs, and humans have three billion.
In July, AI startup Anthropic announced the launch of a platform for scientists called Claude Science, aiming to develop treatments for complex diseases.
