A team from Stanford University and the Arc Institute used an AI model to design new viruses that killed bacteria in the lab. The model, called Evo, proposed 700,000 possible genomes, from which the team selected 285 for chemical synthesis. Sixteen of those produced viruses capable of replicating. The work, initially shared as a preprint, has now been peer-reviewed and published in Science. The New York Times reported on the study's details, particularly the hit rate of the AI's designs. The researchers inserted the synthesized DNA into bacteria and observed the viruses' ability to replicate and destroy their hosts. The findings mark a significant step in AI-driven biological design, according to the study's authors. The project's success has sparked discussions about the implications for medicine and biotechnology. Source: thedecoder

The AI model, Evo, was trained on about two million bacteriophage genomes before focusing on the 11 genes of the phage Phi X-174 and its relatives. This dual-phase training allowed Evo to generate new genetic codes that were not just weak copies of existing viruses but proved as robust as natural ones. Some of the AI-generated viruses even replicated faster than Phi X-174. The team's approach, which involved chemically synthesizing DNA and testing it in bacterial hosts, demonstrated the potential of AI in creating functional viral genomes. The study's co-author, Brian Hie, emphasized the importance of the work in advancing AI's role in biological design. Source: thedecoder

The research team faced challenges in ensuring the safety of their AI model. Evo was trained without data on human pathogens or related pathogens from animals, plants, or fungi, preventing it from generating genomes for such organisms. Despite these precautions, concerns remain about the potential misuse of the technology. Moritz Hanke of the Johns Hopkins Center for Health Security highlighted the gap between the pace of AI research and existing biosafety regulations. The U.S. National Institutes of Health has policies that ban experiments making pathogens more dangerous, but purely computational work, such as designing viral DNA on a computer, is not covered unless it involves an entity of concern. Source: thedecoder