Scientists have developed the first viruses fully designed by artificial intelligence (AI), Nature News reports (open copy). These viruses can seek out and destroy specific types of Escherichia coli, or E. coli - a bacterium often found in the gut that can cause food poisoning or other illnesses. The achievement marks a major advance in using ai to build complete genetic sequences (genomes) - the full set of DNA instructions that define an organism. A preprint is published in bioRxiv.
The researchers used AI models named Evo 1 and Evo 2, which analyze and generate sequences of DNA, RNA, and proteins. They based their designs on ΦX174, a simple virus with single-stranded DNA that infects bacteria and replicates inside them. The ai had been trained on over two million genomes from bacteriophages, or phages - viruses that only attack bacteria. Then, through supervised learning, the models were adjusted to create phages that specifically infect antibiotic-resistant E. coli, strains that no longer respond to common drugs.
Out of thousands of ai-generated sequences, the scientists selected 302 promising ones. They synthesized the DNA - chemically built it in a lab - and inserted it into host bacteria to produce the phages. Testing showed that 16 of these ai-designed phages could infect and kill E. coli, with some combinations working against three resistant strains that the original ΦX174 could not handle.
Testing the new phages
This success highlights ai's ability to manage complex gene interactions - how different parts of a genome work together for functions like replication and regulation. It suggests potential for new therapies in phage therapy - a treatment using phages to combat infections - and could help address antibiotic resistance, a growing health threat.
While exciting, the approach raises biosafety issues, like the risk of designing harmful viruses. The study avoided this by excluding data on viruses that affect humans or other complex organisms and using safe, well-studied systems. Experts note that ai still requires human guidance for safety and testing, but this work points to future uses in creating custom biological tools for medicine and public health.