Researchers from Stanford University and the Ark Institute have successfully used artificial intelligence to design complete genomes of new viruses capable of replicating and killing bacteria. This represents a step beyond using artificial intelligence to design a single protein or gene, to designing a complete genetic system that can function inside cells.
But the viruses designed by the team are not human viruses, but bacteriophages, a type of virus that specializes only in infecting bacteria and does not attack human cells.
The researchers used two artificial intelligence models known as "Evo 1" and "Evo 2". Their workings are similar to the language models that tools like "ChatGPT" are based on, but instead of learning the order of words in sentences, they learn the patterns found in DNA sequences.
The team chose a small phage called ΦX174 as a starting point. It is a known phage that infects Escherichia coli bacteria, and its genome contains only 11 genes.
According to the study published in the prestigious journal Science, the models were trained on thousands of nearby genomes, and then the scientists asked them to produce new sequences that preserve the genes and elements necessary to form a virus capable of replication.
Note that the artificial intelligence did not build the viruses directly, but suggested DNA sequences to the computer, and then the researchers chemically created these sequences and introduced them into bacterial cells in the laboratory.
In addition, the results do not mean that the virus was designed from scratch without any natural basis. Artificial intelligence learned from the genomes of phages found in nature, and then rearranged the patterns and genetic changes in new ways.
According to the study, researchers tested 285 AI-designed genomes, and 16 of them succeeded in producing real phages that were able to reproduce inside bacteria and slow their growth or kill them.
This means that most of the designs did not work, which reveals that the technology is still in its early stages, according to the researchers. However, the result was important because it showed that an artificial intelligence model can suggest complete viral genomes, some of which are actually functional.
The tests also showed that the new bacteriophages did not infect all types of bacteria, but remained highly specialized in infecting specific strains of E. coli, according to the study.
This result suggests the possibility of using artificial intelligence in the future to design phages that help treat antibiotic-resistant bacterial infections.
Theoretically, a large number of designs could be produced, and then the phages best able to infect a specific type of bacteria could be selected. However, the experiment does not represent a ready-made treatment; it was conducted in the laboratory using non-pathogenic bacterial strains.
Tags:
science
