California: Artificial intelligence has been successfully used to design complete genomes for new functional viruses for the first time, marking a significant advance in synthetic biology while prompting fresh concerns about biosafety and security.
Researchers at Stanford University used AI models known as Evo1 and Evo2 to generate designs for bacteriophages, viruses that infect bacteria rather than people. The resulting viruses were capable of functioning and replicating under laboratory conditions.
The researchers selected 302 of the most promising AI-generated designs and synthesised them in the laboratory. Of those, 16 successfully killed E. coli bacteria, demonstrating that generative AI could design complete viral genomes capable of performing biological functions.
The models work in a way broadly comparable to large language models, which learn patterns and predict sequences of text. Instead of predicting words, Evo1 and Evo2 were trained to recognise and generate patterns in genetic sequences.
The models were trained using genetic information from viruses, bacteria, plants and humans before being refined to generate bacteriophages targeting specific bacterial species. Researchers said the development could eventually help scientists create new phages to combat infections that have become resistant to conventional antibiotics.
AI-assisted biological design could also contribute to the development of medicines, enzymes and other therapies. However, the ability to use generative AI to design complete viral genomes has also raised concerns about potential misuse.
Experts from the Johns Hopkins Center for Health Security, writing in a commentary accompanying the research in the journal Science, said the findings raised urgent questions about biosafety and biosecurity as AI systems become increasingly capable of designing novel biological systems.
The Stanford researchers implemented several safeguards during the study. Viruses capable of infecting more complex organisms were excluded from relevant training data, the experiments focused on bacteriophages rather than human-infecting viruses, and laboratory work was conducted under controlled conditions.
The achievement represents a significant increase in the complexity of biological structures that can be designed using generative AI. The bacteriophage genomes contained roughly 5,400 base pairs, considerably smaller than even the simplest living cells.
The breakthrough also demonstrates how AI is moving beyond digital applications and increasingly being used to design biological structures that can be physically created and tested in laboratories.
YOU MAY LIKE | Meta reveals AI model hacked external system during safety testing
