Researchers in the United States have used artificial intelligence to design and create functional viruses with no natural counterparts, in an experiment that significantly expands the capabilities of synthetic genomics.
The study, published in Science, was conducted by scientists from Stanford University and the Arc Institute in California. The researchers used an AI model to design new bacteriophages, viruses that infect bacteria, potentially opening a new avenue for tackling bacteria that have developed resistance to antibiotics.
Stanford chemical engineer Dr. Brian Hie said the approach could eventually contribute to the development of bacteriophage based therapies. The strategy relies on using multiple genetically distinct phages simultaneously, making it more difficult for bacteria to develop resistance against the entire combination.
At the same time, the ability to design viral genomes using generative AI is raising significant biosafety concerns.
Experts from the Center for Health Security at Johns Hopkins University warned that the technology is advancing faster than existing systems of oversight and governance.
The experiment
The researchers used Evo, an open source artificial intelligence model trained on vast amounts of genetic data from animals, plants, microbes and viruses. Genetic sequences from viruses known to infect humans were excluded from the training data.
The team then focused on bacteriophage ΦX174, one of the best studied viruses that infects E. coli. The AI model was trained using the genome of ΦX174 together with sequences from related viruses.
Based on these data, the model generated 700,000 candidate genetic sequences. The researchers selected a small subset for laboratory testing and found that some of the designed sequences could produce functional phages.
According to the scientists, the result demonstrates that AI models can be used not only to analyze known biological sequences but also to design new ones.
A potential tool against antimicrobial resistance
The researchers also investigated whether the newly designed phages could overcome bacterial resistance to natural ΦX174.
They suggest that combining multiple genetically different phages could make it more difficult for bacteria to develop resistance against all of them simultaneously. In theory, this approach could form the basis for future treatments targeting difficult bacterial infections.
The possibility of applying the technology against drug-resistant bacteria is particularly significant given the growing global problem of antimicrobial resistance.
The researchers stress, however, that the work represents a proof of concept, rather than a ready-to-use medical technology. Considerably more research will be required before such approaches can be translated into clinical treatments.
AI is reshaping biology
The study is part of a broader trend toward using artificial intelligence in biological research. Modern AI systems can process enormous genetic datasets and identify patterns that would be extremely difficult to detect using conventional approaches.
One of the best-known examples is Google DeepMind’s AlphaFold, which transformed the prediction of three-dimensional protein structures and helped accelerate the development of computational biology.
The new research takes this capability a step further: from predicting and analyzing biological molecules to designing entirely new biological systems.
Biosafety concerns
The same capabilities that could accelerate the development of new treatments also raise questions about potential misuse.
Biosafety experts warn that generative AI systems are increasingly capable of assisting with the design of genetic sequences in ways that previously required highly specialized expertise and substantial computational resources.
Synthetic genomics engineering professor Tom Ellis of Imperial College London, however, argued that directly modifying existing pathogens could represent a more immediate threat than designing an entirely new pathogen with AI.
At the same time, concerns surrounding gain-of-function research experiments that alter organisms to introduce or enhance specific characteristics have contributed to calls for tighter biosafety policies in the United States.
The central question is therefore no longer simply what AI can design, but how its rapidly expanding capabilities in synthetic biology can be governed and used safely.