Artificial intelligence has crossed a critical threshold in modern biology. For the first time, it has produced entire viral genomes that function in laboratory conditions. This development extends beyond digital tasks and shows how advanced systems described in future technology transforming daily life are entering experimental science. The work focuses on bacteriophages, viruses that attack bacteria, and delivers measurable biological results.
Table of contents:
- Brian Hie and Stanford University research team
- Training AI to write DNA sequences
- Testing bacteriophages against resistant bacteria
- Johns Hopkins Bloomberg School of Public Health perspective
Brian Hie and Stanford University research team
The study confirms that two AI models generated full DNA genomes for 16 bacteriophages capable of infecting Escherichia coli. These viruses were tested in lab dishes and demonstrated active replication. This step marks a shift from earlier AI applications limited to isolated genes or proteins.
Brian Hie, a computational biologist at Stanford University in California, led the research. His team released the findings on September 17 through bioRxiv, a scientific platform that hosts preprint studies. It is the first documented case of AI creating an entire genome, even as researchers continue to debate whether viruses qualify as living organisms.
Training AI to write DNA sequences
The AI systems were trained on billions of DNA base pairs composed of A, C, G and T. These datasets came from existing bacteriophage genomes. The process mirrors how language models learn text patterns, a concept also explored in reinforcement learning research, but here the language is genetic.
To guide the models, the researchers used the well-known bacteriophage ΦX174. Its genome was fully decoded in 1977 and is among the most studied viral DNA sequences. This reference allowed scientists to compare natural viral genomes directly with AI-designed versions. Phages were chosen because they do not infect humans, and the models were not trained on disease-causing viruses.
Testing bacteriophages against resistant bacteria
The AI produced about 300 candidate genomes. From these, 16 resulted in functional viruses that successfully infected Escherichia coli. Several AI-designed phages killed the bacteria faster than ΦX174. Laboratory measurements confirmed these effects.
ΦX174 alone could not eliminate three resistant strains of Escherichia coli. However, mixtures of AI-generated phages adapted rapidly and destroyed those strains. This outcome suggests a practical pathway for addressing antibiotic resistance, a challenge also discussed in more here within emerging biomedical strategies.
- 300 AI-generated genome designs
- 16 functional bacteriophages
- 3 resistant bacterial strains eliminated through phage mixtures
Johns Hopkins Bloomberg School of Public Health perspective
Kimberly Davis, a microbiologist at the Johns Hopkins Bloomberg School of Public Health in Baltimore, reviewed the findings independently. She was not involved in the research.
She stated verbatim
“The need to find a phage that targets [such a ‘superbug’] strain would be very urgent,” says Kimberly Davis. “AI could be a powerful way of rapidly generating a phage match to treat patients.”
The study demonstrates that artificial intelligence can design complete biological systems with real-world laboratory impact.
Source: Science News Explores
FAQ
What did artificial intelligence create in this study?
Artificial intelligence designed complete and functional DNA genomes for bacteriophages that infect Escherichia coli.
Who led the research on AI-designed viral genomes?
The research was led by Brian Hie, a computational biologist at Stanford University in California.
How many AI-designed bacteriophages were able to infect bacteria?
Out of about 300 genome designs, 16 resulted in functional bacteriophages that infected Escherichia coli.
Why were bacteriophages chosen for this experiment?
Bacteriophages were selected because they infect bacteria rather than humans and are safe to study in laboratory conditions.