In a significant development for synthetic biology, scientists have successfully designed and constructed a virus from scratch using artificial intelligence. The new virus, which has never existed in nature, was specifically created to target and destroy harmful E. coli bacteria, opening new possibilities for combating antibiotic resistance.
The project, led by researchers at the startup Genyro, combined a powerful AI model for generating genetic code with a breakthrough technique for assembling DNA. The result is an artificial virus that is 25% more effective at killing bacteria than its natural counterparts, demonstrating a new era where life forms can be designed for specific purposes.
Key Takeaways
- Researchers used an AI model called Evo2 to design a completely new virus, Evo–Φ2147.
- The artificial virus was created to kill infectious E. coli bacteria.
- In tests, the AI-designed viruses were 25% more efficient at destroying bacteria than natural viruses.
- The technology could accelerate the development of new medicines, including personalized cancer vaccines.
- The breakthrough also raises concerns about the potential for AI to be used in designing bioweapons.
The Dawn of AI-Driven Evolution
For billions of years, life on Earth has evolved through natural selection. Now, scientists have introduced a co-author to that process: artificial intelligence. A team at Genyro, a startup founded by British scientist Dr. Adrian Woolfson, has developed a system that can create novel biological entities.
The newly created virus, named Evo–Φ2147, contains just 11 genes and is composed of 5,386 DNA base pairs, a tiny fraction of the 3.2 billion found in the human genome. While simple, its creation marks a pivotal moment in biotechnology.
"For the last 4 billion years, all life on Earth has evolved by the trial-and-error process of Darwinian evolution by natural selection, which lacks any foresight or intention," Dr. Woolfson explained. "Natural evolution now has a co-author."
This achievement was made possible by the convergence of two powerful technologies: an AI that can write genetic code and a new tool that can assemble that code into physical DNA with unprecedented accuracy.
How to Build a Virus with AI
The process of creating Evo–Φ2147 involved two core innovations that work in tandem. The first is an AI model named Evo2, and the second is a DNA assembly method called Sidewinder.
Evo2: The Genetic Code Generator
Similar to large language models like ChatGPT that are trained on text, Evo2 was trained on a massive dataset of nine trillion DNA base pairs. This taught the AI the fundamental rules of genetics, allowing it to generate entirely new and functional genetic sequences for organisms that have never existed.
For this project, the scientists tasked Evo2 with designing viruses based on a natural virus known to infect bacteria. The AI generated 285 new viral designs, from which the team selected the most promising candidates to build in the lab.
Sidewinder: Assembling the Blueprint of Life
Once the genetic code was designed, it needed to be physically constructed. In the past, assembling long strands of DNA was a complex and error-prone process. Dr. Kaihang Wang, an assistant professor at the California Institute of Technology and inventor of the Sidewinder technology, compares the old method to reassembling a book from torn-up pages without knowing their order.
The Sidewinder Analogy
Dr. Wang explained that Sidewinder essentially adds "page numbers" to the small fragments of DNA that are synthesized. This allows them to be arranged in the correct order automatically, dramatically improving the speed and accuracy of genome construction. The "page numbers" are removed after assembly, leaving a complete and correct DNA sequence.
This new method is reportedly 1,000 times cheaper and 1,000 times faster than previous techniques. Together, Evo2 and Sidewinder give scientists the ability to go from a digital design to a physical biological entity in a matter of days instead of months.
A New Weapon Against Superbugs
The primary goal of this research was to address one of the most urgent threats in global health: antibiotic resistance. As bacteria evolve to resist existing drugs, infections are becoming harder to treat, leading to hundreds of thousands of deaths annually.
The scientists wanted to see if they could design phage therapies—viruses that infect and kill bacteria—that could overcome bacterial evolution. The most successful AI-designed viruses were 25% quicker at killing E. coli than the wild variants they were based on.
In a joint blog post, Dr. Samuel King and Dr. Brian Hie, co-creators of the virus, stated their motivation. "Bacterial resistance to antibiotics represents one of the most pressing challenges in modern medicine," they wrote. "We wanted to see if we could one day design phage therapies that could be resilient against bacterial evolution."
Beyond fighting bacteria, the technology holds immense promise for other medical applications. Dr. Woolfson noted its potential for creating personalized cancer vaccines, which typically take 8 to 12 weeks to manufacture. "With the Sidewinder technology... we anticipate being able to do this in two to three days," he said.
The Dual-Use Dilemma
While the potential benefits are significant, the ability to rapidly design and create new life forms also carries substantial risks. Experts have long warned that AI could be used to accelerate the development of bioweapons.
Previous research has demonstrated that AI can design proteins mimicking deadly toxins like ricin and botulinum. Many of these AI-generated DNA sequences could potentially bypass the safety filters currently used by companies that synthesize custom DNA.
Organizations that monitor global threats, such as the Existential Risk Observatory, consider an AI-designed pandemic to be one of the five greatest risks facing humanity. Figures like Elon Musk and the late Stephen Hawking have also voiced concerns that advanced AI could pose an existential threat if not managed carefully.
To mitigate this risk, the creators of Evo2 took a critical precaution. "Evo cannot generate human viral sequences due to deliberate training data exclusions, preventing both accidental and intentional misuse for pathogen design," wrote Dr. King and Dr. Hie. This step ensures their specific model cannot be used to create viruses that infect humans, but it highlights the need for industry-wide safeguards as the technology becomes more widespread.