The artificial intelligence models Evo 1 and Evo 2 designed hundreds of new versions of a tiny bacteriophage. 16 of the 285 genomes constructed in the lab created active viruses capable of replicating in E. coli cells. The achievement could advance future treatments for antibiotic-resistant bacteria, but also raises questions of biosafety.
artificial intelligence She planned viruses Capable of infecting Bacteria and reproduce within them. This is a first-of-its-kind achievement, showing that artificial intelligence is beginning to do more than analyze the genetic code of living things: it is also capable of writing new versions of it.
The viruses created are bacteriophages, or phages for short – viruses that infect bacteria, not humans, animals or plants. The researchers used artificial intelligence to design hundreds of new versions of a well-known phage, and then assembled the designed genomes in the laboratory.
Of the 285 genomes designed by artificial intelligence and tested in the experiment, 16 produced functional phages that were capable of infecting bacteria. Escherichia coli Inaudible.
The study, published in the journal Science, is an important step in the field sometimes called "Creative biology", where artificial intelligence is used to design new molecules and biological systems. However, it is important to understand what exactly was demonstrated in the study – and what has not yet been demonstrated.
The natural predators of bacteria
Phages are essentially predators of bacteria. They are among the most common and diverse biological entities on Earth, having evolved alongside bacteria for billions of years.
Most phages are highly specialized for a particular type of bacteria. They recognize molecules on the surface of the bacteria, attach to it, and inject their genetic material into it. The phage takes over the molecular machinery of the bacteria, forcing it to produce new copies of the virus. Eventually, the infected cell splits and releases the new phages.
These properties make phages a particularly suitable system for testing the capabilities of artificial intelligence. Their genomes can be very small, so the DNA can be produced in the laboratory in a relatively short time and tested to see if it works. The result of the experiment is also clear: the genetic instructions create a functioning phage – or they fail to do so.
The artificial intelligence systems used in the study, Evo 1 and Evo 2, operate on genetic sequences in a way that is reminiscent of the way large language models, such as ChatGPT, operate on text.
A language model learns patterns in words and sentences. A genomic model learns patterns in DNA sequences.
But a genome is not just a string of instructions that can be changed one letter at a time. Its different parts must work together. A phage's DNA must contain instructions for identifying the right bacterium, taking over its machinery, producing the right proteins at the right time, and assembling new virus particles.
More than two million genomes were used for training
The researchers trained the AI models on more than two million phage genomes. They then asked them to design new versions of a small, well-studied phage called ΦX174.
The ΦX174 genome contains only about 5,400 letters of DNA and instructions for making 11 proteins. It infects a strain of E. coli that is not associated with disease, so it serves as a relatively simple system for testing the technology.
The researchers also performed further fitting of the models using the genomes of approximately 15 relatives of ΦX174.
So it's not true to say that AI invented a virus out of thin air. It used patterns found in existing biology to create new combinations within a system that scientists already know well.
After the design phase, the researchers chemically synthesized the DNA sequences and introduced them into E. coli cells to see if they could produce functional phages. Most of the sequences failed, but 16 of the 285 designs were successful.
Several of the phages that were created behaved quite similarly to ΦX174, even though their DNA sequences were significantly different. In one case, the genome created by the AI included a combination of genetic components that could not have functioned in the original genome, but did function within the new, adapted genome.
Will phages be able to bypass bacterial resistance?
The researchers also tested whether the new phages could overcome bacterial resistance. They exposed them to a version of E. coli that was resistant to the original phage. After repeated cycles of exposure, hybrid phages emerged that were able to infect the resistant bacteria as well.
The AI didn't directly design these final phages. Instead, it created a diverse source population, which gave evolution more options to act on.
This approach may help in the future in the fight againstAntibiotic resistanceBacteria are developing increasing resistance to antibiotic drugs, and some of the infections they cause are already becoming very difficult to treat. Phages have been studied for years as a possible alternative, because they are able to attack and destroy bacteria, and even evolve in response to the resistance that bacteria develop against them, without harming human cells.
One of the biggest challenges in phage therapy is finding the right virus for the bacteria causing the infection. In the future, artificial intelligence may help predict which phages are likely to be effective against certain drug-resistant bacteria.
The distance from treating patients is still great
There is a big gap between designing a phage on a computer and using it to treat a patient. Any potential treatment would need to be tested against the strains of bacteria that actually cause infections in humans to ensure it is effective and safe. It would then need to be manufactured to the same stringent standards that apply to pharmaceuticals.
The phage tested in the study is extremely simple. Many phages that might be useful against dangerous infections have much larger genomes and much more complex biological mechanisms. Scientists still don't understand much of these mechanisms.
Using phages tailored to specific patients will also raise complex regulatory questions. Therefore, the study should be seen as a proof-of-concept—not the arrival of a new AI-designed phage-based medical treatment.
Phages have been used in biology for decades to understand the fundamentals ofGenetics and the activity of cells. Now they provide researchers with a way to test whether artificial intelligence can do more than decode existing DNA: can it design an entire genome that actually works?
The next question is whether the method will also succeed in more complex systems. Can artificial intelligence design phages with larger genomes? Can it help deal with bacteria that cause serious diseases in humans? And can it do so with a reliability that allows for use outside the laboratory?
Questions of biosafety
The study involved a relatively simple phage that infects bacteria, not a virus capable of causing disease in humans. It does not prove that you can ask an artificial intelligence to create a dangerous virus and immediately receive a ready-to-use design.
However, it shows that artificial intelligence systems are beginning to move from reading biological sequences to creating new sequences capable of operating in the real world.
Therefore, it is increasingly important to examine how such systems are trained, what biological information they have access to, and how to screen for potentially dangerous designs.
These questions cannot be left to AI researchers alone. Biologists, doctors, regulators, ethicists, and biosafety experts will also need to be involved in the development of the technology.
For now, it is correct to view the achievement as a proof of principle: artificial intelligence was able to create new versions of an entire viral genome, and some of them actually functioned.
The more difficult task will be to discover how far this capability goes – and to ensure that understanding and monitoring of risks advances at a similar pace as the development of technology.
Questions and Answers
Did artificial intelligence create a virus from scratch?
Not exactly. The models were trained on more than two million phage genomes and adjusted using about 15 genomes close to phage ΦX174. They created new combinations based on patterns learned from existing organisms.
How many of the viruses that were designed actually worked?
Of the 285 genomes constructed and tested in the laboratory, 16 produced functional phages that were able to infect E. coli bacteria and reproduce within them.
Are these viruses dangerous to humans?
The phages tested infect bacteria, not human cells. The study does not show that artificial intelligence can create a virus that causes disease in humans at the touch of a button, but it does highlight the need for oversight and biosafety testing.
Can they be used instead of antibiotics?
In the future, artificial intelligence may help design or select phages against antibiotic-resistant bacteria. However, the path from experimental technology to approved treatment for patients is still long and requires efficacy, safety, manufacturing, and regulatory studies.
For the scientific article: Opening the scientific article
More on the subject on the science website
- The miracle sponges of phages
- The common fight of bacteria and plants against viruses
- Gut bacteria are able to change their "software" in response to inflammation
- Changing the code of life: Scientists have created a bacterium with a reduced artificial genome
- When the machine wears a lab coat: Artificial intelligence enters the lab
