This time, the virus is on the patient’s side — and AI helped design it
Bacteria have evolved ways to dodge antibiotics. Could viruses that kill those bugs give doctors some much-needed backup — and what are the risks?
AI can write emails, create seasoning blends or put pets in Renaissance outfits. One adventurous baker used ChatGPT to concoct a cake from sauerkraut and seaweed.
Now, there’s an AI-made virus on the scene.
And no, it’s not time to run for cover or dig out your face mask. While some viruses can make you sick, others — known as bacteriophages — cure disease by invading harmful bacteria.
In October, the biotech company Radical Numerics reported the first use of AI-designed viruses in a patient with an antibiotic-resistant infection. The team used a model called Omnii to design them. The work was a collaboration with another biotechnology company, Tabula Bioscience, which led the broader effort, according to Radical Numerics.
For decades, doctors have relied on antibiotics to treat illnesses like strep throat and bacterial pneumonia.
“They were called the magic bullet,” said Northeastern biotechnology professor Sudhakaran Prabakaran. But bacteria evolved ways to survive. More than 9.3 million drug-resistant infections occur in the U.S. each year, according to the Centers for Disease Control and Prevention, the nation’s health protection agency. Radical Numerics warns that antibiotic resistance could claim nearly 40 million lives between 2025 and 2050.
Bacteriophages, or phages for short, give the disease-fighting task force some much-needed backup. The trick is finding one that attacks the particular strain making a patient sick.
Pitting one microscopic fighter against another is cutting-edge stuff. But does it actually work? And could designing viruses unleash new problems? Northeastern Global News asked Prabakaran, who uses AI to study genomes and recently published a book on the subject, and other experts what the new approach might mean.
How does a virus take on a bacterium?
To create tailor-made phages that get past the microbe’s cellular defenses, scientists first need to identify the exact strain of bacteria infecting the patient.
Each strain has a unique set of molecular equipment — proteins encoded in sections of DNA called genes. To launch an effective attack, a virus needs its own set of corresponding molecular machinery — protein-based “legs” that allow it to grab a foothold of the bacterium, Prabakaran explained.
When making the virus out of an existing prototype, scientists can change the shape of the phage’s legs by tweaking the genes coding for them. The new virus can then latch onto a strain of bacteria it couldn’t infect before.
If the virus gets its DNA inside the cell, it can put the bug’s cellular machinery to work to make copies of itself.
Where does AI come in?
Figuring out which genetic tweaks produce the desired traits is easier said than done — and that’s where AI can help.
First, the model has to learn what that organism’s DNA typically looks like.
A DNA sequence is a chain of chemical building blocks called nucleotides. They come in four types and string together in a myriad possible combinations. For members of the same species, many stretches of DNA match, while others vary.
By comparing lots of sequences, the model learns which nucleotide building blocks tend to show up together. Give it a similar stretch of DNA with a gap, Prabakaran explained, and it can use the surrounding nucleotides to make an educated guess about what belongs there.
In designing a phage, scientists keep what already works and customize the part of the virus that anchors it to the bacterial cell. The model can suggest changes to the DNA that codes for those bits.
Think of it like a prompt, Prabakaran explained. A scientist asks for a virus that can grab onto a particular bacterial strain. Then, out pops several candidates to test in the lab.
“Maybe two of those options result in a viable organism,” he said. “And one of them can actually kill the bacteria.”
Radical Numerics is keeping the way the AI-made virus was administered to the patient private. As for phage therapy in general, treatment details depend on the particular infection. It’s not necessarily a one-shot deal, and doctors often use repeat doses. They may deliver the viruses through an IV, apply them to a wound or prescribe an inhaled treatment. The dose and duration are tailored to each case.
Editor’s Picks
Editor’s Picks
Editor’s Picks
How close is made-to-order virus treatment at this point?
This particular treatment still has to make it through clinical trials and the Food and Drug Administration, or FDA, the agency charged with protecting public health, hasn’t given it the official nod. But it authorizes use on a case-by-case basis by patients who’ve run out of other options want to give an experimental therapy a shot.
That said, the Radical Numerics report hasn’t gone through peer review, and the outcome for the patient is unknown. Details couldn’t be shared due to privacy concerns, the report explained.
Could making new viruses cause new problems?
Prabakaran said he’s “cautiously optimistic” about the development, but acknowledged that creating synthetic organisms can quickly take a wrong turn.
“We’re creating viruses to kill bacteria. But what happens when they evolve in such a way that they go out of our control?” Prabakaran asked.
As the virus makes copies of itself, mutations creep in. Over time, those changes could undermine safeguards meant to protect the patient whose body serves as the battleground in the fight against the bacteria.
Alexander Sulakvelidze, president and chief executive officer of Intralytix, a Maryland-based biotechnology company that develops bacteriophages, expressed excitement about the work and its potential to make treatments more precise and effective. At the same time, he noted broader concerns about AI-driven biological design, flagging unreliable AI predictions, unintended effects and the risk of using the technology to create “harmful biological agents.”
He also stressed the need for human oversight to make the most of what AI has to offer.
“AI should complement, rather than replace, scientific expertise and laboratory testing,” Sulakvelidze said.
Then, there’s the question of FDA approval. If each patient needs a different version of the virus, would the agency have to sign off each time?
“All of these questions are open,” Prabakaran said.
Could an AI-made virus be the next magic bullet after all?
If AI-designed viruses live up to their promise, running out of antibiotics that work may no longer mean running out of options.
Doctors could also use strategic tweaks to go after bugs causing infections while sparing beneficial ones that often get disrupted by traditional antibiotics.
“This could transform infectious disease management … by making treatments more precise, effective and less disruptive to the patient’s normal microbiome,” Sulakvelidze said.
The approach could also help curb outbreaks on a much larger scale.
“Tailored phages also have exceptional potential for addressing the global burden of diseases such as cholera,” said Minmin Yen, co-founder and chief executive officer of PhagePro. The Boston-based biotechnology startup aims to curb the spread of cholera with an oral product. While AI-enabled therapy can enhance personalized medicine, its reach could extend to entire communities, she explained.
