AI Predicts Viral Mutations to Design Future-Proof Vaccines

The wave-shaped chart Ratul Chowdhury pulls up on a computer monitor in his office captures the evolutionary cat-and-mouse game his research lab is up against.

The undulating curves track variants of the porcine reproductive and respiratory syndrome (PRRS) virus, which causes a swine disease that annually costs the global pork industry more than $1 billion – damage attributable in part to how quickly it adapts to escape from immune defenses. The PRRS virus has one of the fastest evolutionary rates among known animal-infecting RNA viruses, said Chowdhury, an assistant professor of chemical and biological engineering at Iowa State University.

The ability of the PRRS virus to change quickly gives it an evasive capacity that impedes effective vaccination. As hogs' immune systems figure out how to grab hold of the virus, new variants emerge even more rapidly. That's what the peaks in the wave graph suggest: the PRRS virus intensifying its mutations in search of designs that won't trigger an immune response, like a burglar testing different ways to bypass a home security system.

"The virus shape shifts," said Chowdhury, a Black and Veatch Building a World of Difference Faculty Fellow and a member of Iowa State's Nanovaccine Institute. "It tries to create a new structure and a new chemistry so that it can't be trapped and neutralized."

An ISU research group led by Chowdhury uses simulations powered by artificial intelligence to learn from and then predict viral mutations to develop vaccines that protect animals from a range of potential variations. With assistance from a state-funded initiative to encourage bioscience innovation and Iowa State's decades-deep trove of animal health genetic data, Chowdhury's work on future-proof vaccines is on a path toward commercial development.

Looking for Legos

Chowdhury's livestock vaccine research is based on advanced analysis of viral-host interactions. A deep understanding of how and where viruses attach to cells and infect them helps researchers identify which protein segments of a virus antibodies can recognize, grab and neutralize – vaccine targets known as epitopes.

As a computational structural biologist, Chowdhury is interested in the molecular configurations of those attachments. Antibodies need certain proteins folded in certain ways to snare a viral attacker. If the pieces don't match, immune defenses struggle.

It's a lot like Legos. But in the biological world, these are soft Legos. A tiny change can have a butterfly effect and give the structure a very different shape."

Ratul Chowdhury, computational structural biologist

AI makes quickly sorting through that complexity attainable. For PRRS, Chowdhury's team screened millions of possible protein shapes and mutations to identify which ones could help the virus avoid detection, a process he estimates was 100 times faster with AI modeling tools.

Armed with libraries of prioritized viral targets, researchers can develop stable forms of antibody-inducing antigens that prepare an animal's immune system for future threats. Ultimately, the goal is to stitch several of these antigens together to create robust multi-epitope vaccines to inoculate animals against numerous adaptations of a virus.

Viral Evolution Offers a Blueprint

In its work on PRRS, described in a study published in Computational and Structural Biotechnology Journal in late 2024, Chowdhury's research team identified 75 epitopes and designed 56 immunogens – short fragments of proteins that could be potential vaccines.

In a study published last fall in the American Chemical Society's Journal of Chemical Information and Modeling, the Chowdhury lab used AI tools to develop possible vaccines for infectious bronchitis virus (IBV), a common chicken pathogen. Researchers designed a digital platform, AstraMEV, built on 56 clusters of similar variations culled from public sequencing data to pinpoint which snippets of IBV's genetic makeup remain stable as it evolves.

"We can read the virus almost like a blueprint, identifying which parts stay the same across many strains and which parts help it enter chicken cells," Chowdhury said.

From the 258 epitopes identified in that analysis, researchers designed recipes for two different types of multi-epitope IBV vaccines: SmallTope (including up to eight epitopes) and BigTope (including up to 23 epitopes).

Multiple animal health companies, from large corporations to small startups, have told Chowdhury they plan to test his published vaccine research in their own labs, he said. He's also planning a small trial of a PRRS vaccine at an Iowa State swine farm later this year.

Industry Experience Next Door

Having Mike Roof in his corner has been essential in designing research well-positioned to translate into a marketable vaccine, Chowdhury said. Roof is Iowa State's chief technology officer for vaccines and immunotherapies, one of the four biosciences platforms the state of Iowa supports to transform existing strengths into innovation-powered economic growth. Roof spent the last decade of his private sector career as the global segment head of vaccine development at Boehringer Ingelheim, one of the world's largest pharmaceutical companies. His office at the Nanovaccine Institute is one door down from Chowdhury's.

"That's been tremendously helpful," Chowdhury said.

One of the benefits of having ready access to an industry veteran is identifying problems to solve. Chowdhury started looking at PRRS after Roof noted it has been a long-standing concern for pork producers. Roof also will oversee the ISU PRRS trial and helps make sure the lab's research meets the needs of biologic companies and regulators.

"Sometimes a little tweak or a slight change in study design can make the results more appealing to industry or for commercial success," Roof said. "For faculty with strong academic backgrounds but maybe not much industry experience, I can mentor them and help accelerate their work."

If the PRRS vaccine proves effective in live testing, CYVAX could be the next step in its progression toward commercial use. The small-scale wet lab in the ISU Research Park is designed to help commercialize university research and launch biotech startups, providing flexible and cost-effective space for federally complaint clinical trials. The intentionally industry-friendly setup can trim development time from a few years to several months, Roof said.

"This isn't a one-man show," Chowdhury said. "There's so much supportive infrastructure here."

A Built-In Advantage

Before coming to Iowa State in 2022, Chowdhury was a postdoctoral researcher at Harvard Medical School for two years studying the computational structural biology of cancer and pain treatment. He saw the promise of his AI-fueled computational approach but also the translational barriers of human health research.

One big obstacle is who has the best information to analyze. In human health, companies usually have the most valued data sets, he said. In animal health, universities are often ahead. Thanks in large part to its Veterinary Diagnostic Laboratory (VDL), which has the largest U.S. livestock caseload, few institutions have better animal health data than Iowa State.

The university's built-in advantage on livestock disease data was a crucial consideration for Chowdhury when pursuing a faculty position. The depth of Iowa State's in-house data makes his simulation models more effective, he said.

"If your data set is detailed, your methods don't need to be funky. We have the benefit of having rich data for understanding patterns, and that puts us in the driver's seat for performing cutting-edge research," he said.

The ongoing stream of anonymized data from the VDL – which conducted more than 1.7 million tests last year, a new annual high – also provides timely inputs for models, rooting them in what's actually happening in livestock herds, Roof added.

"You get large numbers of samples, which is important – not a single genome, but thousands of genomes. But you also get material that is more relevant because it's what's infecting animals today," Roof said. "There's no place in the world that that could feed people like Ratul better data to make their technology powerful."

'The Need is Huge'

Efficient and accurate prediction of viral mutations has many possible applications. In a paper published last spring in PLOS Computational Biology, Chowdhury and his co-authors described a model they built to predict how the virus that causes COVID-19 would evolve – and how that insight could be used to understand any other viral evolution.

Roof said animal health companies have contacted the Chowdhury lab with questions about how the technology could be used for a variety of cases, from reducing methane production by cattle to vaccinating against avian influenza.

"It's a novel way to identify vaccine targets or therapeutics and relatively quick and cost-effective, so he's attracting a lot of commercial interest," Roof said. "It's the new frontier."

Chowdhury said his frameworks for vaccine design are especially potent tools for commercial vaccine development in part because they are adaptable to any combination of virus and livestock, compatible with industrial-scale manufacturing constraints, and designed to be transparent and verifiable.

In the coming years, Chowdhury hopes to develop computational tool kits for a wider variety of livestock, resources companies could license to use for vaccine development or virus detection. He envisions building animal-specific tools like the AstraMEV platform, which in addition to identifying vaccine targets also analyzes the expected effectiveness and tolerance of a proposed vaccine.

Creating AI-driven vaccine-making roadmaps would leverage Iowa State's world-class expertise in veterinary medicine and animal health – and its in-house data assets – to speed responses to emerging diseases and strengthen defenses against more familiar threats, he said.

"We are technologically ready for this, and the need is huge," he said.