How AI could help crack vaccine science’s hardest problem
Updated: Sep 18
AI may be able to help regional biotechs develop the next generation of vaccines by helping more researchers find answers faster.
Vaccine-preventable diseases like rotavirus disproportionately affect children in low- and middle-income countries. Photo: Chidi Young
An infant in a village clinic in Malawi could, within a decade, get a rotavirus vaccine that works just as well as the one infants in London or Boston get every day. That is not the case today, despite rotavirus vaccines having existed for close to two decades. Anyone who works in this field knows the quiet embarrassment behind that gap: the vaccines we currently have tend to protect babies in wealthy countries far better than they protect babies in the low- and middle-income countries that carry most of the rotavirus disease burden.
Poor nutrition, maternal exposure, gut microbiome differences, competing infections—all of it chips away at how well a rotavirus vaccine trains a still-developing immune system. Closing that protection gap costs hundreds of millions of dollars, and takes years we do not have to re-solve a problem we already thought we had solved.
Introducing the AI co-scientist
Much of that cost and delay stems from how vaccines are proven safe and effective in the first place. Normally, it is done the hard way: give people the vaccine, wait, and count how many still get sick compared to those who did not receive it. That can take years. But for some diseases, researchers have found a more efficient way—a signal, often in the blood but sometimes in the stool or other bodily fluids, that reliably shows whether someone is protected, without waiting to find out the hard way.
Scientists call this a “correlate of protection”. It is why an updated Covid-19 or flu vaccine does not need a fresh multi-year trial every time a strain shifts. Regulators can clear it based on a much smaller study, one that only needs to show people had the expected immune response. Correlates let vaccines keep pace with a mutating virus, bringing speed and cost savings.
Many diseases do not have an established correlate of protection. Finding one usually takes years of systematic lab testing or a serendipitous discovery by a scientist in the field. But published scientific literature is now so vast that promising correlates may already sit buried in old papers, described once and abandoned for reasons unrelated to whether they worked.
Searching that volume of information would take a human team years. This is the promise of a new class of AI tools often called “AI co-scientist”. Used well, these tools can comb through decades of scattered research on rotavirus, malaria, or TB in hours, and return dozens of potential candidates for new correlates of protection.
Benchmarking tools on the market
These new AI tools come with a risk. A genuinely new hypothesis has nothing to check itself against, unlike AI-generated clinical advice, which can be measured against existing guidelines. If we cannot tell how good a model’s suggestions really are, a well-meaning researcher might pick up the wrong tool for the job, one that sends them down a dead end for months or years.
This is what PATH is working on, and we see our role as trying to prevent the bad outcomes we cannot afford. One part of that is benchmarking the tools already on the market, so a research team without a dedicated machine-learning group, or the resources to evaluate hundreds of candidate hypotheses themselves, knows in advance where a given tool holds up and where its limits lie. That can make the difference between a false lead caught early and one chased for months before anyone realises it was never real.
Getting a vaccine from an idea to an approved product can take 10 years and cost hundreds of millions of dollars, most of it spent proving to regulators that the product works and is safe. A multinational company can spread that risk across dozens of programmes and shrug off the ones that fail. A regional biotech, closer to the patients but working with a fraction of the money, usually cannot survive even one failed trial. That is a large part of why so much vaccine development for diseases of poverty, such as TB and malaria, has relied on philanthropic and public funding rather than venture capital. The road was too long and too risky for anyone without deep reserves.
Towards a new era of access
If AI tools prove reliable, it will affect not only how quickly science moves, but also who gets to compete. Shorter, sharper trials mean less money is needed to get from candidate to approval, which changes who can realistically participate. A company that could never have absorbed a decade of trial risk might only need to absorb one or two years of it, opening the field to teams that were priced out before.
None of this is guaranteed. Equitable access to these tools is its own problem, and solving it will take real investment. But if we get it right, the next rotavirus vaccine could be built by a team that understands local disease patterns rather than a global average. It could be brought to market by a company founded a few years ago in Nairobi, Hyderabad, or Dakar, close enough to the problem to see what a larger, more distant company would miss.
Go back to that infant. AI alone will not have cured a disease; nobody serious would claim that. What will have changed is that a door once open to only the best-funded institutions is now open to many more. Getting through it will still take years of hard, unglamorous work. But there are potentially more hands able to try than there were even five years ago, and that is worth celebrating.
The opinions expressed are those of the authors and do not necessarily reflect the position of Re:solve Global Health.
Dr Roberto Amato is a computational biologist with years of professional experience in AI and machine learning, and a career built in global health. His research has directly shaped public health policy and practice. He leads the AI for science portfolio at PATH, working to bring frontier AI tools safely and usefully into vaccine and drug development for diseases that have long been underserved.
Professor Bilal A Mateen is a physician by training and an internationally recognised expert in AI applications in health, spanning the implantation of frontier solutions in low-resource settings to biomedical science applications. He serves on the UN’s Independent International Scientific Panel on AI and is the chief AI officer at PATH.
