Could AI help us develop vaccines against viruses that don’t yet exist?
Could AI-designed vaccines help us stay ahead of future pandemics? We asked the vaccinologist behind a new AI-enabled biotech company what the AI revolution could mean for immunisation.
- 17 November 2026
- 7 min read
- by Linda Geddes
Working in Guinea during the 2014–2016 West Africa Ebola epidemic, Professor Jonathan Heeney became frustrated by the amount of time it took to develop and deploy vaccines against emerging infectious diseases. That experience inspired him to rethink how vaccines are designed.
Today, he and his colleagues at DIOSynVax are using AI to design antigens capable of triggering broad immune responses. He spoke to VaccinesWork about how the technology works and how it could help us to stay one step ahead of evolving viruses.
AI is not replacing immunologists or vaccinologists; it is helping us process vast amounts of biological data more efficiently, allowing us to explore biological possibilities at a much greater scale.
What is wrong with the way we currently design vaccines, and what are you trying to do differently?
Many vaccines have been extraordinarily successful. Diseases such as measles, mumps and rubella are controlled because the viruses change relatively slowly and vaccine protection remains durable.
The challenge comes with rapidly evolving RNA viruses such as influenza and coronaviruses. Current vaccines are typically based on a specific viral strain circulating today, but viruses continue to evolve, meaning that tomorrow’s variants may look quite different.
Credit: J.Garget/University of Cambridge
My goal is to reverse that paradigm. Instead of targeting the parts of viruses that change most rapidly, we use AI to computationally engineer novel vaccine antigens, using global sequence data to identify and combine the features that are shared across many virus variants.
The aim is to create broadly protective vaccines that can provide protection not only against today’s viruses, but also against future virus variants and potentially even be used to protect us against future outbreaks.
What exactly is a synthetic antigen, and how could it be incorporated into different types of vaccines?
AI-engineered antigens are genetic sequences that can be transcribed into amino acids. But whereas, say, the spike protein sequence used to create COVID-19 vaccines is representative of one SARS-CoV-2 variant, these AI-engineered antigens represent a whole spectrum of potential variants based on what we know occurs in nature and in humans.
By exposing the immune system to these shared features, we aim to induce antibodies and T-cell responses that recognise and generate protective immune responses against a wider spectrum of viral variants.
The technology is what we call ‘vector agnostic’, meaning we can incorporate these antigens into different vaccine platforms. Which one we choose will depend on the requirements of each vaccine and the populations we’re trying to protect.
For instance, some regions of sub-Saharan Africa lack a reliable cold chain, which is why we’ve chosen a modified poxvirus vector to deliver our pan-haemorrhagic fever vaccine. These are relatively sturdy and robust. In other situations we might decide to use mRNA, or a protein-based vaccine.
How does AI help you design these antigens, and what can it do that human researchers can’t?
Our AI-leveraged platform, DIOSynVax, analyses large numbers of viral sequences simultaneously and identifies the patterns that remain conserved across them. We then instruct it to help us generate new synthetic antigens that optimally represent those shared features that circulate globally.
A human researcher can compare dozens or perhaps hundreds of sequences, but AI can explore an immense design space containing millions of possible combinations and identify solutions that would be difficult or impossible to find by intuition alone.
Are humans still involved in the process?
AI is not replacing immunologists or vaccinologists; it is helping us process vast amounts of biological data more efficiently, allowing us to explore biological possibilities at a much greater scale.
We still draw on biological knowledge and immunology, and spend considerable time checking, validating and quality controlling our antigens through experiments.
Have you read?
We might take the top 10 to 20 antigens, make synthetic genes out of them, test each one in the laboratory, and then take the top three to six and test them in animals to see what kind of antibodies, T cell and broader immune responses they make. We’d then test those antibodies and immune cells against a whole bunch of different variants to see how they respond and, for instance, block entry or neutralise the wider subgroup of viruses they represent.
Does broader protection mean reduced potency against specific viruses?
In theory, focusing exclusively on highly conserved features could risk generating responses that are broader but less potent against a specifically targeted strain. However, we aim not to sacrifice potency for breadth. It is to identify designs that achieve both.
What’s encouraging from our studies is that we have consistently observed strong immune responses alongside broad cross-reactivity. However, these questions can only be answered through rigorous laboratory testing and ultimately through clinical studies.
Credit: J.Garget/University of Cambridge
That’s why we are actively pursuing manufacturing clinical, GMP-grade vaccines suitable for human clinical trials and engaging with regulatory agencies to advance these candidates for clinical use.
Your latest study investigated responses to an AI-engineered bird flu vaccine. What did it show?
The avian influenza virus A/H5 is continuing to rapidly change as it spreads worldwide.
We evaluated an AI-engineered H5Nx influenza vaccine against multiple genetically distinct avian influenza strains that have emerged in different regions and at different times.
The vaccine generated broad immune responses in mice and ferrets, and provided protection across variants that would normally present a significant challenge to conventional strain-matched approaches. In several key measures, including immune breadth and protective efficacy, it outperformed conventional bird flu vaccines stockpiled by many governments.
These protected only around 50% of animals exposed to multiple bird flu variants, whereas they were solidly protected with our vaccine.
We have also developed broadly protective vaccine candidates for seasonal influenza, and a universal influenza candidate covering seasonal as well as pandemic influenza threats.
What other vaccine candidates have you developed, and how far have they progressed through testing?
We have several broadly protective vaccine programmes at different stages of development. Our most advanced clinical programme is a pan-sarbecovirus vaccine designed to provide protection against SARS-related coronaviruses, including SARS-CoV (which causes SARS) and SARS-CoV-2.
This programme has already successfully completed its first in-human clinical testing and demonstrated that computationally designed broad-spectrum antigens can be safely administered while inducing broad immune responses.
We recently expanded this to a multivalent vaccine that could also protect against MERS coronaviruses, resulting in a broadly protective betacoronavirus vaccine targeting the coronavirus family most likely to generate future pandemic threats.
We have also developed a broadly protective haemorrhagic fever vaccine capable of protecting against Ebola Sudan, Marburg and Lassa fever viruses simultaneously. Rather than delaying time-critical outbreak responses, while the specific virus and potential vaccine is identified, a broadly protective vaccine could be used to immunise frontline healthcare workers and high-risk populations in advance.
We’re currently scaling up to get that vaccine candidate manufactured, before starting Phase 1 clinical trials in the UK.
Could this new approach to vaccine design require a different approach to testing and approval?
It is clearly impractical to run separate clinical efficacy trials to test a vaccine against every possible variant. Instead, developers increasingly use a combination of animal studies and tests of immune responses in humans to assess how broadly protective a vaccine might be in an outbreak. They base their designs using immune correlates known to predict protection against disease.
Influenza already provides a useful precedent. Updated influenza vaccines are routinely approved without conducting a new Phase 3 efficacy trial against every emerging strain. I believe broadly protective vaccines may ultimately require a similar framework.
What safeguards prevent AI from designing something harmful?
Safety is non-negotiable. Our AI-engineered antigens do not move directly from the computer into people. Every candidate undergoes extensive computational screening, laboratory characterisation, preclinical and stepwise evaluation in animal models, and then thorough regulatory review before any human testing can begin.
Importantly, these antigens are intended to stimulate broadly protective immune responses. They are not infectious viruses and cannot replicate or cause infection. The safety standards are exactly the same as those applied throughout modern vaccine development.
Could this approach make a difference in the next pandemic?
The greatest advantage is not necessarily designing vaccines faster, although AI certainly helps. The real opportunity is preparedness. Instead of today’s approach of starting from scratch after a new outbreak emerges, we hope to develop broadly protective vaccines in advance, evaluate them in early-stage clinical trials, and potentially have stockpiles available before the next spillover event occurs.