Every year, public health organizations around the world face a critical decision: Which influenza strains should be included in the next season’s flu vaccine? This choice must be made well in advance, often months before flu season begins — and the stakes couldn’t be higher.
The Flu Forecasting Challenge
If the selected strains match those that circulate widely, the vaccine can be highly effective. But when predictions miss the mark, it can lead to reduced protection, increased illness, and overwhelmed healthcare systems. The unpredictable nature of influenza, constantly evolving much like its viral cousin COVID-19, makes accurate forecasting especially difficult.
Introducing VaxSeer: AI-Powered Flu Surveillance
To address this challenge, researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and the MIT Jameel Clinic have created an artificial intelligence tool named VaxSeer. This innovative system is designed to predict dominant flu strains and recommend the most effective vaccine candidates — all months before flu season arrives.
VaxSeer leverages deep learning algorithms trained on decades of viral genome sequences and laboratory test data. Unlike traditional models that examine single mutations in isolation, VaxSeer uses a large protein language model to understand how combinations of mutations impact viral dominance. This allows it to detect and adapt to the dynamic shifts in viral evolution that make flu forecasting so challenging.
Smarter Predictions Through Dual Engines
VaxSeer features two main prediction engines: one estimates how likely a viral strain is to spread (dominance), while the other measures how well a vaccine can neutralize that strain (antigenicity). Together, these engines generate a “predicted coverage score” — a forward-looking metric that forecasts how effective a vaccine will be against future flu variants.
This score ranges from negative infinity to 0, with values closer to zero representing a better match between the vaccine and circulating viruses. It’s essentially a new way to quantify vaccine effectiveness long before patients ever roll up their sleeves.
Outperforming the WHO: A 10-Year Retrospective
To test its capabilities, researchers conducted a 10-year retrospective analysis comparing VaxSeer’s strain selections to those of the World Health Organization (WHO) for two major flu subtypes: A/H3N2 and A/H1N1.
For A/H3N2, VaxSeer’s recommendations beat the WHO’s in nine out of 10 flu seasons. For A/H1N1, it matched or outperformed in six of 10 seasons. In one standout case — the 2016 season — VaxSeer identified a strain that the WHO didn’t recommend until the following year. Its predictions closely aligned with real-world vaccine effectiveness data from the CDC, Canada’s Sentinel Practitioner Surveillance Network, and Europe’s I-MOVE program.
How It Works: Behind the Algorithm
VaxSeer’s model begins by estimating the spread of various flu strains over time using a protein language model that factors in competition among strains. It then uses mathematical simulations based on ordinary differential equations to visualize how dominance evolves.
To evaluate antigenicity, the system predicts how well a particular vaccine strain will perform in hemagglutination inhibition (HI) assays — a standard test that measures how efficiently antibodies can prevent a virus from binding to red blood cells.
A Step Toward Predictive Medicine
By simulating viral evolution and vaccine interactions, VaxSeer offers a powerful new tool for public health decision-making. “AI tools like this one could enable faster, more accurate vaccine development — helping us stay ahead in the battle between infection and immunity,” says MIT researcher Wenxian Shi.
Currently, the model focuses on the hemagglutinin (HA) protein — the main surface antigen of the flu virus. Future versions could expand to include other proteins like neuraminidase (NA), as well as considerations such as immune history, manufacturing constraints, and dosage variations.
The development of VaxSeer is part of a broader trend in the use of AI to predict complex biological systems. For example, efforts to forecast the evolution of antibiotic-resistant bacteria or drug-resistant cancers could similarly benefit from this kind of predictive modeling. These advances reflect a larger shift toward using AI to proactively tackle the world’s most adaptive diseases.
Looking Ahead: Beyond Influenza
Though VaxSeer is tailored for flu forecasting, its underlying framework has potential applications across other rapidly mutating pathogens. However, adapting it for new viruses requires extensive, high-quality datasets — something researchers are actively working to develop in AI-driven health and environmental monitoring as well.
With the speed of viral evolution threatening to outpace therapeutic development, tools like VaxSeer could play a crucial role in bridging the gap between emerging threats and timely interventions.
Read the full open-access study in Nature Medicine.
