Every year, public health officials face a daunting gamble: Which flu strains should go into next season’s vaccine?
They have to decide months before flu season hits — with no guarantees. If they guess right, millions stay healthy. If they’re wrong? Hospitals overflow, lives are lost, and preventable illness spreads like wildfire.
But what if we could predict the flu’s next move; not with guesswork, but with precision? Enter VaxSeer, an innovation of AI from MIT’s CSAIL and the MIT Jameel Clinic. This tool is changing how we create flu vaccines. It uses deep learning to keep up with a virus that keeps evolving.
The Flu Forecast: Science Meets Prediction
Influenza is a master of disguise. It mutates constantly, swapping genetic code like a thief changing disguises. That’s why last year’s vaccine might not work this year. And during the pandemic, we saw this same game play out with SARS-CoV-2 variants — just as vaccines rolled out, new strains emerged.
Flu plays the same game — faster and quieter.
Traditional methods look at single mutations in the hemagglutinin protein. They don’t consider how these mutations work together. And, this can lead to a flu strain that tricks the immune system.
That’s where VaxSeer changes everything.
How VaxSeer Works: AI That Thinks Like a Virus (and a Scientist)
VaxSeer isn’t just another AI model. It’s a dual-engine prediction system trained on decades of viral sequences and lab data:
1. Dominance Predictor
Uses a large protein language model to forecast which flu strains will spread fastest — not by looking at single mutations, but by understanding how combinations of changes give a strain a competitive edge.
Unlike older models that assume viral populations are static, VaxSeer simulates real-time dominance shifts, mimicking the chaos of nature.
2. Antigenicity Forecaster
It predicts how well a vaccine will fight future strains. It does this by simulating results from the hemagglutination inhibition (HI) assay, the top lab test for checking antigenic match.
These engines produce a predicted coverage score. This score forecasts how well a vaccine is likely to perform before flu season starts.
The closer the score to 0, the better the match.
Results That Speak for Themselves
In a 10-year retrospective analysis, VaxSeer went head-to-head with real-world decisions made by the World Health Organization (WHO):
– For A/H3N2 — the most unpredictable and dangerous flu subtype — VaxSeer outperformed WHO choices in 9 out of 10 seasons.
– For A/H1N1, it matched or beat WHO recommendations in 6 out of 10 seasons.
– In 2016, VaxSeer flagged a strain that the WHO didn’t adopt until the following year — proving the model could see the future before we lived it.
Even more impressively, VaxSeer’s predictions closely aligned with real-world vaccine effectiveness data from the CDC, Canada’s surveillance network, and Europe’s I-MOVE program.
This isn’t just theory. It’s validation in the wild.
The Engine Behind the Magic
VaxSeer combines cutting-edge AI with classic epidemiology:
– A protein language model learns patterns in viral evolution.
– Ordinary differential equations (ODEs) simulate how strains compete and spread over time.
– Deep learning predicts lab test outcomes — no physical trials needed.
It’s like running a virtual flu season inside a computer — and doing it months before the first cough echoes in the clinic.
What’s Next? Beyond the Flu
Right now, VaxSeer focuses on the HA protein. But future versions could include:
– Neuraminidase (NA) — another key flu surface protein
– Immune history — how prior infections shape protection
– Manufacturing realities — can this strain even be produced at scale?
And the vision extends far beyond influenza.
As Dr. Jon Stokes of McMaster University puts it:
The implications go far beyond flu. Imagine predicting how antibiotic-resistant bacteria or drug-resistant cancers evolve. This is a new era of proactive medicine.
The team is already working on low-data prediction methods, aiming to apply VaxSeer-like models to viruses with sparse surveillance — opening doors for global pandemic preparedness.
Final Thought: From Reactive to Proactive Health
For decades, we’ve played defense against viruses. We wait, we react, we scramble.
VaxSeer flips the script.
It turns vaccine development from a high-stakes guess into a data-driven forecast — powered by AI, grounded in science, and built for the future.
As Regina Barzilay, MIT’s AI and Health pioneer, says:
Given the speed of viral evolution, current therapeutic development often lags behind. VaxSeer is our attempt to catch up.
And if the results are any indication; we’re starting to win the race.