AI Takes on the Flu: How MIT’s VaxSeer Could Revolutionize Vaccine Predictions

Every year, health officials around the world take on the daunting task of selecting the right flu strains for the subsequent season’s vaccine. This high-stakes decision is made months in advance, often involving educated guesswork about the dominance of certain strains. The repercussions of inaccurate predictions are significant, not only leading to a rise in illness but also burdening healthcare systems.

The flu’s unpredictability is nothing new, but the COVID-19 pandemic has truly amplified the challenges posed by rapid viral evolution. Like the SARS-CoV-2 variants that sprouted worldwide, the influenza virus is continuously mutating, making it tricky to contain. Thankfully, scientific advancements are making strides in this uphill race against mutating pathogens. The researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and the MIT Abdul Latif Jameel Clinic for Machine Learning in Health have designed an AI tool named VaxSeer to outsmart the flu’s relentless evolution.

VaxSeer is like a sophisticated crystal ball forecasting both dominant flu strains and effective vaccine preparations before the flu season starts. Its secret weapon is deep learning models trained on decades of virus genetic sequences and lab test data. These models predict virus evolution and assess potential vaccine effectiveness against future strains.

Unlike traditional models considering singular amino acid mutations, VaxSeer taps into a large protein language model to comprehend the complex interactions between multiple mutations. This approach delivers a more accurate picture of viral dominance changes, making it perfectly suited to deal with rapidly evolving viruses like influenza.

VaxSeer’s forecasting power lies in two main components. One projects the likelihood of a flu strain becoming dominant, and the second appraises how well a vaccine can neutralize that strain — a concept known as antigenicity. These predictions are woven into a ‘predicted coverage score’ that illustrates how well the vaccine matches circulating strains. The closer the score is to zero, the more compatible the vaccine is.

But does VaxSeer really work? A 10-year retrospective study comparing VaxSeer’s predictions with those from the World Health Organization (WHO) indicates a promising outlook. For two main flu subtypes, A/H3N2 and A/H1N1, VaxSeer outperformed or matched WHO recommendations in most seasons. Moreover, VaxSeer’s predictions aligned closely with real-world vaccine effectiveness data from various health departments.

In creating its predictions, VaxSeer operates in a unique way. It first estimates how swiftly a virus strain will spread and then simulates viral competition after calculating dominance. After cruising through this mathematical process, the model estimates the vaccine strain’s effectiveness using a standard lab test known as hemagglutination inhibition (HI) assay, serving as a widely accepted stand-in for measuring vaccine effectiveness.

Future plans for VaxSeer involve widening its focus beyond the influenza virus’s main target protein (or hemagglutinin – HA). Researchers hope to include additional viral proteins, immune history, vaccine manufacturing limitations, and dosage strategies. All this expansion, though, would require extensive datasets, which are not always easy to come by. Nonetheless, the team is hopeful about finding ways to predict viral evolution even in data-scarce environments.

VaxSeer may also have broader implications beyond influenza. Leading researchers envisage it as a game-changer in anticipating how antibiotic-resistant bacteria or drug-resistant cancers might evolve. The idea of forecasting disease progression could substantially shift our approach towards curing diseases. Though this technology is in its early stages, its future applications can extend our understanding of disease management and prevention.

This groundbreaking study was published in Nature Medicine and achieved its momentum with support from the U.S. Defense Threat Reduction Agency and the MIT Jameel Clinic. Now, only time will tell how this innovation shapes our fight against rapidly evolving viruses.

Want to read more about it? Visit the original article on MIT News: https://news.mit.edu/2025/vaxseer-ai-tool-to-improve-flu-vaccine-strain-selection-0828