The flu virus evolves rapidly, making it hard to predict which strains will dominate in a given season. Twice a year, the World Health Organization (WHO) convenes global experts to recommend strains for upcoming vaccines. A new AI tool VaxSeer introduced by MIT researchers might reduce the uncertainty faced by vaccine planners and manufacturers.
How VaxSeer works
VaxSeer uses machine learning to anticipate how influenza viruses mutate and how immune systems might respond. Researchers said the system could help public health agencies and pharmaceutical companies make better-informed choices months in advance, critical timing for vaccine development, and could improve match rates, which often fall between 40% and 60% effectiveness in average seasons.
VaxSeer aims to enhance this process using a two-pronged machine learning model: one component predicts which viral strains are most likely to become dominant, while the other estimates their antigenic similarity, or how well immune systems would recognize them. The model then computes a “coverage score,” which is a forward-looking metric that reflects how effectively a vaccine formulation is expected to perform against future virus populations.
Accuracy of VaxSeer
When tested against a decade of influenza data, MIT researchers found that VaxSeer more accurately predicted dominant H3N2 strains in nine of 10 seasons. For H1N1, its recommendations either equaled or surpassed WHO’s choices in most years. One illustration came in 2016, when the AI tool flagged a strain that global health officials only added to the vaccine lineup the following year.
Implications for vaccine development
VaxSeer’s predictive accuracy could boost vaccine effectiveness by improving strain match, a key driver of public health outcomes. Better forecasts mean vaccine manufacturers could start production earlier with greater confidence, easing supply chain logistics and reducing risks of poor match.
Additionally, the framework isn’t limited to influenza. MIT researchers noted that the same modeling approach could be adapted for other fast-mutating viruses, such as coronaviruses.
Challenges ahead
While VaxSeer shows promise in retrospective evaluations, integrating it into global public health processes poses challenges.
The WHO-driven selection process involves international consensus, regulatory oversight, and proven transparency. MIT researchers acknowledged that rigorous validation, peer-reviewed publication, and trust-building would be necessary for adoption.
Still, the project reflects a broader trend: Computational biology and artificial intelligence are increasingly supporting high-stakes decisions in public health, from diagnostics to drug discovery.