“`html
Each year, international health authorities confront a pivotal choice: Which influenza strains should be included in the forthcoming seasonal vaccine? This decision must be finalized months ahead of time, long before the flu season starts, often resembling a race against time. If the chosen strains align with those in circulation, the vaccine will probably be very effective. However, if the prediction is inaccurate, immunity can diminish significantly, resulting in (potentially preventable) sickness and stress on healthcare systems.
This issue became even more recognized among researchers during the Covid-19 pandemic years. Recall the moments, repeatedly, when new variants appeared just as vaccines were being administered. Influenza acts like a similar, unruly relative, constantly and unpredictably mutating. This makes it challenging to stay ahead, making it difficult to create vaccines that remain protective.
To mitigate this unpredictability, scientists at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and the MIT Abdul Latif Jameel Clinic for Machine Learning in Health aimed to enhance vaccine selection accuracy and reduce dependence on speculation. They developed an AI system called VaxSeer, intended to forecast dominant flu strains and pinpoint the most effective vaccine candidates well in advance. This tool utilizes deep learning models trained on decades of viral sequences and laboratory test findings to simulate potential flu virus evolution and forecast vaccine responses.
Conventional evolution models generally examine the impact of single amino acid mutations in isolation. “VaxSeer employs a comprehensive protein language model to understand the relationship between dominance and the combinatorial effects of mutations,” states Wenxian Shi, a PhD candidate in MIT’s Department of Electrical Engineering and Computer Science, a researcher at CSAIL, and the primary author of a new study on the topic. “In contrast to existing protein language models that presume a static distribution of viral variants, we identify dynamic dominance shifts, making it more suited for rapidly evolving viruses such as influenza.”
An open-access article on the research was released today in Nature Medicine.
The future of flu
VaxSeer comprises two primary prediction engines: one that assesses the likelihood of each viral strain spreading (dominance), and another that estimates how efficiently a vaccine will counter that strain (antigenicity). Together, they generate a predicted coverage score: a forward-looking estimate of how effectively a particular vaccine is anticipated to perform against future viruses.
The range of the score could extend from an infinite negative to 0. The nearer the score is to 0, the better the antigenic match of the vaccine strains to the circulating viruses. (You might think of it as the inverse of some kind of “distance.”)
In a decade-long retrospective analysis, the researchers compared VaxSeer’s recommendations to those made by the World Health Organization (WHO) for two significant flu subtypes: A/H3N2 and A/H1N1. For A/H3N2, VaxSeer’s selections exceeded the WHO’s in nine out of ten seasons, based on retrospective empirical coverage scores (a surrogate measure of vaccine effectiveness, derived from observed dominance in previous seasons and experimental HI test outcomes). The team utilized this information to review vaccine selections, as effectiveness data is only available for vaccines actually administered to the population.
For A/H1N1, it outperformed or matched the WHO in six out of ten seasons. In one noteworthy instance, during the 2016 flu season, VaxSeer identified a strain that the WHO did not select until the next year. The model’s forecasts also demonstrated strong correlations with real-world vaccine effectiveness estimates, as reported by the CDC, Canada’s Sentinel Practitioner Surveillance Network, and Europe’s I-MOVE program. VaxSeer’s predicted coverage scores closely matched public health statistics on flu-related illnesses and medical consultations averted through vaccination.
So how precisely does VaxSeer decipher all this data? Primarily, the model first estimates the speed at which a viral strain spreads over time using a protein language model, and then determines its dominance while considering competition among various strains.
Once the model has derived its insights, they are integrated into a mathematical framework based on ordinary differential equations to simulate viral spread over time. For antigenicity, the system evaluates how effectively a specific vaccine strain will perform in a common laboratory assessment called the hemagglutination inhibition assay. This test measures how efficiently antibodies can inhibit the virus from binding to human red blood cells, serving as a widely used proxy for antigenic match/antigenicity.
Outpacing evolution
“By modeling how viruses evolve and how vaccines interact with them, AI tools like VaxSeer could enable health officials to make improved, faster decisions — and stay one step ahead in the battle between infection and immunity,” states Shi.
Currently, VaxSeer focuses solely on the flu virus’s HA (hemagglutinin) protein, the primary antigen of influenza. Upcoming versions may incorporate other proteins like NA (neuraminidase), as well as factors such as immune history, production limitations, or dosage levels. Adapting the system for other viruses would also necessitate large, high-quality datasets that track both viral evolution and immune responses — information that is not always publicly accessible. Nevertheless, the team is actively developing methodologies that can predict viral evolution in data-scarce environments by leveraging relationships among viral families.
“Given the rapid pace of viral evolution, current therapeutic advancements frequently lag behind. VaxSeer is our endeavor to bridge that gap,” says Regina Barzilay, the School of Engineering’s Distinguished Professor for AI and Health at MIT, AI lead of the Jameel Clinic, and CSAIL principal investigator.
“This paper is noteworthy, but what excites me perhaps even more is the team’s ongoing efforts in predicting viral evolution in low-data conditions,” remarks Assistant Professor Jon Stokes of the Department of Biochemistry and Biomedical Sciences at McMaster University in Hamilton, Ontario. “The implications extend far beyond influenza. Envision being able to anticipate how antibiotic-resistant bacteria or drug-resistant cancers may evolve, both of which can change rapidly. This form of predictive modeling introduces a powerful new approach to understanding disease evolution, allowing us to remain one step ahead and design clinical interventions before evasion becomes a significant challenge.”
Shi and Barzilay collaborated on the paper with MIT CSAIL postdoc Jeremy Wohlwend ’16, MEng ’17, PhD ’25, and recent CSAIL affiliate Menghua Wu ’19, MEng ’20, PhD ’25. Their research was partially supported by the U.S. Defense Threat Reduction Agency and the MIT Jameel Clinic.
“`