MIT researchers have developed cutting-edge AI tool that could revolutionize selection of the most accurate virus strain for flu vaccines. The system, called VaxSeer, uses machine learning to predict the evolution of the virus. And its antigenic properties so that vaccine selection can be based on science rather than guesswork.
Every year, health experts around the world have to decide which virus strains to include in the flu vaccine for the coming winter. This decision has to be made months in advance, when the prediction becomes almost a race. If the selected strains are correct, the vaccine remains effective. Otherwise, the public is less protected and the strain on the health system increases.
VaxSee uses deep learning models based on decades of viral genetic data. And laboratory test results to map the likely evolution of the virus and the vaccine’s ability to respond. Traditional models look at each genetic change in isolation. But this new model also considers the cumulative effects of those changes, making it more suitable for rapidly changing viruses. Researchers found in a 10-year analysis that the VaxSear recommendations were more effective than the
World Health Organization (WHO) recommendations in many seasons. Specifically, the model outperformed the A/H3N2 virus in nine out of nine years, while it outperformed or equaled A/H1N1 in six seasons. At one point, the model selected a strain for 2016 that the WHO included the following year.
The system consists of two primary prediction engines—one that estimates the spread of the virus and the other that measures how well the vaccine will inactivate the virus. Together, the two provide a preliminary “coverage score” that indicates the vaccine’s potential efficacy.
Experts say the technology will allow public health officials to make faster and better decisions, keeping the world one step ahead in the race to contain infections and prevent them. In the future, the use of this model for treatment of other viruses is possible, antibiotic-resistant bacteria and even cancer, where diseases mutate rapidly.
“Current treatments are always lagging behind due to the rapid pace of viral evolution. VaxSear is our attempt to stay one step ahead,” said Regina Barzley, an MIT professor and co-author of the study.
