Limited evidence of AI superiority in seasonal influenza vaccine strain selection

Preview:

"Ensuring seasonal influenza vaccine strains antigenically match circulating viruses is a key challenge for influenza vaccine development. Shi et al. recently introduced VaxSeer1, a novel artificial intelligence (AI) tool for seasonal influenza vaccine strain selection, reporting that their model retrospectively selects vaccine strains with better antigenic match than the World Health Organization (WHO) and that their central metric of vaccine match, the ‘coverage score,’ correlates with real-world vaccine effectiveness (VE) and vaccine-averted disease burden. While we recognize the potential value of AI for predicting virus evolution, we believe the evidence presented does not support the AI tool’s claimed utility for public health decision-making. First, we are concerned that the correlations underlying the claim that VaxSeer’s coverage score is a meaningful surrogate for VE are largely artifacts of statistical confounding. Second, we demonstrate that VaxSeer does not outperform a simple, non-AI baseline model at the key retrospective vaccine strain selection task."

de Jong, S.P.J. and Russell, C.A. (2026). Limited evidence of AI superiority in seasonal influenza vaccine strain selection. Nature Medicine.