Sequence-structured fitness landscapes predict influenza evolution

A generative, mutation-level fitness landscape inferred from temporal A/H3N2 and A/H1N1 hemagglutinin and neuraminidase data predicts fixation/loss and future population composition, outperforming neutral and multinomial-logistic models.

• Why it matters: Genomic surveillance samples only a small fraction of sequence space, making it hard to learn the fitness landscape that drives seasonal influenza evolution and to forecast which variants will spread.
• What they did: The framework infers effective mutation-level contributions to viral fitness from temporal sequence data for hemagglutinin and neuraminidase in A/H3N2 and A/H1N1, then runs forward simulations from the inferred landscapes.
• The result: The inferred landscapes recover known antigenicity determinants and show convergent spatial organization across subtypes consistent with common antibody-mediated selection, enabling more accurate prospective evolutionary forecasting than neutral evolution or multinomial-logistic growth.