Identification of Low- and High-Impact Hemagglutinin Amino Acid Substitutions That Drive Antigenic Drift of Influenza A(H1N1) Viruses
Fig 5
Sequence-based prediction of antigenic phenotype.
Observed and predicted HI titers plotted on log2 scale (antigenic units) using representative models trained with data for 90% of the pairs of virus and antiserum. Predictive models contained terms for A) Average titers for each reference virus, B) Antigenic cluster-defining substitutions ΔK130 and K141E, C) All 18 antigenic substitution(s) shown in Table 1, D) All 18 antigenic substitution(s) shown in Table 1 with additional terms that estimate differences in test virus receptor-binding avidity (non-antigenic variation in titer associated with each virus). Each model was fitted to the same training dataset comprising 90% of all pairs of virus and antiserum and predictions for the remaining data are shown. Incremental improvements in mean absolute prediction error are shown alongside SEM and 95% upper limits in S5 Table.