Fig. 5: Dissecting prediction of beth-1 by consensus strain.

a Full-length protein. b Epitope. The bar-plots display average genetic mismatch (amino acids, AA) over seven seasons in 2012/13-2018/19 for pH1N1 and 17 seasons in 2002/03-2018/19 for H3N2 (n = 70 region × season strata for each HA and NA of pH1N1, n = 144 and n = 146 available strata for HA and NA of H3N2, respectively). Panel (b) left Y-axis: average genetic mismatch of pH1N1; right Y-axis: average genetic mismatch of H3N2. We dissect prediction performance of the beth-1 by showing its accuracy achieved at multiple steps. The beth-1 (single protein) is compared to the future-consensus (consensus strain of the predicted future by the beth-1) and the current-consensus (consensus strain of the current virus population). Two-sided p-value is calculated using paired t-test on log mismatch between two methods matched by region and season. The p-value of LBI versus beth-1(single protein) are 6.7e–10, 1.5e–12, 1.7e–6, 7.9e–5, respectively (a), and 9.3e–6, 1.1e–8, 6.7e–5, 4.8e–4, respectively (b). The p-value of current consensus versus beth-1 future consensus are 0.061, 6.1e–5, 0.005, 0.011, respectively (a), and 1, 0.001, 0.009, 0.016, respectively (b). The future-consensus generally advances prediction of the current-consensus on the genomic segments, while beth-1 (single protein) gives slightly higher mismatch compared to the future-consensus. The LBI is displayed to replicate the previous finding involving the current-consensus strain²⁸. Error bar: standard deviation of the average genetic mismatch by region and season. ^*: p-value < 0.05; ^**: p-value < 0.01; ^***: p-value < 0.001.