Table 1.
The site-specific amino-acid preferences measured in the new experiments offer an improved description of HA evolution in nature. Aikake information criterion (AIC) [26] was used compare the maximum likelihood phylogenetic fit of several models to an alignment of seasonal human H1N1 and classical swine H1N1 HAs. The experimentally informed substitution models are of the form described in [24] with the data from the average of all three replicates of the new or old experiments, or the average of the two. These models are compared to the variants of the substitution model of Goldman et al. [27] denoted as M0 and M8 in Yang et al. [28] with the equilibrium codon frequencies estimated empirically using the F3X4 method. The best model is the one that combines all experimental data, but a model informed by the new experiments alone is better than one informed by the old experiments alone. To confirm that the experimentally informed models are superior because they are site specific, we fit a control model in which the experimental data is averaged across sites. The tree topology was fixed to that inferred by maximum likelihood using the M0 version of the Goldman–Yang model. The free parameters for each model were then optimized along with the branch lengths; optimized parameters are in the last column.
| Model | ΔAIC | Log Likelihood | Parameters (Optimized + Empirical): Optimized Values |
|---|---|---|---|
| new data + old data | 0.0 | −14933.5 | 6 (6 + 0): β = 1.82, ω = 0.51, κ = 4.95, = 0.40, = 0.18, = 0.20 |
| new data | 197.6 | −15032.3 | 6 (6 + 0): β = 1.80, ω = 0.46, κ = 5.06, = 0.40, = 0.18, = 0.20 |
| old data | 341.2 | −15104.1 | 6 (6 + 0): β = 1.40, ω = 0.46, κ = 4.90, = 0.39, = 0.18, = 0.20 |
| Goldman–Yang M8 | 2156.8 | −16003.9 | 14 (5 + 9): = 0.01, = 1.91, = 0.02, = 0.76, κ = 4.94 |
| new data + old data, averaged across sites | 2971.6 | −16419.3 | 6 (6 + 0): β = 0.50, ω = 0.20, κ = 5.38, = 0.38, = 0.18, = 0.21 |
| Goldman–Yang M0 | 2980.8 | −16418.9 | 11 (2 + 9): ω = 0.19, κ = 4.88 |