Inferring pandemic growth rates from sequence data

de Silva, Eric; Ferguson, Neil M.; Fraser, Christophe

Real and simulated HA sequences were first aligned using CLUSTALX [26] and the multiple alignments visually confirmed. The aligned sequences were then imported into BEAST together with the dates each sample was collected. Given the small temporal extent over which samples were collected, we used a nucleotide substitution model of Hasegawa et al. [27] (which subsumes the two-parameter Kimura model used to generate simulated data) to avoid over-parametrization. Mutation rate heterogeneity among aligned sites was described by a gamma distribution. BEAST then uses the sequence alignment to construct a Jukes -Cantor distance matrix from which a starting unweighted pair-group method with arithmetic mean tree is made. A relaxed clock molecular model [28, 29] where branch rates are drawn from an underlying lognormal distribution is assumed. For two parameter combinations, we verified that BEAST converged to the same equilibrium distribution of parameters for a range of randomly starting trees (analysis not shown).