Inferring pandemic growth rates from sequence data

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

The sharpness of the transition in the flattening of the LTT curve in figure 3b is striking and figure 3d illustrates the effect of sampling density on the shape of the LTT. Figure 3d ,e are the LTT and BSP, respectively, of the log-proportionally sampled R ¼ 1.5, k ¼ 1 simulated dataset in figure 2 using a piecewise-linear skyline model. The greater sampling density still results in a relatively sharp transition in the LTT, and the BSP starts to flatten at the same time as the LTT slope flattens and the period after which there are no new lineages corresponding to little growth on the BSP. Figure 3f ,g shows the LTT and BSP that result from randomly sampling respectively three and 10 sequences per generation (for generations with more than 100 sequences-see later). As expected, increasing the sampling density means the cumulative LTT slows down at later times corresponding to the last coalescent event and the time at which the BSP also slows down and/or flattens.