Inferring pandemic growth rates from sequence data

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

This then provides an explanation for why the BSPs flatten at later times: for an exponentially growing viral population (as is the case early on in an epidemic and in the simulated datasets we have been exploring) the phylogenetic tree is 'star-like'. That is to say, it has longterminal branches as in the tree in figure 3. This means that the final coalescent event will occur quite some time before the final sampling event (in figure 3a , there are 11 sampling events after the last coalescent event: gen23-gen33). From this period onwards, there are no new lineages (see LTT) and therefore no change in the effective population size is imputed. This issue (a form of left censoring in the 'backwards in time' coalescent context) represents a limitation to how far forwards in real-time the BSP can confidently be used to estimate N e . In light of this, we suggest that the BSP could be truncated at the time of the last coalescent event, rather than at the time of the last sample collection event.