Speaker
Description
The COVID-19 pandemic has seen an order of magnitude increase in numbers of pathogen genomes sequenced. The increased availability of genomes also raised the importance of genomic data in public health policy decisions. One key use is in phylodynamic inferences, which allows estimating epidemiologically relevant parameters, such as the effective reproduction number, directly from genomic data. Although individual sequenced samples are more informative than individual case reports, the proportion of confirmed cases sequenced is still low, with most countries sequencing <5%. It stands to reason that models that can draw on both sequenced and unsequenced samples to infer epidemiological parameters would be more powerful. However, existing exact methods are currently restricted to small datasets and of limited use in real-world scenarios.
We developed the time series integration method through approximation of moments (Timtam) to overcome this limitation by using a negative binomial approximation to the number of lineages in the transmission tree not represented in the reconstructed phylogeny. The approximation agrees well with exact methods and scales linearly in the number of datapoints. Timtam is available as a BEAST2 package, making it easily accessible and straightforward to combine with a wide range of model components. The method also supports piecewise-constant rates (allowing to infer changes in the effective reproduction number) and inferring historical prevalence estimates.
Using the emergence of the Alpha variant of concern in the United Kingdom as a case study, we show that Timtam is practical to use on even very large datasets. By subsampling this dataset we investigate how much can be gained by increased surveillance, and how the proportion of sequenced to unsequenced cases affect estimates. Finally, we examine the trade-off between accuracy and computational efficiency from aggregating samples into a time series.
