Hill numbers at the edge of a pandemic: rapid SARS-COV2 surveillance using clinical, pooled, or wastewater sequence as a sensor for population change

Abstract

The COVID-19 pandemic has highlighted the critical role of genomic surveillance for guiding policy and control strategies. Timeliness is key, but rapid deployment of existing surveillance is difficult because current approaches are based in sequence alignment and phylogeny. Millions of SARS-CoV-2 genomes have been assembled, the largest collection of sequence data in history. Phylogenetic methods are ill equipped to handle this sheer scale. We introduce a pan-genomic measure that examines the information diversity of a k-mer library drawn from a country's complete set of sequenced genomes. Quantifying diversity is central to ecology. Studies that measure the diversity of various environments increasingly use the concept of Hill numbers, or the effective number of species in a sample, to provide a simple metric for comparing species diversity across environments. The more diverse the sample, the higher the Hill number. We adopt this ecological approach and consider each k-mer an individual and each genome a transect in the pan-genome of the species. Applying Hill numbers in this way allows us to summarize the temporal trajectory of pandemic variants by collapsing each day's assemblies into genomic equivalents. We do this quickly, without alignment or trees, using modern genome sketching techniques to accommodate millions of genomes in one condensed view of pandemic dynamics. Using data from the UK, USA, and South Africa, we trace the ascendence of new variants of concern as they emerge in local populations. This history of emerging variants uses all available data as it is sequenced, intimating variant sweeps to dominance or declines to extinction at the leading edge of the COVID19 pandemic.

What do we gain when tolerating loss? The information bottleneck wrings out recombination

“…With respect to our initial random conditions, this algorithm is guaranteed to converge to a local optimum. We mitigate the risk of getting trapped in local optima by testing several random initializations, and we reduce the overall computation by implementing a k-mer sketch 27,28 .…”