Inferring the effects of mutations on SARS-CoV-2 transmission from genomic surveillance data

Pathogens can acquire mutations that affect how easily they are transmitted to others. Rapidly identifying more transmissible variants could aid in public health decisions. However, it is challenging to disentangle the effects of individual mutations from complex data, and the popular phylogenetic approaches for analyzing viral sequences are computationally intractable for large data sets. Here we describe a Bayesian inference method to infer the effects of mutations on viral transmission. Our model extends the standard susceptible-infected-recovered (SIR) epidemiological model by including features such as superspreading and the spread of infection through travel, which are relevant for SARS-CoV-2. Using a path integral approach originally applied in population genetics [1], we obtain analytical estimates for the transmission effects of mutations that best explain genomic surveillance data. Our analysis reveals hotspots for transmission-affecting mutations throughout the SARS-CoV-2 genome, highlighting well-known Spike mutations as well as less-studied mutations in other proteins. Crucially, we also show that our model is capable of rapidly detecting new variants with enhanced transmissibility, which we demonstrate through an analysis of the rise of the Alpha variant in the UK.