Learning and forecasting selection dynamics of SARS-CoV-2 variants from wastewater sequencing data using Covvfit

The COVID-19 pandemic has been driven by the emergence and spread of SARS-CoV-2 variants that confer a selective advantage over previously circulating strains. Estimating these selective advantages typically involves analyzing a large number of positive test samples through genomic sequencing. In this study, we present Covvfit, a statistical model and software package for estimating the fitness advantages of multiple competing variants using sequencing data derived from wastewater samples from different locations. We use our model to reconstruct the dynamics of variant competition across successive waves of the pandemic using over 5,000 samples from wastewater sequencing data collected between 2021 and 2025. We show through a comparison with clinical data that wastewater-based estimates of fitness advantages are efficient and accurate. Furthermore, we demonstrate that once variants surpass a low detection threshold, Covvfit can accurately predict their future dynamics over long prediction horizons.