Inferring viral variants and fitness advantages from longitudinal wastewater sequencing data

During the SARS-CoV-2 pandemic, wastewater-based epidemiology emerged as an effective tool for monitoring infection dynamics. However, wastewater represents a pooled sample of incomplete viral genomic fragments originating from multiple individuals in the catchment area of a wastewater treatment plant, preventing the assembly of complete viral genomes. Existing computational pipelines are therefore limited in detecting new variants that are absent from clinical sequencing data. We developed FitNMF, an integrated viral dynamics and non-negative matrix factorization model that leverages the temporal structure of wastewater sequencing data and the underlying variant competition to jointly infer mutational signatures, relative abundances, and fitness advantages of circulating variants directly from wastewater next-generation sequencing data. Using simulated data as well as wastewater time-series data collected between 2021 and 2024 across six Swiss wastewater treatment plants, we demonstrate FitNMF's ability to identify and quantify emerging variants de novo, offering a powerful tool for real-time variant surveillance in settings with limited clinical sequencing.