VirusWarn: A mutation-based early warning system to prioritize concerning SARS-CoV-2 and influenza virus variants from sequencing data

Abstract

The rapid evolution of respiratory viruses is characterized by the emergence of variants with concerning phenotypes that are efficient in antibody escape or show high transmissibility. This necessitates timely identification of such variants by surveillance networks to assist public health interventions. Here, we introduce VirusWarn, a comprehensive system designed for detecting, prioritizing, and warning of emerging virus variants from large genomic datasets. VirusWarn uses both manually-curated rules and machine-learning (ML) classifiers to generate and rank pathogen sequences based on mutations of concern and regions of interest. Validation results for SARS-CoV-2 showed that VirusWarn successfully identifies variants of concern in both assessments, with manual- and ML-derived criteria from positive selection analyses. Although initially developed for SARS-CoV-2, VirusWarn was adapted to Influenza viruses and their dynamics, and provides a robust performance, integrating a scheme that accounts for fixed mutations from past seasons. HTML reports provide detailed results with searchable tables and visualizations, including mutation plots and heatmaps. Because VirusWarn is written in Nextflow, it can be easily adapted to other pathogens, demonstrating its flexibility and scalability for genomic surveillance efforts.

Keywords: Genomic surveillance; Influenza virus; SARS-CoV-2; Variant prioritization; Warning system.

Graphical abstract

Fig. 1

(a) VirusWarn steps: assembled genome sequences and metadata (sampling dates, geographical locations) are used as query samples, which are pre-processed to infer mutation profiles for the monitored proteins. This is achieved either by aligning the sequences to a reference genome (e.g., in covSonar [27]) or via direct database queries (e.g., in Nextclade [28]). The mutation profiles are then ranked according to their concern level and results for query samples are summarized in interactive HTML reports using Jinja2 or RMarkdown . (b) Frequency heatmaps show mutation frequencies over time (calendar weeks on the x axis). A histogram showing the number of sequences per calendar week is shown at the bottom.

Fig. 2

Steps during annotation and ranking by VirusWarn. (a) Selected proteins to monitor (blue) for SARS-CoV-2 and Influenza. (b) Example annotations and genomic variation for a monitored protein. Annotations used for scoring: Mutation of Concern (MOC, yellow), Region of Interest (ROI, green), and Private Mutation (PM, blue). Genomic variation types (pink): Substitution, Deletion, or Insertion. (c) Example mutation profile: counts of variation types in the different annotation groups (MOC, ROI, PM). (d) Ranking to alert color levels: high impact (pink/red), medium impact (orange), low impact (yellow), and no alert (grey).

Fig. 3

SARS-CoV-2 partial spike protein (the first 650 amino acids, incl. receptor-binding sites) The MOC, ROI, and PM color annotations appear as in Fig. 2), and are contrasted to sites under positive selection (PS, blue bars on top) and lineage-defining mutations for selected VOCs from the ECDC table (red bars).