Identification and masking of artifactual and misleading within-host variants in deep-sequencing SARS-CoV-2 data

Deep-sequencing data are increasingly used to study within-host viral diversity and to inform evolutionary inference. For SARS-CoV-2, analyses based on intra-host single-nucleotide variants (iSNVs) have been widely applied to quantify within-host diversity and infer transmission dynamics. However, these applications critically depend on the reliable identification of low-frequency variants, which remain vulnerable to systematic and technical artifacts. In this study, we show that recurrent artifactual iSNVs are common in large-scale SARS-CoV-2 sequencing data and can persist even under conservative minor allele frequency thresholds. Using data from the UK's Office for National Statistics COVID-19 Infection Survey, we demonstrate that such artifacts are predominantly sequencing center-specific rather than primer-specific. Each center exhibits a modest, distinct set of recurrent artifactual variants showing little overlap with sites routinely masked at the consensus level. To address this, we developed a systematic, dataset-aware framework that uses recurrence within sequencing datasets to identify small, noise-adapted sets of artifactual iSNVs to mask. Applying this framework reduces spurious sharing of low-frequency variants between samples and qualitatively alters downstream inferences, including estimates of within-host diversity and transmission bottleneck sizes. Although this study focused on SARS-CoV-2, it is likely that recurrent artifactual iSNVs will be problematic for other viruses as mass-sequencing becomes increasingly routine. Together, these findings highlight the importance of explicit, dataset-aware artifact control for robust inference from within-host variation, particularly as genomic studies increasingly seek to exploit sub-consensus diversity in rapidly evolving pathogens.