Zebra

Zebra filters false taxonomic hits in shotgun metagenomic data by using genome cover and saturated genome cover metrics to distinguish genuine microbial taxa from ambiguous read mappings against large reference collections such as the Web of Life (WoL) database.

Key Features:

• Ambiguous read filtering: Filters false taxonomic hits arising from reads that overlap multiple reference genomes in shotgun sequencing, addressing expansion of the Web of Life (WoL) database with over 10,575 reference genomes.
• Genome Coverage Metric: Defines "genome cover" as the fraction of a reference genome overlapped by reads to help distinguish true microbial presence from overlap artifacts.
• Dynamic Prediction Model: Predicts genome cover from read count to differentiate genuine taxa from false positives and was validated using a Staphylococcus aureus monoculture.
• Saturated Genome Cover: Computes "saturated genome cover" as the actual fraction of a reference genome overlapped by sample contents to represent low-abundance or low-prevalence bacteria across samples.
• Thresholding for Spurious Hits: Applies thresholds on saturated genome cover to identify spurious reference hits or distant relatives.
• Composite Metric Application: Composites genome cover metrics across similar samples to achieve saturation for rare species and improve reproducibility of taxonomic assignments.

Scientific Applications:

• Microbiome taxonomic assignment: Improves accuracy of taxonomic assignment in shotgun metagenomic microbiome studies by reducing ambiguous and spurious hits.
• Interpretation of microbial presence: Enables more biologically plausible interpretation of microbial presence and associations with environmental or clinical conditions by filtering false positives.
• Pathogen artifact mitigation: Mitigates overemphasis on implausible artifact hits (for example, anthrax or bubonic plague) by focusing on nearest sample relatives and appropriate cover thresholds.

Methodology:

Calculates genome cover as the fraction of a reference genome overlapped by reads, predicts genome cover from read counts, computes saturated genome cover, applies thresholds on saturated cover to flag spurious hits, and composites cover metrics across similar samples; the dynamic model was validated on a Staphylococcus aureus monoculture.