Optimizing Pathogen Detection in Clinical Metagenomics: A Comprehensive Benchmarking of Bioinformatic Pipelines for Single-End Short Reads.

Abstract

Metagenomic next-generation sequencing for identifying pathogens has been increasingly utilized as a diagnostic tool. Prior research mostly emphasizes the selection of classification software based on reads above 100bp, neglecting the influence of sequencing data pre- and post-processing stages. Independently assessment was performed assessed the efficacy and effects of tools for each phase, encompassing quality control, host read elimination, taxonomic classification, and false positive removal, utilized in a metagenomics analysis pipeline employing single-end 50bp reads derived from sequencing 13 positive-control mock community samples. The efficacy of tool combinations was assessed using single-end 50 bp data from identical mock samples processed using different wet-lab methodologies. The findings indicate that the selection of quality control and host-removal techniques is essential, significantly influencing the extraction of microbial reads from clinical metagenomic sequencing data and resulting in elevated alpha diversity. While false-positive filtering enhances precision for trimming tool-based combinations without impacting recall, its effect is limited for combinations utilizing the bt2 aligner. The tool combination pipelines integrated with various false positive methodologies for four microbial categories: no filtering for bacteria, bracken for fungi, relative abundance (>0.1%) for DNA viruses, and z-score (z > 3) for RNA virus identification, as implemented in the prior R Shiny application for forthcoming benchmarking. This study highlights the critical role of preprocessing and classifier selection in accurate metagenomic pathogen identification.