Performance of shotgun metagenomic sequencing for detection of fungi and parasites across clinical sample types: a multicenter retrospective study.
To evaluate the diagnostic performance of clinical shotgun metagenomic sequencing (SMg) for detecting medically relevant fungi and parasites compared with standard of care (SoC), and to define read-based thresholds for interpretation, we retrospectively analyzed 198 clinical samples from 187 patients across four university hospitals (2018-2022): blood (n=37), faeces (n=63), respiratory fluids (n=54), other biological fluids (n=24), and tissue biopsies (n=20). Total nucleic acids were sequenced ([≥]10 million reads per library) and processed with MetaMIC v2.2.1. Data were normalized as reads per million (RPM). Receiver operating characteristic analyses were used to derive optimal RPM thresholds by sample type. SoC identified microorganisms in 152/198 samples (76.8%). All 46 SoC-negative samples were also negative by SMg. At the genus level, SoC identified 187 taxa and SMg 175. Of these, 147 (78.6%) were detected by both methods, 40 (21.4%) by SoC only, and 28 (14.9%) by SMg only. The overall genus-level F1-score was 0.84. Quantification cycle (Cq) values (n=57) correlated inversely with RPM (p<0.001), and no false negatives occurred with Cq<28.6. Optimal thresholds were 0.06 RPM for faeces (AUC 0.89), 0.07 for respiratory fluids (AUC 0.93; sensitivity 88.9%, specificity 90.7%), 0.09 for blood (AUC 0.99), 0.19 for other fluids (AUC 0.94), and 0.57 for biopsies (AUC 0.89). A global threshold of 0.06 RPM yielded an AUC of 0.92 (sensitivity 88.9%, specificity 88.5%). A pragmatic uniform 0.1 RPM threshold maintained performance, while sample-type specific thresholds further improved accuracy, supporting standardized implementation of clinical metagenomics for fungal and parasitic diagnostics.