Benchmarking next- versus third-generation sequencing in metagenomics: performance metrics and diagnostic efficacy

This study aimed to compare the analytical characteristics and diagnostic performance of short-read next-generation sequencing (NGS) and long-read third-generation sequencing (TGS) for metagenomic pathogen detection, using defined mock communities and clinical bronchoalveolar lavage fluid (BALF) samples. Mock evaluations included microbe-host gradient mixtures (D1/D2) and six complex microbial panels (M1-M6). Sequencing was performed on Illumina, MGI, and Oxford Nanopore Technologies (ONT) platforms. Clinical validation was conducted on 62 BALF samples. Diagnostic performance was assessed against culture, clinical microbiological tests (CMT), and a composite reference standard (CRS). Turnaround times for Illumina and MGI were approximately 18-20 h and 14-19 h, respectively, whereas the ONT workflow was completed within 4-6 h. The microbe-to-host DNA ratio significantly influenced sequencing performance. Depletion of host DNA notably enhanced ONT detection, reducing the false-negative rate for low-abundance microorganisms from 43.3% to 6.7%. For all mock samples, both the Illumina and MGI platforms demonstrated 100% sensitivity and showed highly concordant detection profiles. In clinical specimens, when evaluated against the composite reference standard, the positive percent agreement (PPA) values of NGS and TGS were 93.3% and 90.7%, respectively, with corresponding negative percent agreements (NPAs) of 77.6% and 83.3%. Both platforms identified numerous pathogens that were missed by culture, especially in polymicrobial infections. Among 22 CRS-defined polymicrobial samples, culture identified all pathogens in only 2 cases, whereas NGS and TGS achieved full pathogen recovery in 18 and 17 cases, respectively. Within the evaluated workflows, short-read sequencing showed slightly higher sensitivity and overall stability, whereas host-depleted ONT offered a substantial turnaround-time advantage and may serve as a useful complementary approach in complex or time-sensitive clinical scenarios. IMPORTANCE: Rapid and accurate identification of the microbes causing pneumonia is essential for choosing effective treatment, yet current diagnostic tests are slow and often miss important pathogens. We systematically compared two major DNA sequencing strategies-established short-read platforms and newer long-read nanopore sequencing-using both carefully designed mock communities and real bronchoalveolar lavage samples from patients. We show when removal of human DNA is essential, how mixed infections are best captured, and what trade-offs exist between speed and sensitivity. Our results provide practical guidance on how hospitals can implement sequencing-based diagnostics, when rapid nanopore testing can complement conventional short-read workflows, and how to interpret sequencing read counts in day-to-day clinical decision-making.