Short- and long-read hybrid-capture viral metagenomics reach high sensitivity and specificity with data-driven cut-offs compared with routine diagnostics in CSF and BAL specimens

Background Hybrid-capture viral metagenomics improves sensitivity over untargeted metagenomics, but its performance across sequencing platforms and the definition of data-driven analysis cut-offs using real clinical samples have not been systematically evaluated. This study benchmarks the hybrid-capture workflow on mock and clinical specimens across short- and long-read platforms while introducing a new, reproducible approach for setting pathogen-calling cut-offs. Methods The Twist Comprehensive Viral Research Panel was used with both AVITI short-read and Oxford Nanopore Technologies (ONT) long-read sequencing on 12 mock cerebrospinal fluid (CSF) external quality assessment (EQA) panel samples and on 14 CSF and 21 bronchoalveolar lavage (BAL) clinical specimens. Data-driven cut-offs for pathogen calling were derived using Latin hypercube sampling across genome coverage, abundance, and bleed-ratio parameters, then validated on the independent EQA datasets. Results After cut-off application, both platforms detected 22 of 23 expected viral targets in EQA samples (95.7% detection-level sensitivity) with 100% specificity in negative controls. In clinical specimens, AVITI corroborated 26 of 30 (86.7%) and ONT 25 of 30 (83.3%) routine viral detections while maintaining 100% diagnostic specificity. Sensitivity was 100% for CSF but lower in BAL, particularly for low viral load targets, and all detections below a Cq value of 32 were identified by both platforms. Metagenomics also identified a potentially clinically relevant pathogen not tested by routine testing in two CSF samples. Conclusion Hybrid-capture viral metagenomics performed comparably on AVITI and ONT for routine viral targets, with AVITI showing slightly higher genome recovery. In addition, the newly developed data-driven cut-off selection strategy offers a practical, reproducible method other laboratories can apply to their own workflows.