Table 2.

Short-read (Illumina) versus long-read (Oxford Nanopore) sequencing for the pediatric neuromuscular airway microbiome

Feature	Illumina short-read	Oxford Nanopore (ONT) long-read
Sensitivity and specificity across kingdoms (bacteriome, virome, archaeome, mycobiome)	Very high depth and per-base accuracy. Shotgun covers all kingdoms well. Genus-level specificity is strong, and 16S species-level calls can be ambiguous [24]	Full-length 16S from V1 to V9 and long shotgun reads improve species and sometimes strain assignment, and direct RNA sequencing aids the virome. Confident species calls need high-accuracy basecalling [26, 35]
Required microbial biomass	Deep sequencing with in silico host removal handles a bacterial fraction of about 1 to 5 percent [20, 33]	Low-input kits and adaptive sampling enrich targets, but very low biomass magnifies contamination and host DNA [27]
Detection of the rare biosphere	High depth favours detection of low-abundance taxa at scale [32]	Places detected rare taxa more precisely, but lower throughput per run can undersample them. Adaptive sampling can target specific taxa [27]
False-positive and false-negative taxon assignment	Low per-base error, but chimeras and multi-mapping cause misassignment, and contamination drives false positives in low biomass [36, 37]	Long reads reduce ambiguous mapping and chimeras, but higher per-read error can cause false calls without high-accuracy basecalling [35]
Subclonal and low-frequency variants	High depth and low error make this a short-read strength [38]	Higher per-read error limits low-frequency variant calling, and this is improving with duplex and Q20 chemistry [28, 29]
Virulence and antimicrobial resistance determinants	Detects genes but fragments the resistome and loses genomic context [38]	Resolves full-length resistance operons, plasmid versus chromosomal location, and links to mobile elements [38]
Replication activity (growing versus dormant)	Coverage-based replication-rate estimation is better established with deep short-read data [39, 40]	Possible with adequate coverage. Neither platform separates dormant cells without RNA-based measures [39, 40]
Sequence and structural variants	Excellent for single-nucleotide and small variants, weaker for large structural variants and repeats [35]	Excellent for structural variants, repeats, phasing and genome assembly [35, 41]
Network analysis of polymicrobial communities	High sample throughput and depth support robust co-occurrence networks	Lower throughput per run limits large-scale networks, but long reads add molecule-level linkage [35]
Time from sampling to report	Batch-oriented, usually next day at the earliest [24]	Real-time, with interim results during the run and about 6 h in optimized protocols. Point-of-care friendly [20, 24]

Comparison for the pediatric neuromuscular disease, low-biomass airway context. Platform performance evolves rapidly with chemistry and instrument version