Sukhum, Kimberley V.

Abstract Accurate taxonomic profiling of human microbiomes is essential for advancing research and understanding the complex role microbial communities play in human health. When using shotgun metagenomics, the sequencing data is analyzed through metagenomic pipelines, which incorporate various open-source tools and classify microbes based on matched paired-end DNA reads. However, differences in sequencing and computational approaches can produce substantially different microbiome profiles from the same sample, making validation critical. One approach for validation is benchmarking with realistic mock communities, but this remains relatively rare. Additionally, existing benchmarks often overlook microbiome variability across life stages and body sites, limiting their clinical and research utility. Here, we developed age- and body site-stratified synthetic metagenomes, enabling context-aware benchmarking of microbiome pipelines. Using novelty-based sampling to prioritize microbial diversity and minimize redundancy among selected samples, we selected 300 representative, real biological samples spanning six categories: adult, child, toddler, and infant (>6 months and <6 months) gut samples, as well as adult vaginal samples. We validated three pipelines, Tiny Health’s proprietary Metagenomic Classifier v2 (THMCv2), MetaPhlAn4, and Kraken2+Bracken, using precision, recall, F1 score, and area under the precision-recall curve (AUPR) across age groups and sample types. THMCv2 demonstrated higher recall and F1 scores, detecting more taxa across sample types and ages, while MetaPhlAn4 achieved the highest precision. THMCv2 also achieved the highest area under the precision-recall curve, reflecting peak performance across both abundant and rare species. When analyses were weighted by abundance, THMCv2 and MetaPhlAn4 each characterized the mock community nearly perfectly. Errors for THMCv2 were largely restricted to very low-abundance taxa (<0.001%), whereas MetaPhlAn4 occasionally produced false positives for higher-abundance taxa. Species-level analyses of clinically relevant microbes confirmed these patterns, with THMCv2 demonstrating higher sensitivity, MetaPhlAn4 higher specificity, and Kraken2 lower overall performance. These results demonstrate clear precision-recall trade-offs in metagenomic profiling. This benchmarking framework provides a reproducible approach for evaluating pipeline performance across diverse microbiome contexts and life stages.

Introduction Eczema and food allergy commonly emerge during infancy and are linked to changes in the gut microbiome, yet it remains unclear when microbiome differences associated with allergic disease first appear during development. Methods We analyzed age-stratified shotgun metagenomic data from 97 children aged 4–36 months, including physician-confirmed cases of eczema or food allergy and non-allergic controls, excluding recent antibiotic or probiotic exposure. Microbial taxa, functional pathways, and composite microbiome metrics were evaluated across three developmental stages: early infancy (4–6 months), mid-infancy (6–12 months), and toddlerhood (12–36 months). Results Differences between allergic and non-allergic children were minimal before 6 months of age but became more apparent during mid-infancy and persisted into toddlerhood. Allergic conditions were associated with reduced abundance of fiber-fermenting and butyrate-producing taxa, enrichment of facultative and inflammation-associated microbes, lower microbiome maturation scores, and shifts in metabolic and inflammatory functional capacity. Discussion These findings suggest that gut microbiome divergence associated with allergic disease becomes more apparent during mid-infancy, highlighting a developmentally relevant period for understanding early immune disruption. The results support further longitudinal and interventional studies aimed at clarifying whether earlier microbiome-targeted strategies may help modify progression along the atopic march.