Corpusome

A harmonized cross-body-site human microbiome corpus for representation learning — 190,601 samples across 7 body sites and 2 sequencing modalities, integrating MGnify, curatedMetagenomicData, AGP, and disease-specific cohorts (Alzheimer's, lung cancer, T2D).

16S amplicon · 165,046 samples Shotgun metagenomics · 25,555 samples 782 studies · 42+ countries 7 body sites · Alzheimer · Lung Cancer · T2D
📄 bioRxiv Preprint 🔗 Zenodo DOI 💻 GitHub
—
Total Samples
16S + Shotgun combined
—
Studies / Projects
MGnify, cMD, AGP
—
Body Sites
stool · oral · skin · respiratory · urogenital · milk
—
Countries
Shotgun tier (cMD)
—
16S Taxa Tokens
Terminal genus, harmonized
—
Shotgun Species
MetaPhlAn3 species table
Corpus at a Glance
Body site 16S Shotgun Total
stool74,35421,03095,384
respiratory40,7409340,833
oral23,60685724,463
skin14,13150414,635
urogenital12,1279612,223
milk—88
Total 165,046 25,555 190,601
Samples by Body Site and Data Source

Stacked by contributing database

16S vs Shotgun per Body Site

Log scale — both modalities kept separate

Samples per Body Site (Total)
Health Status Breakdown

Shotgun tier only — 16S tier is entirely unlabeled

Note on taxonomic vocabularies: The MGnify full pull uses SILVA lineages; AGP uses Greengenes 13_8. The merged 16S genus table (corpus_16s_genus_relab_expanded.parquet) harmonizes both on the terminal genus token (e.g. g__Bacteroides). Do not merge the 16S and shotgun tiers naively — they use different profilers. Use the modality column as a mandatory split key.
Release History
v1.2 2026-09-25 +3,055 samples
Disease Cohort Integration
stool respiratory brain
  • +164 Ferreiro 2023 — preclinical AD gut shotgun (MetaPhlAn3, USA)
  • +88 AD brain tissue 16S — post-mortem brain microbiome (new body site)
  • +1,549 TCGA Poore 2020 — lung cancer WGS-derived microbial profiles (LUAD + LUSC)
  • +400 BALF mNGS — lung cancer vs. pulmonary infection (Kraken2/Bracken, China)
  • +576 Hartwig / Battaglia 2024 — metastatic lung WGS (Netherlands)
  • +134 MLRepo Qin 2012 — T2D gut shotgun (China cohort)
  • +144 MLRepo Karlsson 2013 — T2D / IGT / NGT gut shotgun (Sweden cohort)
Total: 190,601 samples · 7 body sites · 782 studies
v1.1 2026-07 +156,866 samples
MGnify Full Catalogue Pull
stool oral skin respiratory urogenital
  • +156,866 EBI MGnify — full human body-site catalogue pull (708 studies, SILVA SSU r138)
  • Per-rank abundance tables added (phylum → genus)
  • All non-stool body sites grew by 1–2 orders of magnitude
Total: 187,546 samples · 6 body sites · 775 studies
v1.0 2024 initial release
Initial Release
stool
  • 22,588 curatedMetagenomicData — shotgun, MetaPhlAn3 species + HUMAnN3 pathways
  • 6,588 American Gut Project — 16S amplicon (V4, Greengenes 13_8)
  • 1,504 MGnify selective pull — preliminary 16S
Total: 30,680 samples · 2 body sites · 93 studies
Quick Links
Browse Samples Quality Control Batch Effects Download Data Documentation How to Cite