What this entity is
Canonical name: NCBI environment metadata. 1
Known aliases: NCBI environmental metadata; genome environment metadata. 1
Stable external identifier: Not reported in the source documents. 1,2
NCBI environment metadata is genome-associated environmental information used to classify bacterial genomes into categories such as clinical, human gut, soil, aquatic, host-associated, and other environmental sources. 1 It includes fields such as geo_loc_name, isolation_source, host, env_broad_scale, env_local_scale, and env_medium, supporting habitat harmonization and continuous-environment analyses. 2
The PGP Gene Distribution Across Environments & Pangenomes extends these uses to plant-growth-promoting gene distributions: its KBase Data Lakehouse pangenome analysis used isolation-source metadata for 291,279 genomes, of which 93.5% were classifiable, and conservatively identified 1,637 species (5.9% of species with an environment label) as soil/rhizosphere dominant. 3 The plant-microbiome analysis extends this use to compartment-specific classification, identifying 7,995 plant-associated genomes among 293,059 GTDB r214 genomes across root, rhizosphere, phyllosphere, endophyte, and other plant contexts. 4 This supports concerns that free-text environmental metadata are incomplete and noisy, while showing that the resource can test soil-associated gene enrichment and plant-compartment associations. 3,4
Key facts in the source
The assembled dataset contained 280,337 genomes with NCBI environment metadata. 1 Of those genomes, 93.5% were classified at the genome level. 1 Species-level majority-vote environment classifications covered 95% of species, corresponding to 13,981 species. 1
These classifications were used to compare AMR gene-cluster diversity across environments in the Environmental Resistome at Pangenome Scale. 1 Clinical-source species had a median of 5 AMR gene clusters, compared with 2 in soil, aquatic, and host-associated species. 1 The environment–AMR association remained present at 50%, 60%, 75%, and 90% majority-vote thresholds, with η² values from 0.044 to 0.056. 1
Environment classifications were integrated with kbase-ke-pangenome to analyze 82,908 AMR gene clusters across 14,723 bacterial species and 293K genomes. 1 The newer atlas refines this integration: among 14,723 AMR-carrying species, 7,838 (53.2%) received a non-Other/Unknown classification across 6 categories, while 46.8% remained Other/Unknown because of sparse and inconsistent free-text isolation_source metadata in NCBI BioSample records. 5
The atlas reports 10.6 AMR clusters per Human/Clinical species, versus 4.6 for Soil/Terrestrial, 3.9 for Aquatic, and 3.0 for Animal; this supports the clinical-versus-environmental contrast while its restricted classification coverage qualifies the comparison. 5 A separate NCBI keyword-based classifier assigned environments to 1,190/1,307 species (91%) and found higher AMR burden in host-associated species than in terrestrial or aquatic species, with Kruskal-Wallis p < 0.05; human-clinical isolates had the highest burden. 6 A rule-based approximation of BacDive categories classified 459/1,307 species (35%) and agreed in direction, but both classifiers are approximate. 6
The ecotype analysis extends metadata use from categorical AMR comparisons to genome-wide gene-content similarity: among 172 species with sufficient data, the median partial correlation was 0.0025 for environment versus 0.0143 for phylogeny, and phylogeny dominated in 60.5% of species. 7 A significant positive environmental effect occurred in 12 species (7.0%), a significant negative effect in 4 species (2.3%), and no significant effect in 156 species (90.7%). 7 This refines interpretation by suggesting that environmental effects may be concentrated in specific gene subsets rather than genome-wide gene content. 7
The reanalysis supports the original ecotype null conclusion while refining classification: among 224 species meeting thresholds of >=20 genomes with AlphaEarth embeddings and >=30% coverage, 106 (47%) were majority human-associated, 47 (21%) majority environmental, and 71 (32%) Mixed/Other. 8 Environmental species had a median partial correlation of 0.051, versus 0.084 for human-associated and 0.109 for Mixed/Other species; the one-sided Mann-Whitney U test for Environmental > Human-associated gave U=1536 and p=0.83. 8 Continuous Spearman analyses found no relationship between environmental fraction and partial-correlation strength (rho=-0.085, p=0.25), or between human-associated fraction and correlation strength (rho=0.030, p=0.69). 8
These results contradict the hypothesis that clinical sampling bias in the AlphaEarth subset explains the weak environment–gene-content relationship. 8 Of 30 species with NaN partial correlations, the NaN rate was 10/47 = 21% for Environmental, 13/66 = 20% for Mixed/Other, and 7/100 = 7% for Human-associated species. 8
The AlphaEarth environment explorer refines the metadata context by analyzing 83,287 pangenome genomes with 64-dimensional satellite-derived embeddings: 38% were human-associated, including 16,390 Human clinical, 13,466 Human gut, and 1,669 Human other genomes; Soil, Marine, and Freshwater contributed 6,073, 5,850, and 5,840 genomes, respectively. 2 Isolation_source was available for 76,295 (91.6%), host for 53,091 (63.7%), env_broad_scale for 34,800 (41.8%), env_local_scale for 31,541 (37.9%), and env_medium for 31,483 (37.8%). 2
A keyword workflow mapped 5,774 unique isolation_source values to 12 broad categories, capturing 71% of genomes with a label; 17% were Other and 12.5% were Unknown because isolation_source was missing or null. 2 The NCBI ncbi_env table contained 334 distinct harmonized attribute names across 4.1M rows; collection_date, geo_loc_name, and isolation_source were populated for 273K, 272K, and 245K genomes, respectively. 2 These figures support metadata-sparsity concerns while showing that structured fields provide a cleaner but incomplete fallback. 2
The Prophage Gene Modules and Terminase-Defined Lineages Across Bacterial Phylogeny and Environmental Gradients extends these metadata applications from AMR and ecotype analyses to prophage-module ecology: across 27,702 bacterial species, environmental categories derived from genome-associated metadata were tested alongside genome size and family-level phylogeny. 9 In a 1,773-species subsample, environment significantly affected prophage-module composition (PERMANOVA p=0.01, F=30.04) beyond phylogeny (F=6.17), while genome size was dominant (F=212.99); within genome-size quartiles, environmental effects remained significant with all p < 6.5e-78. 9 Constrained analysis across 18,031 species identified 8 significant module-by-environment enrichments at FDR < 0.05, including tail, head morphogenesis, and anti-defense enrichment in human-associated environments. 9 This supports the use of environment metadata for module-level ecological comparisons, while the source’s 9,659-species (35%) other_unknown category refines interpretation by showing that coarse or missing labels still limit resolution. 9
The prophage study also extends metadata integration to 6,365 NMDC metagenomic samples: taxonomy-based prophage-burden inference achieved 87.2% median matching coverage and found 57 significant module–abiotic correlations at FDR < 0.05, with the strongest reported association being packaging with pH (Spearman rho=0.519). 9 This cross-collection result is an environmental association rather than a causal demonstration, and the indirect genus-level burden inference had not been independently validated for prophage genes. 9
The new pitfalls refines these coverage estimates by documenting that ncbi_env is entity-attribute-value data rather than a flat genome-metadata table. 10 Metadata extraction therefore requires joining genome.ncbi_biosample_id to ncbi_env.accession, filtering attributes, and pivoting the long format. 10 A separate per-genome classification yielded 52.7% unknown labels, or 94,957 of 180,025 genomes, because most BioSample records lacked structured isolation metadata; species-level majority-vote keyword classification reached 91% coverage. 10 These figures are dataset- and workflow-specific and do not contradict the higher coverage reported for the filtered AlphaEarth and pangenome subsets. 10
The plant-microbiome analysis refines the interpretation of compartment labels: among 607 root, rhizosphere, and phyllosphere species, refined Jaccard profiles produced PERMANOVA R² = 0.071 and db-RDA location-only R² = 0.060, with 84% of the PERMANOVA signal attributable to centroid shifts and 16% to dispersion. 4 The earlier R² = 0.527 fell to R² = 0.072 after excluding three genome-rich species per compartment, indicating that much of the initial effect reflected taxonomic sampling rather than a community-wide compartment signal. 4 This supports treating isolation_source-derived plant categories as weak, taxonomically confounded habitat evidence rather than species-level ecological discriminators. 4
AlphaEarth cosine distance increased monotonically with geographic distance, from 0.41 at <100 km to 0.82 at 10K–20K km, with a plateau above 5,000 km. 2 Environmental samples showed a 3.4x geographic gradient, versus 2.0x for human-associated samples, refining interpretation of geographic interchangeability. 2 The pitfalls analysis supports stratifying by environment category or excluding human-associated samples: the pooled geographic distance–embedding distance curve had a 2.0x near-versus-far ratio, compared with 3.4x for environmental samples and 2.0x for human-associated samples. 10 This does not establish a stronger environment–gene-content relationship and instead motivates analyses of environmental-only samples with sampling and metadata-quality controls. 2
The PGP pangenome analysis extends environment-stratified gene-distribution analysis beyond AMR: among 1,039 soil/rhizosphere species and 10,233 species from other environments, acdS, pqqC, and hcnC were soil-enriched, whereas nifH was depleted; acdS enrichment remained strong after phylum-level fixed effects and in a strict rhizosphere-only sensitivity analysis. 3 This result is observational and does not remove the classification limitations described above. 3
The plant-microbiome analysis further extends metadata-linked ecological analysis to host and plant-compartment structure: it identified 1,307 plant-associated species across 19 testable hosts, but only 11.7% Jaccard overlap between 487 pangenome-derived plant genera and 438 MGnify rhizosphere genera. 4 This supports the existing warning that isolation metadata and metagenomic detection measure different phenomena and should not be treated as interchangeable habitat labels. 4
Limitations in interpretation
NCBI sampling overrepresented clinical isolates, while soil and aquatic species were undersampled, which may inflate the clinical-versus-environmental AMR contrast. 1 Majority-vote species classifications collapsed within-species environmental variation, although threshold sensitivity analyses partly assessed this limitation. 1 Environment and phylogeny remained deeply entangled despite family-level controls, and only 14% of tested families showed significant within-family environment effects after false-discovery-rate correction. 1 The source supports an observational association rather than a causal conclusion about environmental exposure and AMR. 1
The atlas supports metadata-sparsity concerns: 46.8% of AMR-carrying species were Other/Unknown, and comparison was restricted to 7,838 non-Other/Unknown species. 5 The within-species analysis refines this limitation: 52.7% of genomes had no classifiable isolation_source, leaving only 2 species with sufficient within-species environmental diversity for strict environment–ecotype chi-squared tests. 6 AlphaEarth embeddings covered only 28% of genomes and were biased toward genomes with geographic metadata. 5
The prophage analysis supports these warnings: only 28% of genomes had AlphaEarth environmental embeddings, the 10 environmental categories collapsed substantial within-category variation, and other_unknown contained 9,659 species (35%). 9 Its environmental effects therefore demonstrate metadata-linked association, not a fully resolved habitat mechanism; dedicated prophage detection was also absent, and NMDC burden estimates assumed genus-level conservation. 9
The ecotype analysis supports incomplete environmental representation: AlphaEarth embeddings covered only 28.4% of genomes, coordinates were often missing or imprecise, and host-associated coordinates may describe collection sites rather than actual microenvironments. 7 It found no significant difference in environmental effects between environmental and host-associated bacteria (p=0.66), but this result is limited by those constraints. 7 Its partial-correlation approach assumes linear relationships between distance matrices and may miss nonlinear effects. 7
The explorer refines the coordinate limitation: among 83,286 genomes with coordinates, 50,109 (60.2%) were Good, 30,469 (36.6%) Suspicious cluster, and 2,708 (3.3%) Low precision. 2 Suspicious-cluster heuristics can flag legitimate field sites as well as institutional addresses, so isolation_source homogeneity is needed before treating geographic metadata as reliable habitat evidence. 2 Its AlphaEarth subset contained 3,838 genomes with at least one NaN among 64 dimensions, with cause unresolved. 2 The pitfalls quality check likewise found 3,838 of 83,287 genomes, or 4.6%, with at least one NaN among 64 dimensions; filtering left 79,449 genomes. 10
The reanalysis refines cross-method interpretation: its median partial correlation across all 183 species was 0.081 versus 0.003 originally, characterized as a 27x difference, because it used all genomes with embeddings, including up to 3,505 genomes per species, rather than diversity-maximizing downsampling capped at 250 genomes. 8 Absolute correlations are not comparable across methodologies, but the Environmental versus Human-associated comparison used one consistent methodology. 8 Klebsiella pneumoniae was excluded after exceeding Spark’s maxResultSize during gene-cluster extraction and consequently had no correlation data. 8
The reanalysis suggests, rather than establishes, that AlphaEarth embeddings may capture regional epidemiological patterns in human-associated species rather than ecologically relevant variation; environmental embeddings had a 3.4x geographic-signal ratio versus 2.0x for human-associated embeddings. 8 Unresolved tests include downsampled versus full-genome extraction, functional gene subsets, structured ENVO terms, and genome-count covariate control. 8 The PGP analysis likewise notes that soil classification is conservative and noisy, and that its observed enrichment may underestimate true rhizosphere enrichment. 3
The plant-microbiome analysis refines these limitations: only 7,995 of 293,059 genomes, or 2.7%, had plant-associated annotations, the endophyte category contained only 29 species, and plant labels were inferred primarily from variable NCBI isolation_source metadata. 4 Its corrected compartment effect was weak, and its host/subclade analysis was testable for only 18 of 65 candidate species because of incomplete phylogenetic-tree coverage. 4 Thus, plant-associated classifications provide useful hypotheses but do not establish causal habitat specialization or broad species-level ecotypes. 4
The pitfalls document further refines reproducibility requirements: NCBI environment joins should begin with live schema inspection, because table schemas, naming conventions, access, and metadata contents can change; missing attributes or zero joins must not be interpreted as biological absence. 10 Free-text isolation-source labels require explicit keyword or synonym reconciliation, and classifications should retain the original accession, attribute, and transformation provenance. 10
Related pages
- kbase-ke-pangenome — pangenome collection linked to environment metadata. 1
- amrfinderplus — AMR annotation resource used in the resistance-gene analysis. 1
- alph-aearth — supplementary continuous environmental embedding dataset. 1
- environmental-resistome — cross-project synthesis of environment-associated AMR patterns. 1
- ecotype-environment-gene-content — environmental similarity and bacterial gene-content variation. 2
- environment-embedding-geography — geographic and environment-stratified AlphaEarth behavior. 2
- provenance-aware-resource-discovery — live schema discovery, entity-attribute-value extraction, and provenance-preserving metadata joins. 10
- Pan-Bacterial AMR Gene Landscape — pan-bacterial AMR atlas integrating environmental metadata, pangenomes, and fitness data. 5
- Within-Species AMR Strain Variation — within-species AMR variation and environmental classification. 6
- Ecotype Correlation Analysis — environmental and phylogenetic predictors of gene-content similarity. 7
- Ecotype Reanalysis — Environmental vs Human-Associated Species — reanalysis of clinical sampling bias and environment–gene-content relationships. 8
- AlphaEarth Embeddings, Geography & Environment Explorer — AlphaEarth embeddings, geography, coordinate quality, and environment harmonization. 2
- PGP Gene Distribution Across Environments & Pangenomes — plant-growth-promoting gene distribution across environments and pangenomes. 3
- Plant Microbiome Ecotypes — Compartment-Specific Functional Guilds and Their Genetic Architecture — plant compartments, host associations, and metadata-linked genomic ecology. 4
- Prophage Gene Modules and Terminase-Defined Lineages Across Bacterial Phylogeny and Environmental Gradients — prophage-module composition, environmental gradients, TerL lineages, and NMDC validation. 9
- pitfalls — database, metadata, join, coverage, and reproducibility pitfalls affecting environmental classification. 10