TaxonMech taxonmech

active seeded x-mech-suite kg-microbe

TaxonMech is a knowledge base of microbial taxa and the strains behind them. One record per taxon, keyed by NCBI Taxonomy and harmonized with GTDB, LPSN, BacDive, MediaDive, GOLD, Madin et al. and BactoTraits through the kg-microbe transforms. Records are species and below. Higher taxa appear only in a record’s lineage. Each record keeps every source’s own identifier, mapping predicate and assertion count, the LPSN nomenclature with type strain designations, the GTDB species mapping with its genome count, and the BacDive strains with their culture-collection deposits. A primary focus is linking strain identifiers to genome identifiers: NCBI GenBank and RefSeq assemblies first, then GTDB, BV-BRC, IMG and AllTheBacteria records, each link carrying the source record that asserts it. Phenotypes, media and isolation sources are left to TraitMech, CultureMech and HabitatMech.

One record is

a microbial taxon at species rank or below

625960 records as of 2026-10-12

Counted from data/taxa/*.yaml at 56dcc69. Refreshed weekly.

Links

Homepage

Repository

LinkML schema

License

Records: CC0 1.0

Domains

microbiology organisms genomics

Taxa: NCBITaxon:2 NCBITaxon:2157

Curation model

Agent-curated	Yes. AI agents generate or maintain most of the content.
Human review	`pull_request`. Agent changes arrive as pull requests that a person reviews and merges.
Agents and tools	Claude Code, culturebotai-claw
Evidence policy	Every causal edge must carry a citation. Seeded descriptive fields carry source attestations instead. A strain-to-genome link must name the source record that asserts it. Species membership and strain-name matching never supply a link.
Validation in CI	schema corpus_reproduction
Append-only history	Yes
Record identifiers	Records are keyed by their NCBITaxon CURIE. Filenames are pinned by data/taxa/PATHS.tsv so a re-seed never renames an existing record.

Features

causal_mechanism_graphs Records carry directed, evidence-backed graphs of mechanism.
ontology_grounding Entities bind to public ontology terms with labels.
append_only_history Every change is an event in an append-only history directory.
static_browser A static site lets a reader browse every record.

Grounding and sources

Ontologies and vocabularies: NCBITaxon GTDB BacDive GOLD insdc.gca refseq biosample bioproject img.taxon brc.genome

Source	Relation	What is taken
NCBI Taxonomy	prov:hadPrimarySource	The identity of every record, its rank and its lineage, from a pinned snapshot.
GTDB	prov:wasDerivedFrom	Species mappings with the assemblies behind each, via kg-microbe.
LPSN	prov:wasDerivedFrom	Nomenclature, authority, correct-name status and type strain designations, via kg-microbe.
BacDive	prov:wasDerivedFrom	Strain records with culture-collection deposits and genome assertions.
GOLD	prov:wasDerivedFrom	Organism, sequencing-project and analysis-project links from the public workbook.
AllTheBacteria	prov:wasDerivedFrom	Assembly records joined through BioSample evidence.
StrainInfo	prov:wasDerivedFrom	Deposit-to-NCBI assertions and typed strain, deposit and nucleotide references.
MediaDive	prov:wasDerivedFrom	Media attestations per taxon.
Madin et al.	prov:wasDerivedFrom	Prokaryotic phenotypic trait compilation, as a source attestation.
BactoTraits	prov:wasDerivedFrom	Functional trait database, as a source attestation.
kg-microbe	prov:used	The harmonized KGX inputs and the GTDB-to-NCBI and LPSN-to-NCBI mappings.

Products

Product	Kind	Format	Description
Taxon records `taxonmech.records`	RecordCorpus	yaml	One YAML file per taxon under data/taxa//.yaml.
TaxonMech LinkML schema `taxonmech.schema`	DataModelProduct	linkml linkml mech_shared
TaxonMech browser `taxonmech.browser`	GraphicalInterface	html
Strain assembly inventory `taxonmech.strain_assemblies`	RecordCorpus	tsv	Uncapped inventory of strain-to-NCBI-assembly links, joined to BacDive strains on strain_id. Unlisted strains live here and not in the records.

Relations to other Mechs

follows_pattern_of dismech
hands_off_to traitmech: Strain phenotypes are not held in TaxonMech.
hands_off_to culturemech: Growth media are not held in TaxonMech.
hands_off_to habitatmech: Isolation sources are not held in TaxonMech.

Contacts

CultureBotAI (Organization) CultureBotAI
Marcin P. Joachimiak (Individual) ORCID 0000-0001-8175-045X mjoachimiak@lbl.gov realmarcin

Registry entry

Created 2026-09-16. Last modified 2026-09-16. Edit this entry on GitHub.

On 2026-09-16 the record count was the generated corpus figure in the README at commit 7d1756a. It is now counted from the repository tree. Of 625,960 records then, 605,695 are Bacteria, 14,013 Archaea and 6,243 Eukaryota. 572,305 are species and 46,325 are strains. No record is REVIEWED and 14 are DEPRECATED. The rest are SEEDED or PROPOSED. Records are generated and just verify-corpus requires that they reproduce from the inventories plus curation/seed_scope.tsv. 21,071 listed strains carry genome identifier links.

LPSN and StrainInfo have no Bioregistry prefix, so they are listed under data sources and not under ontologies. BV-BRC (PATRIC) genome identifiers are listed as brc.genome.