Metadata with Ontologies¶

MESA's metadata workflow is ontology-aware. Instead of free-text tags like "soil", MESA encourages you to use ontology terms like ENVO:00001998 ("soil"), which are stable, machine-readable, and interoperable across domains.

Why ontologies?¶

Disambiguation. "Bear" can mean a mammal, a stuffed toy, or "to carry." NCBITaxon:9632 (the species Ursus) is unambiguous.
Cross-domain integration. A genomicist's GO:0008152 ("metabolic process") and an environmental scientist's ENVO:00002259 ("soil environment") can be linked through shared ontology relationships.
Search and inference. Asking "which datasets are sampled from any type of forest?" works only if your tags are linked to a hierarchy (ENVO:00000111 and its descendants).

How `mesa-mcp` helps¶

The mesa_ols_* tool family is backed by the EMBL-EBI Ontology Lookup Service (OLS) and exposes:

Tool	What it does
`mesa_ols_search`	Free-text search over OLS — returns candidate terms
`mesa_ols_get_term`	Fetch a term's metadata (label, definition, parents)
`mesa_ols_get_descendants`	Get all descendants — useful for "any kind of X"
`mesa_ols_suggest_avu`	Heuristic mapping from a free-text suggestion to a canonical term

A typical agent prompt:

I want to tag these soil-microbiome files with the right ENVO biome. Help me pick.

The agent calls mesa_ols_search("biome") → presents candidates → after you pick one (e.g. ENVO:00000446 "sandy desert"), calls ds_add_avu to write it as {attribute: "ENVO:biome", value: "sandy desert", unit: "ENVO:00000446"}. The change is mirrored into the Lakehouse catalog automatically.

Common ontologies¶

Ontology	Use for
ENVO — Environment Ontology	Biomes, habitats, environmental contexts
GO — Gene Ontology	Biological processes, molecular functions, cellular components
ChEBI	Chemical compounds
NCBITaxon	Organism taxonomy
EDAM	Bioinformatics data formats, operations
PATO	Phenotypic qualities
DOID	Human diseases
UBERON	Cross-species anatomy

The OLS hosts hundreds more. Use mesa_ols_search to discover.

AVU conventions¶

MESA standardizes on a simple AVU schema for ontology terms:

Field	Convention	Example
`attribute`	`<ONTOLOGY>:<predicate>`	`ENVO:biome`
`value`	Human-readable label	`sandy desert`
`unit`	Ontology IRI or CURIE	`ENVO:00000446`

This keeps free-text human labels searchable while preserving the canonical IRI for machine reasoning.

Browsing AVU history¶

After tagging, you can ask:

Show me how the biome tags on this dataset have changed over the last three months.

The agent calls mesa_ducklake_get_avu_history and renders a timeline. See Time-Travel Queries for more patterns.