Relations

A relation is a typed source of biological facts: a table whose columns carry a declared kind, namespace, and coordinate system. The ontology declares 54 of them, and each also fixes its grain, primary key, default order, and recommended filters. Relations are the nouns of the ontology. An RQL recipe begins by importing the relations it needs, and every join connects one relation to another.

Importing a relation

Recipes pull relations in by name. Importing a relation brings its typed columns into scope — nothing about a source is available to a query until it is imported.

imports:
  - relation: variant_features
  - relation: ligand_neighbors
from: variant_features

The from line names the relation the query starts from; every other imported relation reaches it through the join tree.

The relation families

Relations are organized into families that mirror how biological evidence is produced. These are the sources recipes draw on most often.

Structures

Everything anchored to a solved or predicted 3-D structure.

Relation	What it holds
`structure_summary`	Structure identity and per-structure quality (e.g. `mean_plddt`)
`residue_features`	Per-residue signals along a structure
`pocket_features`	Measured binding pockets (`volume_angstrom3`, `mean_local_plddt`)

Sequences

The sequence layer and its bridges into structure.

Relation	What it holds
`source_sequence_records`	Canonical sequences
`protein_variant_mappings`	Variant positions mapped sequence → structure
`variant_features`	Per-variant annotations (`variant_raw`)

Evidence

Relationships derived between entities — similarity and contact.

Relation	What it holds
`fold_hits`	Structure fold-similarity hits (`query_structure_id`, `target_structure_id`)
`homology_search_hits`	Sequence-similarity hits
`interface_features`	Chain–chain interfaces (`chain_a_id`, `chain_b_id`)
`interface_contacts`	Residue contacts across an interface (`distance_angstrom`)

Ligands

Small molecules and their spatial relationship to protein.

Relation	What it holds
`ligand_neighbors`	Ligands near a residue (`ligand_id`, `distance_angstrom`)
`pocket_features`	Pockets as ligand-binding candidates

Relations are typed, not raw tables

The difference between a relation and the underlying database table is the types. Two relations may both expose a structure_id column, but each column declares its kind and coordinate system — so the ontology knows whether they can be joined directly, need a lossy or range mapping, or must never be compared at all.

One schema, many sourcesEvery relation is validated against one shared, versioned ontology schema before it can be imported. That single schema is what makes a join between two independently produced sources safe to write.

Next: Joins & join safety — how relations are connected, and what the ontology checks before a single row is read.

Getting started

CLI

Homology search

Ontology

AI Foundry

Evals

Importing a relation

The relation families

Structures

Sequences

Evidence

Ligands

Relations are typed, not raw tables

​Importing a relation

​The relation families

​Structures

​Sequences

​Evidence

​Ligands

​Relations are typed, not raw tables

Importing a relation

The relation families

Structures

Sequences

Evidence

Ligands

Relations are typed, not raw tables