Goal – pull every "canonical" LinkML schema (Biolink, SSSOM, NMDC,
MIxS, GA4GH, Bioregistry) so you can import: them rather than
redefine them.
Time – 30 minutes.
Prereqs – chapters 01, 02.
``mermaid`
mindmap
root((Core LinkML schemas))
Biolink
bmt
Translator KP API
SSSOM
sssom-schema
sssom-py
NMDC
microbiome metadata
MIxS
genome metadata
GA4GH
Phenopacket
Pedigree
Bioregistry
curies
prefixmaps
OBO
via OAK
You almost never write these from scratch. Treat them as third-party
libraries: install, pin a version, import.
The "OS-level" schema for biomedical knowledge graphs. Everyone in the
NCATS Translator ecosystem uses it.
bash
YAML schema
curl -L -o downloads/biolink/biolink-model.yaml \
https://raw.githubusercontent.com/biolink/biolink-model/master/src/biolink_model/schema/biolink_model.yaml
Python toolkit
pip install bmt
`
1.2 What's inside
| Layer | Examples |
| --- | --- |
| Top-level entities |
BiologicalEntity, ChemicalEntity, Disease, Gene, PhenotypicFeature |
| Associations (edges) | GeneToDiseaseAssociation, ChemicalToGeneAssociation |
| Predicates | treats, affects, interacts_with (mapped to RO/SIO) |
| Mixins | ThingWithTaxon, MacromolecularMachineMixin |
1.3 Use bmt to introspect
`python
from bmt import Toolkit
tk = Toolkit() # downloads model on first use
print(tk.get_descendants("disease"))
print(tk.get_element("treats")) # predicate metadata
print(tk.get_all_classes()[:10])
`
1.4 Import into your schema
`yaml
imports:
- linkml:types
- https://w3id.org/biolink/biolink-model
classes:
CytognosisDisease:
is_a: biolink:Disease
slot_usage:
id:
pattern: "^MONDO:[0-9]+$"
`
Checkpoint –
linkml-validate -s your_master.yaml succeeds with
the Biolink import;
gen-pydantic produces a Pydantic class that
inherits from
biolink_model.Disease.
2. SSSOM schema
Already covered in 14_sssom_snomed_workflow.md
and
sssom_tooling_for_cytognosis.md. The summary:
`bash
pip install sssom-schema
python -c "
import sssom_schema, os
print(os.path.dirname(sssom_schema.__file__) + '/schema/sssom_schema.yaml')
"
`
In your master schema:
`yaml
imports:
- sssom_schema:sssom_schema
`
3. Other LinkML schemas worth knowing
| Schema | Domain | Repo |
| --- | --- | --- |
| Biolink Model | KG core types/predicates | https://github.com/biolink/biolink-model |
| SSSOM Schema | mappings | https://github.com/mapping-commons/sssom |
| NMDC Schema | microbiome data | https://github.com/microbiomedata/nmdc-schema |
| MIxS | genome/metagenome metadata | https://github.com/GenomicsStandardsConsortium/mixs |
| Phenopacket Schema | clinical case rep. | https://github.com/phenopackets/phenopacket-schema |
| GA4GH Pedigree | family pedigrees | https://github.com/ga4gh/pedigree-standard |
| ROBOT/OBO patterns | DOSDP templates | https://github.com/ontodev/robot |
| Bioregistry LinkML | identifier registry | https://github.com/biopragmatics/bioregistry-schema |
| LinkML metamodel | LinkML for LinkML | https://github.com/linkml/linkml-model |
You can pull any of them with
git clone --depth 1 into downloads/
and them, or install from PyPI when packaged
(pip install nmdc-schema, pip install phenopackets).
4. The "import" pattern
`yaml
schemas/cytognosis/master.yaml
id: https://cytognosis.org/schemas/master
name: cytognosis_master
imports:
- linkml:types
- https://w3id.org/biolink/biolink-model
- sssom_schema:sssom_schema
- ../bioschemas/dataset
- ../bioschemas/computationalworkflow
- ../sosa_ssn/sosa_ssn
- ../cellxgene/cellxgene
- ../openalex/openalex
- ./scholarly # local subschema
- ./artifacts # local subschema
`
Two takeaways:
1. **Imports can be local files, PyPI-installed schemas (resolved by
name), or absolute URLs.**
2. Imports are transitive – Biolink already imports a lot, so be
careful not to override its names.
5. Verify the assembled schema
`bash
SCHEMA=schemas/cytognosis/master.yaml
linkml-validate --schema "$SCHEMA" --target-class CytognosisDisease \
test_data/disease.yaml
gen-pydantic "$SCHEMA" > build/master.py
gen-erdiagram "$SCHEMA" > build/master.mmd
gen-json-schema "$SCHEMA" > build/master.schema.json
`
If
gen-erdiagram produces a Mermaid file with all your imported types
visible, your imports are wired correctly.
6. Hands-on
1. Add Biolink and SSSOM imports to a fresh
schemas/cytognosis/master.yaml.
2. Define one class that inherits from biolink:Disease.
3. Run linkml-validate and gen-pydantic end-to-end.
4. Use bmt to print all descendants of biolink:ChemicalEntity.
7. Pitfalls
Biolink is big. First codegen takes a couple of minutes; subsequent
runs are fast.
Slot name collisions – if you redeclare name or id your schema
silently overrides Biolink's. Use slot_usage instead.
bmt's default model URL points at GitHub master. Pin a version
for reproducibility: Toolkit(schema='path/to/biolink-model.yaml').
Phenopacket schema publishes both Protobuf and LinkML; for
LinkML pulls use phenopacket-schema/src/schema/`.