| name | biocypher |
| description | Build knowledge graphs with BioCypher — adapters, schema_config.yaml, biocypher_config.yaml, and multi-backend export (Neo4j, CSV, PostgreSQL, RDF). Use for standalone KG ETL pipelines, writing schema_info.yaml, Neo4j bulk import, or tuning dbms/output settings. Trigger on BioCypher, schema_config, biocypher_config, write_nodes, write_edges, knowledge graph construction, Biolink schema, or graph database export. |
BioCypher: build a knowledge graph
BioCypher is a Python library for ontology-grounded KG construction. Three pieces: adapters (data in), schema_config.yaml (what the graph means), biocypher_config.yaml (where it goes).
pip install biocypher
pip install "biocypher[neo4j]"
Project template: https://github.com/biocypher/project-template
Workflow
1. Define scope (entities, relations, ontologies)
2. Write adapters → yield node/edge tuples
3. Author schema_config.yaml (input_label ↔ ontology types)
4. Configure biocypher_config.yaml (dbms, paths)
5. BioCypher().write_nodes / write_edges → export
Read references/schema-config.md before authoring schema. Read references/outputs.md for dbms choice, import scripts, and write_schema_info().
Adapters
Yield tuples — BioCypher maps _type via input_label in schema:
def node_generator():
for row in rows:
yield (row["uniprot_id"], "protein", {"name": row["name"]})
def edge_generator():
for row in edges:
yield (None, row["src"], row["tgt"], "interaction", {"score": row["score"]})
Reuse existing adapters: https://github.com/orgs/biocypher/projects/3/views/2
Run
from biocypher import BioCypher
bc = BioCypher()
bc.write_nodes(node_generator())
bc.write_edges(edge_generator())
bc.write_import_call()
bc.write_schema_info()
BioCypher() reads biocypher_config.yaml from cwd or config/. Override paths via constructor args if needed.
Config essentials
biocypher:
dbms: neo4j
schema_config_path: config/schema_config.yaml
offline: true
output_directory: biocypher-out
head_ontology:
url: https://github.com/biolink/biolink-model/raw/v3.2.1/biolink-model.owl.ttl
root_node: entity
protein:
represented_as: node
preferred_id: uniprot
input_label: protein
Small graphs / agent prototyping
create_workflow() — in-memory, JSON export, no DBMS. See references/outputs.md "Agent API". Not for large ETL.
Verify
After export, check file counts and spot-read CSVs or run import dry-run. For Neo4j, compare expected vs actual edge counts — silent drops often mean id namespace mismatches at the adapter layer.
Optional next step
With schema_info.yaml and a loaded graph database, ask the user before loading the biochatter skill for natural-language querying.
