Skip to content

README.md

Latest commit

 

History

History
77 lines (52 loc) · 2.49 KB

README.md

File metadata and controls

77 lines (52 loc) · 2.49 KB

Scripts Directory

Utility scripts for the biomedical knowledge graph system.

Analysis

benchmark_scispacy_models.py

Benchmarks scispaCy models for speed, memory, and entity extraction coverage.

poetry run python scripts/benchmark_scispacy_models.py

Output: JSON file with timing, memory usage, and entity counts per model.

Manual Verification

Scripts for testing against live infrastructure (Neo4j, LLMs). These are not automated tests.

Note: Test entities (breast cancer, BRCA1, HER2, trastuzumab) are examples from a breast cancer-focused KG. Modify entity parameters in each script to match your KG's data.

verify_agent_queries.py

Tests low-level Cypher query functions against live Neo4j database. Validates that agent query methods (disease -> genes, gene -> diseases, neighbors, explain relationships) work correctly against real Neo4j schema.

Requires: Neo4j with data, env vars NEO4J_URI, NEO4J_USER, NEO4J_PASSWORD

Usage:

export NEO4J_URI=bolt://localhost:7687
export NEO4J_USER=neo4j
export NEO4J_PASSWORD=your_password
poetry run python scripts/verify_agent_queries.py

Output: Query results with entity counts, relationship counts, and sample evidence sentences.

verify_agent.py

Tests high-level LangChain agent with LLM reasoning and tool routing. Validates that BiomedKGAgent correctly routes questions to appropriate tools and generates quality answers.

Requires: Neo4j with data, Anthropic API key, env vars NEO4J_*, ANTHROPIC_API_KEY

Usage:

export ANTHROPIC_API_KEY=your_key
poetry run python scripts/verify_agent.py
poetry run python scripts/verify_agent.py --model "claude-haiku-4-5-20251001"

Output: Agent responses with tool routing decisions, query results, and cited provenance.

verify_model_entity_types.py

Validates entity types produced by each scispaCy model. Quick verification that model behavior matches descriptions in docs/validation.md.

Requires: scispaCy models installed (bc5cdr, bionlp13cg, craft)

Usage:

poetry run python scripts/verify_model_entity_types.py

Output: Entity labels and examples from each of the 3 NER models, plus overlap analysis.

Utility

load_env.sh

Loads environment variables from .env file into current shell.

Usage:

# Load default .env
source scripts/load_env.sh

# Load custom file
source scripts/load_env.sh path/to/custom.env

Note: Must be sourced (not executed) to affect current shell environment.