We use the pole dataset to demonstrate the building of a BioCypher knowledge graph and the facilitation of querying the graph using BioChatter and BioChatter Light. The dataset is a public synthetic dataset from Manchester, U.K. Check our vignette for more details: https://biochatter.org/vignettes/kg/
To start a local instance of the Neo4j database and BioChatter Light, clone the repository and run the Docker compose setup. Docker needs to be installed and running on your machine.
Important
For using the OpenAI GPT model that is called from
BioChatter Light, you need to provide your OpenAI API key through the environment
variable OPENAI_API_KEY. If you do not provide a key, the query generation
will fail.
git clone https://github.com/biocypher/pole.git
cd pole
export OPENAI_API_KEY=sk-... # or add it to your .bashrc or .zshrc
docker compose up -d
The Docker compose workflow will take care of building the database, importing
and deploying in Neo4j, and starting the BioChatter Light app. Given you have provided
your API key, you can now open the BioChatter Light app in your browser at
http://localhost:8501. Via the configuration in the
docker-compose.yml, we already set BioChatter Light to only display the knowledge
graph tab, and the correct connection details should be set. Entering a question and
confirming with Enter will generate a Cypher query (via BioChatter) and execute it on
the database. The generated query and results will be displayed in the space below the
interface. You can modify the query and rerun it (CMD+Enter on Mac, Ctrl+Enter on
Windows) without having to call the LLM again. To generate a new query, simply update
the question text.
Some example questions:
where happened most crimeswho committed the most crimeswho knows people that committed many crimes while not being criminal themselves
You can also visit http://localhost:7474 to access the
Neo4j browser interface. It requires no authentication (simply press Connect)
and allows you to explore the database and run Cypher queries.