Nguyendat-bit/UMLS-KG

Build a knowledge graph from UMLS Knowledge Sources (2022) with load, visualize and query with Neo4j and Scispacy

Load, Visualize, Query UMLS Knowledge graph with Neo4j

Overview

On this repo, you can load UMLS knowledge graph and Visualize with Neo4j and you can use Neo4j python driver to query nodes or relationships what you want.

First, you need to install UMLS dataset and load it into mysql.

Usage

Requirement

• python >= 3.9
• pymysql >= 1.0.2
• csv >= 1.0
• neo4j python driver >= 4.4.5
• Docker

Create your graph

Run this code below:

python build_graph.py --host ${host} --user ${user} --password ${password} --database ${database}

Example:

python build_graph.py --host localhost --user root --password root --database umls2022

Ouput:

3618459 concepts
7577026 atoms
25004792 relationships

Import UMLS data into Neo4j

Run this code below:

docker run -it -v /d/umls/2022AA/META:/data -v /d/umls/2022AA/META:/var/lib/neo4j/import neo4j:3.5 bin/neo4j-admin import --nodes=import/MRCONSO.processed.csv /--nodes=import/MRAUI.processed.csv --relationships=import/MRREL.processed.csv

In the above script, --nodes represents specifying location of node tables used, and --relationships represents specifying location of relationship tables. You can specify multiple nodes or relationships table for importing various types of nodes and relationships.

Now you can visualize and query nodes or subgraphs what you want.

Befor to do that, you should shutdown docker container of neo4j image and run this code below:

docker run -it -p7474:7474 -p7687:7687 -v /d/umls/2022AA/META:/data -v /d/umls/2022AA/META:/var/lib/neo4j/import --env NEO4J_AUTH=neo4j/test neo4j:3.5 

In the above script, neo4j is user and test is password.

For example, I will query nodes that have relationship with CUI C0000039.

Using Neo4j python driver to query database

First you need to install neo4j package:

pip install neo4j

Run this code below:

import neo4j 
from neo4j import GraphDatabase
graphdb= GraphDatabase.driver(uri= "bolt://localhost:7687", auth=("neo4j", "test"))
session= graphdb.session()
q1="match (p: Concept {CUI: 'C0000039'}) return p"
nodes= session.run(q1)
for node in nodes: 
    print(node)

Output:

<Record p=<Node id=1 labels=frozenset({'Concept'}) properties={'name': '1,2-dipalmitoylphosphatidylcholine', 'CUI': 'C0000039'}>>