UMLS Semantic Network (CSV & Parquet)

This repository contains dataset exports of the UMLS Semantic Network, available in both CSV and Parquet formats.

These files provide a machine-readable reference for the high-level categories (Semantic Types) and relationships (Semantic Relations) used to classify concepts within the Unified Medical Language System (UMLS) Metathesaurus.

About the Data

The UMLS Semantic Network is one of three Knowledge Sources developed by the National Library of Medicine (NLM). It provides a consistent categorization of all concepts represented in the UMLS Metathesaurus.

The network consists of:

1. Semantic Types: Broad categories (e.g., "Disease or Syndrome", "Clinical Drug", "Organism").
2. Semantic Relations: Relationships that exist between these types (e.g., "treats", "causes", "location_of").

File Contents

The repository includes the following files:

1. Semantic Types

• Files: UMLS Semantic Network - type.csv, umls_semantic_network_type.parquet
• Description: Defines the nodes in the Semantic Network.

Column	Description
semantic_type	The name of the category (e.g., "Disease or Syndrome").
definition	A text description of what the type encompasses.
identifier	The unique Type Unique Identifier (TUI) (e.g., "T047").
note	Additional usage notes or guidelines for assigning this type.

2. Semantic Relations

• Files: UMLS Semantic Network - relation.csv, umls_semantic_network_relation.parquet
• Description: Defines the edges/relationships available in the Semantic Network.

Column	Description
semantic_relations	The name of the relationship (e.g., "treats", "part_of").
definition	A text description of the relationship's meaning.
identifier	The unique identifier for the relation (e.g., "T154").
note	Context on how the relationship is applied.

Usage Examples

Python (Pandas)

You can easily load these datasets using Python and Pandas. The Parquet format is recommended for faster I/O and preserved data types.

import pandas as pd

# Load Semantic Types (Parquet)
types_df = pd.read_parquet('umls_semantic_network_type.parquet')

# View the definition of "Disease or Syndrome" (T047)
disease_type = types_df[types_df['identifier'] == 'T047']
print(disease_type[['semantic_type', 'definition']])

# Load Semantic Relations (CSV)
relations_df = pd.read_csv('UMLS Semantic Network - relation.csv')

# Find all relations regarding "treatment"
treatment_rels = relations_df[relations_df['semantic_relations'].str.contains('treat', case=False)]
print(treatment_rels)

Data Accuracy

Please note that these files were copy-pasted by hand. If you encounter any errors or inconsistencies, please create an issue in this repository so it can be fixed.

Source and Attribution

This data was sourced from the National Library of Medicine (NLM) Unified Medical Language System (UMLS).

• Source URL: https://uts.nlm.nih.gov/uts/umls/semantic-network/root
• Publisher: U.S. National Library of Medicine

License & Disclaimer

Please Note: The UMLS Semantic Network and Metathesaurus are subject to the UMLS Metathesaurus License.

This repository provides these files for educational and ease-of-access purposes and is not officially affiliated with the NLM.