UniD3 - Drug Discovery Knowledge Graph

Overview

UniD3 is a multi-knowledge-graph built from 150,000+ PubMed articles, stored as 6 GraphML files. It supports drug-disease matching, effectiveness assessment, and drug-target analysis.

• Source: https://github.com/QSong-github/UniD3
• Local path: resources_metadata/drug_knowledgebase/UniD3
• Format: GraphML (6 files, e.g. UniD3_L1T1.graphml)
• Dependency: networkx

Node Schema

Each node contains:

Field	Description
entity	Node name (e.g. RESPIRATORY DISEASES)
entity_type	Type label (e.g. DISEASE, DRUG, GENE, HOST, BIOLOGICAL PROCESS)
description	Free-text description from PubMed articles
source_id	Chunk ID linking back to source article

Edge Schema

Each edge contains:

Field	Description
source / target	Connected entity names
weight	Relation strength (float)
description	Relationship description
keywords	Associated keywords
source_id	Source chunk ID

API Reference

list_graphs() → list[str]

Return names of all 6 GraphML files.

from UniD3 import list_graphs
list_graphs()
# → ["UniD3_L1T1", "UniD3_L1T2", "UniD3_L2T1", ...]

query_entities(entities, graph_names=None) → list[dict]

Look up one or more entities by name (case-insensitive).

from UniD3 import query_entities

# Single entity
query_entities("RESPIRATORY DISEASES")

# Multiple entities
query_entities(["CALVES", "INFLAMMATION MODULATION"])

# Restrict to specific graph
query_entities("CALVES", graph_names=["UniD3_L1T1"])

Returns list of dicts: {entity, entity_type, description, source_id, graph}

get_neighbors(entity, graph_names=None) → list[dict]

Get all direct neighbors and connecting edge info for an entity.

from UniD3 import get_neighbors
get_neighbors("CALVES")

Returns list of dicts: {graph, neighbor: {entity, entity_type, description, source_id}, edge: {source, target, weight, description, keywords, source_id}}

search_by_type(entity_type, graph_names=None, limit=50) → list[dict]

Filter entities by type.

from UniD3 import search_by_type
search_by_type("DISEASE", limit=10)
search_by_type("DRUG", graph_names=["UniD3_L1T1"])

search_by_keyword(keyword, graph_names=None, limit=50) → list[dict]

Substring match over entity names and descriptions.

from UniD3 import search_by_keyword
search_by_keyword("inflammation")
search_by_keyword("cancer", limit=20)

Typical Workflow

from UniD3 import query_entities, get_neighbors, search_by_type

# Step 1: Find a drug entity
hits = query_entities("ASPIRIN")

# Step 2: Explore its neighborhood (related diseases, targets, etc.)
neighbors = get_neighbors("ASPIRIN")

# Step 3: Filter neighbors by type
diseases = [n for n in neighbors if n["neighbor"]["entity_type"] == "DISEASE"]