petagraph

is a Knowledge Graph.

Petagraph is a large-scale biomedical knowledge graph that encompasses over 32 million nodes and 118 million relationships, integrating multi-omics and clinical data. Built on the Unified Biomedical Knowledge Graph (UBKG), Petagraph leverages more than 180 ontologies and standards to embed millions of quantitative genomics data points within a richly connected annotation environment.

Evaluation
Domains

health, biomedical, genomics

License

UMLS License Agreement

Homepage

petagraph

Repository

GitHub

Infores ID

Unknown

FAIRsharing ID

Unknown

Product Summary

Products

From this Resource
ID Name URL Category Format Description
petagraph.graph Petagraph Knowledge Graph (Neo4J) ubkg-downloads.xconsortia.org GraphProduct ❔ A comprehensive multi-omics biomedica...

Details

Petagraph: A Large-Scale Biomedical Knowledge Graph

Petagraph is a comprehensive biomedical knowledge graph that encompasses over 32 million nodes and 118 million relationships. It provides a cohesive data environment that enables users to efficiently analyze, annotate, and discern relationships within and across complex multi-omics datasets.

Overview

Constructed on the foundation of the Unified Biomedical Knowledge Graph (UBKG), Petagraph leverages more than 180 ontologies and standards to support a diversity of genomics data types. The graph significantly extends UBKG by adding over 12 million nodes (an increase of almost 44%) and more than doubling the number of relationships from 52 million to 118 million.

Petagraph was originally developed as a resource for rapid feature selection to identify, annotate, and explore gene candidates for human diseases, but has expanded to serve as a base module for user-customized knowledge graphs supporting various biomedical use cases.

Architecture and Components

Petagraph has five main data-specific node types:

  • Concept nodes: The central backbone organizing specific conceptual meanings
  • Code nodes: Identifiers for references in particular ontologies or standards
  • Term nodes: Human-readable annotations about linked Code nodes
  • Definition nodes: Provide definitions for Concept nodes
  • Semantic nodes: Provide wider semantic type classifications for Concepts

The knowledge graph contains 1,861 distinct edge types, with the majority falling into the category of Concept-to-Concept edges, which are bidirectional and form the primary traversable component of the graph.

Data Integration

Petagraph integrates 21 sources of supporting omics and annotation data, including:

  1. Genomic Data:
    • GTEX expression and eQTL data
    • 4D Nucleome program data
    • Single-cell fetal heart data
    • ClinVar genetic variants
  2. Mapping Data:
    • Human-to-mouse ortholog mappings
    • Human gene-to-phenotype mappings
    • Human-to-mouse phenotype mappings
    • Mouse gene-to-phenotype mappings
    • Molecular signatures database
  3. Protein Interaction Data:
    • STRING (protein-protein interactions)
    • GlyGen (glycosylation information)
  4. Chemical-Gene Interactions:
    • Connectivity MAP (CMAP)
    • LINCS L1000

Installation Options

Petagraph can be installed in two ways:

  1. Neo4j dump file (preferred method):
    • Requires a UMLS license and API key
    • Uses a pre-built database dump (4.5 GB) that can be installed on local machines with at least 16 GB of memory
    • Simple installation through Neo4j Desktop
  2. Build from source:
    • Provides more customization options
    • Requires generating UMLS CSVs and UBKG CSVs
    • More time-intensive but allows for custom data integration

Use Cases

Petagraph supports diverse biomedical research applications, including:

  • Identifying genomic features functionally linked to genes or diseases
  • Linking across genetics data between human and animal models
  • Linking transcriptional perturbations by compound in tissues of interest
  • Identifying cell types from single-cell data associated with diseases or genes
  • Applying machine learning methods for link prediction and node property prediction
  • Supporting integration with Large Language Models (LLMs) to enhance biomedical data analysis

Access and Requirements

  • A UMLS license is required to access and use Petagraph
  • The database dump is available at UBKG Downloads
  • Documentation and source code are available on GitHub
  • The graph database requires Neo4j (version 5.x recommended)

Future Development

Ongoing work focuses on:

  • Expanding Petagraph’s capabilities
  • Enhancing automated validation techniques
  • Developing standardized benchmarks for knowledge graph comparisons
  • Integration with Large Language Models
  • Supporting the NIH Common Fund Data Ecosystem Data Distillery Project

Evaluation

Is this information incorrect or incomplete? Request an update.

Created: March 09, 2025 | Last modified: December 13, 2025