BiomarkerKB Knowledge Graph (KG)

Repository Structure & Setup

Server Path: /data/shared/KG

This directory contains the BiomarkerKB Knowledge Graph (KG) setup and all supporting documentation with instructions for how to integrate Biomarker Knowledge into the CFDE UBKG. PLease make sure to follow all licenses from UBKG when integrating the biomarker knowledge graph into UBKG. The instructions for integration are below

• README.md – Documentation on running the KG and LLM testing performed so far.

• distribution/ and csv/ – Build configurations and files for constructing the BiomarkerKB KG.

• Supplementary Files:Additional resources supporting the build process.

KG Build Process

The BiomarkerKB KG is constructed by ingesting biomarker data into the CFDE Knowledge Graph (KG) managed by the Data Distillery team

Preliminary steps

Convert json files to nt on the dev.biomarkerkb.org server:

/data/shared/repos/biomarker-data-transformation/nt/json_to_nt.sh

Copy NT files from /data/shared/biomarkerdb/generated/nt into /data/shared/KG/nt on the KVM2 server.

Workflow

1. Biomarker Data Conversion into Biomarker KG

   • Biomarker data is first converted into the OWLNETS_edgelist.txt and OWLNETS_node_metadata.txt formats required by the CFDE KG.
   • Conversion handled using the nt-owlnets-kg-converter.
   • This converts BiomarkerKB data into Ntriples. This format can then be used to generate CSVs files needed to build neo4j knowledge graphs.
   • (available in BiomarkerKG)
   • Generated files are available in:

     /data/shared/repos/ubkg-etl
     

2. Converting Ntriples into CSVs

• Biomarker Ntriples are formatted into CSVs for neo4j knowledge graph
• The generation framework is a suite of Extract-Transform-Load (ETL) scripts that
  • extract assertion data from sources, including OWL files
  • appends assertions to the set of UMLS CSV
• This should allow for an internal BiomarkerKB KG to be developed
• Generation framework instructions are provided by the UBKG team UBKG ETL.
• Repo availabe in KVM2 server:

  /data/shared/repos/ubkg-etl
  

3. Integration into CFDE KG

• Once generated, these files are re-ingested into the CFDE KG.
• Full build and integration instructions are provided by the UBKG team:
  UBKG Build Instructions.
• Repo availabe in KVM2 server:

  /data/shared/repos/ubkg-neo4j
  

4. Data Updates
   • Each time new biomarker data is pushed, the edge list and node files must be recreated and re-ingested into the CFDE KG.

Notes & Requirements

• Building the KG requires significant disk space:

  Free disk space equal to 3–4× the size of the ontology CSVs is recommended.

• Repository copies and scripts are stored here:

  /data/shared/repos/ubkg-neo4j
  

• Supplementary files are available in the biomarker-partnership repo.

BiomarkerKG Use Cases

1. Identify multi-disease biomarkers
   • Retrieve biomarkers that are associated with multiple diseases, enabling the identification of shared biological indicators across disease domains.
2. Identify multi-biomarker diseases
   • Find diseases linked to multiple biomarkers to reveal complex molecular signatures and potential diagnostic panels.
3. Detect biomarkers with multiple measurement contexts
   • Query biomarkers that have distinct LOINC and UBERON identifiers.
   • This highlights biomarkers measurable in different specimen types or anatomical locations, suggesting disease-specific measurement modalities.

CFDE Knowledge Graph (KG) Use Cases

1. Explore drug target and therapy connections
   • From a single biomarker entry, expand to identify related drug targets, therapeutic agents, and associated treatment contexts.
2. Cross-disease treatment discovery
   • Starting from one biomarker–disease association, identify drug therapies linked to that biomarker and expand to find other diseases treated by the same therapy, revealing shared treatment and molecular pathways.

RENCI Knowledge Graph (KG) Use Cases

1. Link metabolite biomarkers to protein sequences
   • Connect metabolite biomarkers with related protein sequences to uncover molecular mechanisms and enzymatic interactions.
2. Integrate patient and laboratory data
   • Retrieve associated patient-level and laboratory data for selected metabolites to support translational and clinical insights.
3. Identify interacting genes and proteins
   • Discover genes and proteins interacting with a given biomarker to elucidate biological pathways, reactions, and post-translational modifications that may underlie disease processes.