Novo Nordisk®
From knowledge graphs via
Lego bricks to scientific
conversations
Dennis Madsen
Roland Hangelbroek
Vinay Jethava
NN GraphDay May 7, 2024
Novo Nordisk®
From knowledge graphs via
Lego bricks to scientific
conversations
Dennis Madsen
Roland Hangelbroek
Vinay Jethava
NN GraphDay May 7, 2024
Novo Nordisk®
Our knowledge graph principles
Why Clear purpose with KG
Relevant Pipelines to update
Interoperability Entities and ontologies
Novo Nordisk®
Named Entity Recognition
Entity Linking
Relation Extraction
NovoLinker
Novo Nordisk®
Named Entity Recognition
Novo Nordisk®
Entity Linking
Ontology Vocabulary
Novo Nordisk®
Relation Extraction
Novo Nordisk®
Currently in NovoLinker - Entities
• Genes / Proteins – Linked to HGNC, Interpro and
Protein Ontology
• Chemicals & Drugs – Linked to Chebi and NCIT
• Diseases & Phenotypes – Linked to MONDO and
HPO
• Cell lines – Linked to cellosaurus
• Cell types – Linked to cell type ontology (CL)
• Geographical locations – Linked to Geonames
• Sequence features – Linked to sequence ontology
• Organizations – Linked to Wikidata, Crunchbase,
ROR
• Organisms - Linked to NCBITaxon
• Anatomy – Linked to Uberon
Gene ontology:
• Biological process
• Molecular function
• Cell component
Others:
• Assays, medical procedures, devices,
miRNA, sequence variants, persons
Novo Nordisk®
Currently in NovoLinker – Relations
• Protein – protein interactions
• Chemical – protein interactions
• Adverse drug effects
• Gene – disease relations
• Drug– drug interactions
• Causal relations
Novo Nordisk®
Use cases | Patents
 Long complex documents
 Rich relationships
• Patent Families
• CPC/IPCR classification hierarchy
• References (Patents/Articles)
• Inventors/Owners
 Additional information (figures, tables, sequences)
Novo Nordisk®
Use cases | Patents
11
• Basic schema
• Text-based embeddings
• Node embeddings
• Entity detection using
Novo Linker
• Enabling agentic
interaction with the
patent-graph
Novo Nordisk®
Patents (VNJE)
anansi (RLHB)
(Literature, news, conference notes,
pharma pipelines)
N L
Computational
Biology (NYYL)
N L
Clinical Data (KEGS)
NNRCO
Screening Data (VMNZ)
N
L
N
L
Lego bricks
Novo Nordisk®
Chat with data
Novo Nordisk®
14
Novo Nordisk®
LLM Functions - API
Patents (VNJE)
anansi (RLHB)
(Literature, news, conference
notes, pharma pipelines)
N L
Computational
Biology (NYYL)
N L
Clinical Data (KEGS)
NNRCO
Screening Data
(VMNZ)
N
L
N
L
A
P
I
Novo Nordisk®
Use cases
Novo Nordisk®

Adobe Acrobat Reader DC 2025.001.20458 free

  • 1.
    Novo Nordisk® From knowledgegraphs via Lego bricks to scientific conversations Dennis Madsen Roland Hangelbroek Vinay Jethava NN GraphDay May 7, 2024
  • 2.
    Novo Nordisk® From knowledgegraphs via Lego bricks to scientific conversations Dennis Madsen Roland Hangelbroek Vinay Jethava NN GraphDay May 7, 2024
  • 3.
    Novo Nordisk® Our knowledgegraph principles Why Clear purpose with KG Relevant Pipelines to update Interoperability Entities and ontologies
  • 4.
    Novo Nordisk® Named EntityRecognition Entity Linking Relation Extraction NovoLinker
  • 5.
  • 6.
  • 7.
  • 8.
    Novo Nordisk® Currently inNovoLinker - Entities • Genes / Proteins – Linked to HGNC, Interpro and Protein Ontology • Chemicals & Drugs – Linked to Chebi and NCIT • Diseases & Phenotypes – Linked to MONDO and HPO • Cell lines – Linked to cellosaurus • Cell types – Linked to cell type ontology (CL) • Geographical locations – Linked to Geonames • Sequence features – Linked to sequence ontology • Organizations – Linked to Wikidata, Crunchbase, ROR • Organisms - Linked to NCBITaxon • Anatomy – Linked to Uberon Gene ontology: • Biological process • Molecular function • Cell component Others: • Assays, medical procedures, devices, miRNA, sequence variants, persons
  • 9.
    Novo Nordisk® Currently inNovoLinker – Relations • Protein – protein interactions • Chemical – protein interactions • Adverse drug effects • Gene – disease relations • Drug– drug interactions • Causal relations
  • 10.
    Novo Nordisk® Use cases| Patents  Long complex documents  Rich relationships • Patent Families • CPC/IPCR classification hierarchy • References (Patents/Articles) • Inventors/Owners  Additional information (figures, tables, sequences)
  • 11.
    Novo Nordisk® Use cases| Patents 11 • Basic schema • Text-based embeddings • Node embeddings • Entity detection using Novo Linker • Enabling agentic interaction with the patent-graph
  • 12.
    Novo Nordisk® Patents (VNJE) anansi(RLHB) (Literature, news, conference notes, pharma pipelines) N L Computational Biology (NYYL) N L Clinical Data (KEGS) NNRCO Screening Data (VMNZ) N L N L Lego bricks
  • 13.
  • 14.
  • 15.
    Novo Nordisk® LLM Functions- API Patents (VNJE) anansi (RLHB) (Literature, news, conference notes, pharma pipelines) N L Computational Biology (NYYL) N L Clinical Data (KEGS) NNRCO Screening Data (VMNZ) N L N L A P I
  • 16.
  • 17.