Surfacing Hidden Chemical Risks in Pharma – GraphTalk Boston

Kim Adler, Manager & Technical Product Owner at Pfizer, shares how her team built a knowledge graph to track chemicals of concern across Pfizer’s pharmaceutical supply chain – cutting response time from days to minutes.
When regulators ban a substance, Pfizer must rapidly identify every product it touches. Traditional tabular data makes this nearly impossible: chemicals enter manufacturing genealogies at multiple stages, and inconsistent naming conventions (like “Red Dye No. 3” vs. “Erythrosine”) cause keyword searches to miss critical matches.

To solve this, the team modeled material movements in a graph database, enabling a single Cypher query with variable-length path matching to trace any ingredient downstream – regardless of how many steps removed it is from the finished drug.

For synonym handling, they use vector embeddings from biomedical language models (BioBERT) to enable semantic search via cosine similarity. And with a GraphRAG approach, stakeholders can ask natural language questions and get full supply chain paths in return – with fewer hallucinations and full interpretability.