Analyzing a Healthcare Knowledge Graph with Cypher and Graph Data Science

This assignment felt like a real analytics engagement: incomplete documentation, a complex healthcare graph, and the need to make method choices that I could justify.

Assignment Focus

I worked with a FAERS-inspired healthcare dataset restored from a Neo4j dump, then analyzed it in three passes:

1. Cypher-based schema and data exploration
2. Deeper analytical querying and aggregation
3. GDS workflows for structural and similarity analysis

I used Python + Neo4j driver + pandas to keep experiments iterative and reproducible.

What I Built

• A practical EDA workflow for discovering labels, relationship patterns, and property quality
• Analytical Cypher queries for non-trivial domain questions
• GDS projections and algorithm runs aligned to assignment questions (instead of running algorithms blindly)

I documented not just outputs, but why each query/algorithm choice matched the analytical goal.

Key Findings

• Schema discovery first saved time later. Without that step, I would have built analyses on false assumptions.
• Cypher before GDS produced better algorithm decisions because I understood graph semantics first.
• Projection design is the real GDS work: node sets, relationship directions, and weighting determine result quality.
• Driver-based iteration in Python made query refinement much faster than manual-only browser experimentation.

Lessons Learned

This project taught me to treat graph analytics as a staged process: understand the graph, validate assumptions, then apply algorithms with purpose. I also learned how quickly “interesting but wrong” results can appear when projections are underspecified.

Skills I Gained

• Cypher EDA for unfamiliar production-like graph datasets
• GDS projection design and algorithm execution
• Python-driven graph analysis workflow with Neo4j driver + pandas
• Query/result interpretation with explicit methodological tradeoffs

Artifacts

• Notebook: P_Mangoro_C2_assn.ipynb