Drugs, their gene targets, side effects, indications and bioactivity, integrated from five open pharmacology sources. The graph turns clinical-decision-support questions — "what's the side-effect burden of this drug?", "which two drugs compete at the same target?" — into single traversals.
cd case_studies/drug-interactions && ./run.sh # validate every query
RECORD=1 ./run.sh # also regenerate demo.gifScale: 244,783 nodes · 387,577 edges (from an 8 MB snapshot)
| Node label | Count | Key properties |
|---|---|---|
| Bioactivity | 208,025 | target_name, pchembl_value, standard_type |
| Drug | 19,842 | name, drugbank_id, cas_number |
| Gene | 6,449 | gene_name |
| SideEffect | 5,858 | name, meddra_id |
| Indication | 2,844 | name, meddra_id |
| AdverseEvent | 1,765 | term |
Relationships: BIOACTIVITY_TARGET (180K), HAS_SIDE_EFFECT (139K),
INTERACTS_WITH_GENE (35K), HAS_ADVERSE_EVENT (18K), HAS_INDICATION (15K).
See queries.cypher: side-effect burden → busiest drug-target
genes (CYP enzymes surface naturally) → polypharmacy risk (drug pairs sharing
the most gene targets — a two-hop graph join that's the backbone of interaction
checking) → most widespread side effects → most-indicated drugs → adverse-event
volume. Every query returns real rows (DoD-gated).
Sources: DrugBank, DGIdb, SIDER, ChEMBL, OpenFDA (see the
druginteractions-kg
loader for per-source licensing). Snapshot druginteractions.sgsnap on release
kg-snapshots-v5
(sha256 pinned in case.env).