Discovery ModuleHypothesis generation · Phase 3 API liveStructurally separate from clinical modules
What is structurally plausible, but not yet documented?
Every clinical PetriDish module answers what is documented. Discovery answers what is plausible — ranking undocumented phytochemical-enzyme candidate edges and estimating pathogenicity for variants absent from gnomAD. Backed by graph topology in Phase 1; GraphSAGE (Neo4j GDS) swap-in behind the same contract when compute is allocated.
⚠️
HYPOTHESIS GENERATION — NOT VALIDATED FACT
Every output on this page is a computational prediction scored on graph topology, not a literature-sourced or clinically confirmed relationship. Predictions are stored in schema-separated
PREDICTED_TARGETSedges (never merged with documented TARGETS) and predicted_effectproperties (never merged with documented effect). Top-ranked predictions require literature cross-check before any external or wet-lab use.Phytochemical → Enzyme candidate edgesLINK PREDICTION · v0
Ranks undocumented phytochemical-enzyme pairs by topology-heuristic plausibility. Leave "phytochemical" blank to rank across the whole substrate, or pick one to focus. Documented literature edges are excluded from output.
Why Discovery is a separate module
Schema separation
PREDICTED_TARGETS edges and predicted_effect properties are never merged with their documented counterparts. Zero code path where a prediction can be returned as a fact.
No orchestrator wiring
Discovery output does not route through the orchestration layer. It is queryable only via /discovery/*, never via IndoPGx / CardioRisk / AyurBridge read paths.
Mandatory acknowledgement
Every prediction endpoint refuses to respond unless the caller sets experimental_include_predictions=true — an explicit acknowledgement that the output is not a validated fact.
Swappable backend
Phase 1 scoring is a documented topology heuristic (model_version topology-heuristic-v0). GraphSAGE + Neo4j GDS pipeline per PRD §5 plugs in behind the same contract when compute is allocated.