Pearl Causal Hierarchy
Every relationship in Galen's knowledge graph is annotated with a causal layer from Judea Pearl's three-level framework. This annotation tells you not just that two entities are related, but how strong the causal evidence is.
Layer 1: AssociationSeeing
Statistical correlations observed in data. “Patients with mutation X tend to have outcome Y.” These are observational — they don't tell you whether X causes Y, or whether some confounding factor Z causes both.
Sources: Literature mining, co-occurrence analysis, STRING protein interactions, gene set membership.
API question: P(Y | X) — What is the probability of Y given that we see X?
Layer 2: InterventionDoing
Evidence from experimental interventions. “When we knock out gene X with CRISPR, cell viability drops by 80%.” This is causal evidence — we intervened and measured the effect.
Sources: DepMap CRISPR screens, GDSC drug sensitivity, ChEMBL bioactivity assays, cBioPortal driver gene analysis.
API question: P(Y | do(X)) — What happens to Y if we intervene on X?
Layer 3: CounterfactualImagining
Evidence validated through counterfactual reasoning using Structural Causal Models. “If EGFR had NOT been inhibited, what would the tumor response have been?” L3 relationships have been tested against multiple independent data sources with consistent causal directionality.
Sources: SCM counterfactual validation, cross-evidence convergence across 2+ independent experimental provenances.
API question: P(Y_x | X', Y') — What would Y have been if X were different?
Why it matters
Most biomedical APIs only provide L1 (association) data. Galen is the first cancer API to provide all three layers, enabling:
- L1: Literature review, hypothesis generation, biomarker discovery
- L2: Drug target validation, essential gene identification, treatment selection
- L3: Resistance prediction, treatment failure analysis, personalized counterfactual reasoning
Next: Evidence Quality
How confidence scores, grounded sources, and evidence quality ratings work in API responses.
Read about evidence quality →