Fix: Neo4j 5.26.0 (APOC available) + correct graphrag schema from seeder

Dockerfile

@@ -44,7 +44,7 @@ RUN curl -fsSL https://deb.nodesource.com/setup_20.x | bash - \
     && rm -rf /var/lib/apt/lists/*
 
 # ── Neo4j Community 2026.04.0 ─────────────────────────────────────────────────
-ENV NEO4J_VERSION=2026.04.0
+ENV NEO4J_VERSION=5.26.0
 ENV NEO4J_HOME=/opt/neo4j
 ENV PATH="${NEO4J_HOME}/bin:${PATH}"
 

backend/graphrag.py

@@ -57,25 +57,41 @@ _CYPHER_GENERATION_TEMPLATE = """You are an expert Neo4j Cypher query writer for
 Schema:
 {schema}
 
-Node property conventions (IMPORTANT — use these exact property names and value formats):
-- Patient: id (e.g. "P-001"), name, age (integer), sex ("M"/"F"), ethnicity, city, state, ecog_score (integer)
-- Trial: id (NCT id), title, condition (lowercase, e.g. "breast cancer"), phase, status, sponsor
-- Diagnosis: id, name (e.g. "Breast Cancer"), icd10 (e.g. "C50")
-- Biomarker: id (e.g. "HER2_POS", "EGFR_MUT", "BRCA1_MUT", "PD_L1_POS"), name (e.g. "HER2 Positive", "EGFR Mutation")
-- Medication: id (e.g. "TAMOXIFEN"), name (e.g. "Tamoxifen")
-- StudySite: id, name, city, state, lat, lon, trials (integer), enrolled (integer), capacity (integer)
+Node labels and their exact property names:
+- Patient: id (e.g. "P_C50_000001"), name, age (integer), sex ("MALE"/"FEMALE"), ecog (integer 0-3),
+           condition (lowercase, e.g. "breast cancer"), stage ("I"/"II"/"III"/"IV"),
+           city, state, ethnicity, insurance, icd10_prefix,
+           biomarkers (list of biomarker ids), medications (list of drug names),
+           comorbidities (list), prior_chemo (boolean), prior_radiation (boolean),
+           prior_surgery (boolean), prior_lines_of_therapy (integer), source
+- Trial: id (NCT id, e.g. "NCT04567890"), title, condition (lowercase), phase, status,
+         brief_summary, eligibility_criteria, min_age, max_age, sex, enrollment,
+         start_date, completion_date, sponsor, location_count, source
+- Diagnosis: code (ICD-10, e.g. "C50.919"), name (e.g. "Malignant neoplasm of breast"), source
+- Biomarker: id (e.g. "HER2_POS"), name (e.g. "HER2 Positive"), gene (e.g. "ERBB2"), loinc, source
+- Medication: rxcui, name, tty, generic_name, source
+- StudySite: facility, city, state, country, lat, lon, source
+- ConditionNode: name (e.g. "breast cancer")
+- Publication: pmid, title, journal, pub_date, authors, source
 
 Relationships:
-- (Patient)-[:ELIGIBLE_FOR {{score: float}}]->(Trial)
+- (Patient)-[:ELIGIBLE_FOR {{score: float, matched_at: datetime}}]->(Trial)
 - (Patient)-[:HAS_DIAGNOSIS]->(Diagnosis)
 - (Patient)-[:HAS_BIOMARKER]->(Biomarker)
-- (Patient)-[:TAKES_MEDICATION]->(Medication)
-- (Trial)-[:LOCATED_AT]->(StudySite)
+- (Trial)-[:CONDUCTED_AT]->(StudySite)
+- (ConditionNode)-[:HAS_TRIAL]->(Trial)
+- (Diagnosis)-[:MAPS_TO_CONDITION]->(ConditionNode)
+- (Biomarker)-[:RELEVANT_TO]->(ConditionNode)
+- (Biomarker)-[:MAY_QUALIFY_FOR]->(Trial)
+- (Publication)-[:SUPPORTS_RESEARCH_ON]->(ConditionNode)
 
 Rules:
-- For biomarker lookups, use the `id` property with uppercase underscore format, e.g. `{{id: 'HER2_POS'}}`
-- For condition lookups on Trial nodes, use lowercase: `t.condition = 'breast cancer'`
-- Always use relationship pattern (Patient)-[:ELIGIBLE_FOR]->(Trial) to find eligible patients
+- Biomarker lookups use the `id` property: `{{id: 'HER2_POS'}}`
+- Diagnosis lookups use `code` (not `id`): `{{code: 'C50.919'}}`
+- Medication lookups use `rxcui` or `name` (not `id`)
+- Condition lookups on Trial nodes use lowercase: `t.condition = 'breast cancer'`
+- Patient-to-trial eligibility: `(p:Patient)-[:ELIGIBLE_FOR]->(t:Trial)`
+- ecog property on Patient is `ecog` (integer), NOT `ecog_score`
 - Limit results to 25 unless asked for more
 
 Question: {question}