Example: Target Characterisation — EGFR

Status: Illustrative. Same caveat as examples/01-variant-triage/: the numbers below are consistent with what the server emits when its upstream clients (Open Targets, ChEMBL, UniProt, AlphaFold DB, gnomAD) return data we have unit tests for. They have NOT been verified against live API responses for this exact target. End-to-end validation lands in v1.2.0; see STATUS.md.

User prompt

Characterise EGFR as a drug target for me. How tractable is it, what approved drugs hit it, what's the gnomAD constraint, and is there an AlphaFold structure I can look at?

What the model calls

{
  "tool": "assess_target_druggability",
  "params": {
    "gene_symbol": "EGFR",
    "include_drugs": true,
    "include_alphafold": true,
    "include_constraint": true
  }
}

What the server returns (schema-accurate; values illustrative)

{
  "input": {"gene_symbol": "EGFR"},
  "uniprot": {
    "uniprot_id": "P00533",
    "protein_name": "Epidermal growth factor receptor",
    "length": 1210,
    "subcellular_location": ["Cell membrane", "Endosome", "Nucleus"],
    "function_summary": "Receptor tyrosine kinase binding EGF family ligands. Activates ERK, AKT, JAK-STAT cascades."
  },
  "open_targets": {
    "target_id": "ENSG00000146648",
    "association_score": 0.92,
    "diseases_top": [
      {"id": "EFO_0003060", "label": "Non-small cell lung carcinoma", "association": 0.99},
      {"id": "EFO_0000305", "label": "Breast carcinoma", "association": 0.81},
      {"id": "MONDO_0008549", "label": "Glioblastoma multiforme", "association": 0.71}
    ],
    "tractability": {
      "labels": ["Approved Drug", "Advanced Clinical", "Discovery_small_molecule", "Antibody clinical"],
      "small_molecule": true,
      "antibody": true,
      "modality_score_sm": 4,
      "modality_score_ab": 4
    }
  },
  "chembl": {
    "chembl_id": "CHEMBL203",
    "approved_drugs_count": 14,
    "approved_drugs_top": [
      {"chembl_id": "CHEMBL554", "name": "GEFITINIB", "first_approval": 2003},
      {"chembl_id": "CHEMBL553", "name": "ERLOTINIB", "first_approval": 2004},
      {"chembl_id": "CHEMBL1173655", "name": "OSIMERTINIB", "first_approval": 2015},
      {"chembl_id": "CHEMBL1801", "name": "CETUXIMAB", "first_approval": 2004}
    ]
  },
  "gnomad_constraint": {
    "loeuf": 0.45,
    "mis_z": 3.21,
    "syn_z": 0.78,
    "constraint_note": "LOEUF 0.45 — moderately constrained; loss-of-function tolerance is limited but not absent."
  },
  "alphafold": {
    "uniprot_id": "P00533",
    "model_url": "https://alphafold.ebi.ac.uk/files/AF-P00533-F1-model_v4.pdb",
    "plddt_mean": 73.6,
    "plddt_note": "High mean confidence; kinase domain (residues 712-979) is well-resolved (pLDDT > 90).",
    "n_residues": 1210
  },
  "druggability_tier_draft": {
    "tier": "HOT",
    "score": 5,
    "score_components": {
      "approved_drugs_>=3": "+3",
      "tractability_small_mol": "+2",
      "alphafold_plddt_>=70": "+1",
      "loeuf_<0.35": "0 (LOEUF 0.45, no penalty)"
    },
    "rationale": "Strong drug precedent (14 approved drugs), tractable small-molecule and antibody modalities, well-resolved kinase domain. Tier is a HEURISTIC, not a validated prediction — see LIMITATIONS.md L2."
  },
  "sources_cited": ["UniProt P00533", "Open Targets ENSG00000146648", "ChEMBL CHEMBL203", "gnomAD v4.1.0 constraint table", "AlphaFold DB v4 (P00533)"],
  "report_generated_at": "2026-05-11T14:35:12Z"
}

What Claude says back (paraphrased)

EGFR (P00533) is a textbook "HOT" target by our heuristic:

• Indications: Open Targets ranks it strongest for non-small cell lung carcinoma (association 0.99), breast carcinoma, and glioblastoma.
• Modalities: both small-molecule (gefitinib, erlotinib, osimertinib, …) and antibody (cetuximab) routes are clinically validated — 14 approved drugs total in ChEMBL.
• Tractability: Open Targets labels include Approved Drug, Advanced Clinical, Discovery_small_molecule, and Antibody clinical.
• Constraint: LOEUF 0.45 — moderately constrained. Loss-of- function tolerance is limited but not absent, so toxicity from on-target inhibition is plausible but the drug record shows it's manageable.
• Structure: AlphaFold pLDDT mean 73.6; the kinase domain (712–979) is well-resolved (pLDDT > 90), which is what you want for pocket-finding.
• ⚠ Caveat: The "HOT" tier is a 4-line heuristic, not a calibrated prediction. Treat as a triage signal alongside medicinal-chemistry review.

Reproducing this example for real

Same setup as Example 01.

Files

• transcript.jsonl — MCP exchange.
• expected_output_schema.json — JSON Schema for the report shape.