Cancer Curation Skill

Overview

Curate cancer and neoplastic disease entries in the dismech knowledge base with a focus on:

• Genetic drivers (oncogenes, tumor suppressors, fusion genes)
• Pathophysiology with causal chains
• Histopathology findings
• Targeted therapies and their molecular targets
• NCIT terms for biomarkers, gene products, and histologic findings

When to Use

• Creating new cancer/neoplasm entries
• Adding genetic driver information (BCR-ABL, KIT, RET, RB1, etc.)
• Structuring pathophysiology as atomic processes with causal links
• Adding histopathology findings (grades, patterns, rosettes)
• Linking treatments to their molecular targets
• Adding CHEBI terms for chemotherapy drugs
• Adding NCIT terms for biomarkers and fusion proteins

When creating a new cancer or neoplasm entry, first run the duplicate preflight from initiate-new-disorder-creation: check the latest origin/main knowledgebase, all PRs, and all issues by MONDO ID, preferred label, and major synonyms. Do not create a separate cancer entry if an existing KB file, PR, or issue already covers the same disease concept.

Granularity: which level gets its own entry?

Follow the cancer granularity ladder in design decisions §3a (docs/explanation/design-decisions.md) — it is the ratified policy and supersedes the older "molecular subtypes as discrete entities" framing in projects/CANCER.md:

1. Default entry level = histologic entity (WHO blue-book / ICD-O level: PDAC, SCLC, DLBCL) or WHO/ICC molecularly defined entity (IDH-wildtype GBM, APL with PML::RARA, NPM1-mutant AML).
2. Biomarker/therapy strata (EGFR-mutant NSCLC, MSI-H CRC, TNBC): default has_subtypes; a separate entry requires ≥2 stratum-specific pathophysiology nodes AND a distinct first-line therapy/diagnostic pathway. When promoted: skos:narrowMatch mondo_mapping if no exact MONDO term exists (file an NTR), record overlap with non-disjoint sibling strata, add the entry to a covering grouping, and leave a pointer subtype in the parent.
3. Variant tiers stay has_subtypes inside the stratum entry unless therapy is variant-specific (KRAS G12C).
4. Stage/metastasis is never an entry — use stages: + conforms_to: "invasion_and_metastasis#...". Do not create Metastatic_X files.
5. Pathways/hallmarks are never entries — modules + groupings.
6. Germline predisposition syndromes follow the Mendelian rules and stay separate from the somatic cancers they predispose to.

Mark the cell of origin (no new slot)

The cell of origin is derived, not stored: put a genetic_context carrying variant_origin: SOMATIC on the pathophysiology node with the initiating lesion, and the cell of origin is that node's cell_types.

- name: BCR-ABL1 Fusion Oncogene Formation
  genetic_context:
    variant_origin: SOMATIC
    functional_impact_category: GAIN_OF_FUNCTION
  cell_types:
  - preferred_term: hematopoietic stem cell
    term:
      id: CL:0000037
      label: hematopoietic stem cell

• role: trigger marks nothing. role is free text with ~90 values in the KB, and the derivation reads structured markers only.
• For a cancer with no host lesion to mark (HPV, H. pylori, asbestos, UV), the marker is the environmental[].influences_mechanisms link carrying environmental_effect: TRIGGERS. It only speaks when no lesion is recorded.
• Do not mark a microenvironment or chronic-inflammation node. Those bind macrophage, Treg and fibroblast, which are where the tumor lives, not where it came from.
• Deriving more than one cell of origin is a lump/split prompt, not a defect.
• NCIT asserts its own cell of origin (NCIT:R104) and transformed cell state (NCIT:R105, the Abnormal Cell branch) per disease, quotable from references_cache/NCIT_*.md. Useful as a cross-check and as evidence; never as the term: of cell_types, which is CL-only.

just check-cancer-origin                  # summary + multi-origin worklist
just list-cancer-origin                   # per-entry census

Worked examples: Chronic_Myeloid_Leukemia, Pancreatic_Ductal_Adenocarcinoma. Full guidance: docs/cancer-cell-of-origin.md.

Cancer-Specific Schema Features

Disease Stages (not Subtypes)

For cancers with disease phases (chronic → accelerated → blast crisis) — and for localized vs. metastatic disease — use stages not has_subtypes:

stages:
- name: Chronic Phase
  description: >-
    Initial indolent phase with <10% blasts. Most patients diagnosed here.
- name: Accelerated Phase
  description: >-
    Transitional phase with 10-19% blasts, additional cytogenetic abnormalities.
- name: Blast Crisis
  description: >-
    Terminal phase resembling acute leukemia with ≥20% blasts.

Use has_subtypes for true molecular/histologic subtypes (KIT-mutant vs PDGFRA-mutant GIST).

Pathophysiology: Atomic Processes with Causal Chains

Split bundled pathway descriptions into atomic processes linked by downstream edges:

pathophysiology:
- name: BCR-ABL1 Fusion Oncogene Formation
  description: >-
    The t(9;22) translocation creates the Philadelphia chromosome with
    constitutively active BCR-ABL1 tyrosine kinase.
  cell_types:
  - preferred_term: hematopoietic stem cell
    term:
      id: CL:0000037
      label: hematopoietic stem cell
  gene_products:
  - preferred_term: BCR-ABL1 fusion protein
    term:
      id: NCIT:C16325
      label: BCR/ABL1 Fusion Protein
  downstream:
  - target: Constitutive Tyrosine Kinase Activation
    description: BCR-ABL1 exhibits ligand-independent kinase activity

- name: Constitutive Tyrosine Kinase Activation
  description: >-
    BCR-ABL1 activates RAS-MAPK, PI3K-AKT, and JAK-STAT pathways.
  biological_processes:
  - preferred_term: protein tyrosine kinase activity
    modifier: INCREASED
    term:
      id: GO:0004713
      label: protein tyrosine kinase activity
  downstream:
  - target: Uncontrolled Myeloid Proliferation
    description: Activated signaling drives excessive myeloid expansion

Key principles:

1. One mechanism per entry (not "Oncogenic Signaling" combining MAPK + PI3K)
2. Use downstream to connect cause → effect
3. Include cell_types, biological_processes, locations as appropriate
4. Add gene_products for fusion proteins/oncoproteins (NCIT terms)

Histopathology Section

Use the dedicated histopathology section for microscopic findings:

histopathology:
- name: Flexner-Wintersteiner Rosettes
  finding_term:
    preferred_term: Flexner-Wintersteiner rosette
    term:
      id: HP:0031927
      label: Flexner-Wintersteiner rosette
  frequency: FREQUENT
  diagnostic: true
  description: >-
    Characteristic rosettes with central lumen surrounded by tumor cells
    showing photoreceptor differentiation. Pathognomonic for retinoblastoma.

- name: Spindle Cell Morphology
  finding_term:
    preferred_term: Spindle Cell Pattern
    term:
      id: NCIT:C53643
      label: Spindle Cell Pattern
  frequency: VERY_FREQUENT
  description: >-
    Elongated cells arranged in fascicles. Typical of KIT-mutant GIST.

Histopathology term sources:

• NCIT:C35867 (Morphologic Finding) - patterns, dysplasia, necrosis
• NCIT:C18000 (Histologic Grade) - Fuhrman, Nottingham, GIST grades
• HP:0025461 (Abnormal cell morphology) - rosettes, inclusion bodies

Biomarkers (NCIT)

Add biomarker_term to biochemical entries:

biochemical:
- name: BCR-ABL1 Fusion Transcript
  biomarker_term:
    preferred_term: BCR-ABL1 fusion protein
    term:
      id: NCIT:C36715
      label: BCR-ABL1 Fusion Protein Expression
  notes: >-
    RT-PCR or FISH detection is diagnostic and used for molecular monitoring.

Gene Products (NCIT)

Add gene_products to pathophysiology for fusion proteins and oncoproteins:

gene_products:
- preferred_term: BCR-ABL1 fusion protein
  term:
    id: NCIT:C16325
    label: BCR/ABL1 Fusion Protein

Look up NCIT gene product terms:

uv run runoak -i sqlite:obo:ncit descendants NCIT:C26548 --predicates rdfs:subClassOf | grep -i "fusion\|kinase"

Therapeutic Agents (CHEBI)

Add therapeutic_agent to treatments with CHEBI terms:

treatments:
- name: Imatinib
  description: First-generation TKI targeting BCR-ABL1.
  treatment_term:
    preferred_term: pharmacotherapy
    term:
      id: NCIT:C15986
      label: Pharmacotherapy
    therapeutic_agent:
    - preferred_term: imatinib
      term:
        id: CHEBI:45783
        label: imatinib

Ontology Lookups

NCIT Gene Products

# Search for fusion proteins
uv run runoak -i sqlite:obo:ncit search "fusion protein"

# Check specific term
uv run runoak -i sqlite:obo:ncit info NCIT:C16325

# Get descendants of Gene Product
uv run runoak -i sqlite:obo:ncit descendants NCIT:C26548 --predicates rdfs:subClassOf | head -50

NCIT Histopathology

# Morphologic findings
uv run runoak -i sqlite:obo:ncit descendants NCIT:C35867 --predicates rdfs:subClassOf | head -50

# Histologic grades
uv run runoak -i sqlite:obo:ncit descendants NCIT:C18000 --predicates rdfs:subClassOf | head -50

CHEBI Drugs

uv run runoak -i sqlite:obo:chebi search "imatinib"
uv run runoak -i sqlite:obo:chebi info CHEBI:45783

HP Rosettes and Cell Morphology

uv run runoak -i sqlite:obo:hp descendants HP:0025461 --predicates rdfs:subClassOf | grep -i rosette

Common NCIT Terms

Fusion Proteins

Term	ID	Cancer
BCR/ABL1 Fusion Protein	NCIT:C16325	CML
Mast/Stem Cell Growth Factor Receptor Kit	NCIT:C17328	GIST
Proto-Oncogene Tyrosine-Protein Kinase Receptor Ret	NCIT:C18539	MTC

Biomarkers

Term	ID	Use
BCR-ABL1 Fusion Protein Expression	NCIT:C36715	CML monitoring
Calcitonin	NCIT:C2281	MTC tumor marker
Carcinoembryonic Antigen	NCIT:C16384	MTC, colorectal

Histologic Grades

Term	ID	Cancer
Fuhrman Nuclear Grade	NCIT:C62411	Renal cell carcinoma
GIST Histologic Grade	NCIT:C160731	GIST
Nottingham Grade	NCIT:C138986	Breast cancer

Morphologic Findings

Term	ID	Finding
Spindle Cell Pattern	NCIT:C53643	Elongated cell morphology
Low Mitotic Activity	NCIT:C35961	<5 mitoses/50 HPF
Fleurette Formation	NCIT:C35950	Retinoblastoma differentiation

Common CHEBI Drug Terms

TKIs

Drug	CHEBI ID
imatinib	CHEBI:45783
dasatinib	CHEBI:49375
nilotinib	CHEBI:52172
ponatinib	CHEBI:78543
sunitinib	CHEBI:38940

Chemotherapy

Drug	CHEBI ID
carboplatin	CHEBI:31355
vincristine	CHEBI:27375
etoposide	CHEBI:4911
doxorubicin	CHEBI:28748
cyclophosphamide	CHEBI:4026

Cancer Tiers (from CANCER.md project)

Tier 1: Paradigmatic Single-Gene Drivers

Clear driver mutations with targeted therapies:

• CML (BCR-ABL1) → imatinib
• GIST (KIT/PDGFRA) → imatinib/avapritinib
• MTC (RET) → selpercatinib
• ccRCC (VHL) → belzutifan
• Retinoblastoma (RB1) → two-hit paradigm
• Ewing Sarcoma (EWS-FLI1) → fusion transcription factor

Tier 2: Hereditary Cancer Syndromes

Germline mutations with defined progression:

• Li-Fraumeni (TP53)
• FAP (APC)
• VHL disease
• MEN2 (RET)
• HBOC (BRCA1/2)
• NF1

Tier 3: Molecular Subtype Cancers

• AML (FLT3, NPM1, IDH1/2)
• Colorectal (CMS subtypes)
• Breast (HER2+)
• Melanoma (BRAF)
• Glioblastoma (IDH-mutant vs wildtype)

Validation Workflow

# 1. Schema validation
just validate kb/disorders/MyCancer.yaml

# 2. Term validation (NCIT, CHEBI, HP, CL, GO)
just validate-terms kb/disorders/MyCancer.yaml

# 3. Snippet check against the reference cache (seconds, offline)
just count-verified-snippets kb/disorders/MyCancer.yaml

# 4. Before opening the PR: the batched sweep CI runs (slow — once, not per edit)
just validate-disorders kb/disorders/MyCancer.yaml

# 5. Full QC
just qc

Example: Creating a New Cancer Entry

1. Run duplicate preflight using initiate-new-disorder-creation Step 1. Confirm the target is absent from the latest knowledgebase and not already covered by any PR or issue.

2. Start with deep research (if available):

   # Check for existing research
   ls research/*MyCancer*.md
   

3. Create YAML structure:

   • name, description, categories, parents
   • disease_term (MONDO)
   • has_subtypes or stages (as appropriate)
   • pathophysiology (atomic, with causal chains)
   • histopathology (if relevant)
   • phenotypes
   • biochemical (with biomarker_term)
   • genetic
   • treatments (with treatment_term + therapeutic_agent)

4. Add evidence for key claims (PMID references)

5. Validate all term bindings

Integration with Other Skills

• dismech-terms: For general ontology term lookups (HP, CL, GO, MONDO)
• dismech-references: For evidence validation and PMID snippet checking
• dismech-compliance: For checking field coverage

Project Tracking

Cancer curation is tracked in projects/CANCER.md. Update status after completing entries.