Cancer Curation SOP: MONDO/NCIT Usage and Subtype Axes

Curator-facing SOP for using MONDO and NCIT consistently across cancer entries in dismech: disease anchors, mappings, histopathology, treatments, and subtype schemes. Resolves #1198 as the cancer-specific extension of the broader MONDO-anchoring SOPs in #795 and #1007.

The mechanics of cancer pathophysiology, histopathology, and biomarker curation live in .claude/skills/cancer-curator/SKILL.md. This document covers only the modeling-decision questions that recur across oncology entries: when to split, when to anchor, when to ground subtypes in NCIT.

Curation unit: one entry per mechanism graph

A dismech entry is a curation unit centered on one coherent mechanism graph. Ontology classes are not themselves the unit of curation.

A cancer ontology subclass should become a separate dismech entry only when its causal program is genuinely distinct — different cell of origin, different primary driver, different signaling cascade.

Split into separate entries when: - different driver class or cell of origin (Alveolar_Rhabdomyosarcoma PAX3/7-FOXO1 fusion vs. Embryonal_Rhabdomyosarcoma) - distinct pathway-defined groups (Medulloblastoma_SHH_Activated vs. Medulloblastoma_WNT_Activated) - distinct molecular contexts that drive different therapy programs (IDH_Mutant_Astrocytoma vs. IDH_Mutant_Oligodendroglioma)

Do not split — keep as a faceted subtype within one entry — when the difference is: - histologic grade or pattern within the same mechanism (Favorable Histology vs. Anaplastic Wilms — same nephroblastoma program; anaplasia adds TP53 loss as a modifier, not a new cell-of-origin program) - anatomic facet of the same disease (Bilateral vs. Unilateral Wilms) - stage, age qualifier, or laterality alone

If you are unsure: prefer one entry with multiple subtype axes over fragmenting the mechanism graph across files.

`disease_term`: MONDO-first

Always anchor disease_term to MONDO when a MONDO:0000001 (disease) descendant exists:

disease_term:
  preferred_term: Wilms tumor
  term:
    id: MONDO:0006058
    label: Wilms tumor

When MONDO has only a broad parent and NCIT has the precise oncology concept: keep MONDO as disease_term and add NCIT to mappings (see next section).
When no MONDO disease term exists at all: follow the #795 SOP (skos:closeMatch to the closest MONDO term + Mondo NTR).

The point is to keep one stable disease anchor type across the KB so that downstream tools (G2P, MONDO comparators) do not have to special-case oncology.

Disease-level `mappings`: routinely include NCIT

Cancer entries should populate all four mapping slots when the corresponding ontology has a relevant term:

mappings:
  mondo_mappings:
  - term: {id: MONDO:0006058, label: Wilms tumor}
    mapping_predicate: skos:exactMatch
    mapping_source: MONDO
  icd10cm_mappings:
  - term: {id: ICD10CM:C64.9, label: Malignant neoplasm of unspecified kidney, except renal pelvis}
    mapping_predicate: skos:closeMatch
    mapping_source: ICD-10-CM
  ncit_mappings:
  - term: {id: NCIT:C3267, label: Wilms Tumor}
    mapping_predicate: skos:exactMatch
    mapping_source: NCIT

NCIT mappings at the disease level are especially valuable for downstream CCDI interoperability. Do not omit them just because MONDO is the primary anchor.

The schema slot ncit_mappings ranges over NCITMapping and accepts disease, subtype, or disease/finding terms — use the most specific NCIT class available.

Subtypes: flat axes, not a lattice

Multi-dimensional cancer classification (histology × stage × laterality × predisposition × age) should be modeled as flat has_subtypes entries with explicit classification axes, not as nested hierarchies and not as separate disease files.

Use classification to label the axis that the subtype belongs to. The Wilms tumor entry is the canonical worked example:

`classification` axis	Wilms subtypes
`anaplasia_status`	Favorable Histology, Anaplastic
`histological_pattern`	Blastemal Predominant, Epithelial Predominant, Stromal Predominant, Mixed Cell Type
`laterality`	Bilateral, Unilateral
`predisposition_context`	Hereditary Predisposition-Associated, Sporadic
`age_group`	Childhood, Adult

Other axes seen across cancer entries: molecular_driver, stage, anatomic_site, treatment_associated, mutation_class.

When you introduce a new axis, prefer reusing an existing axis name. Axis vocabulary is open but should converge over time. Keep axes as close to closed/explicit as practical — for example, the complement of "hereditary / predisposition-associated" is Sporadic, not "somatic", because hereditary tumors still acquire somatic alterations.

`subtype_term` and subtype `mappings`: ontology grounding only

A subtype with an NCIT or MONDO term is ontology-grounded, not a separate curation target.

Use subtype_term when the subtype itself has a MONDO identifier (disease-level grounding), and use mappings.ncit_mappings for NCIT clinical/oncology grounding. The two slots are complementary and can both appear on the same subtype:

# NCIT-grounded subtype (no MONDO term for this facet)
- name: Anaplastic
  classification: anaplasia_status
  mappings:
    ncit_mappings:
    - term: {id: NCIT:C6952, label: Anaplastic Kidney Wilms Tumor}
      mapping_predicate: skos:closeMatch
      mapping_source: NCIT

# MONDO-grounded subtype with an additional NCIT mapping
- name: Childhood
  classification: age_group
  subtype_term:
    preferred_term: childhood kidney Wilms tumor
    term: {id: MONDO:0024676, label: childhood kidney Wilms tumor}
  mappings:
    ncit_mappings:
    - term: {id: NCIT:C27730, label: Childhood Kidney Wilms Tumor}
      mapping_predicate: skos:closeMatch
      mapping_source: NCIT

subtype_term and subtype-level mappings provide ontology grounding for the subtype facet within the parent disease entry.
They do not imply that a separate dismech disease page exists or should exist for that subtype.
Renderers should display these as outbound hyperlinks to the source ontology (NCIT, MONDO), not as cross-entry "Not Yet Curated" indicators.

If you find that downstream tooling treats a grounded subtype as a missing disease entry, that is a tooling bug — fix the comparator, not the curation.

NCIT vs. MAXO selection in cancer entries

NCIT often provides clinically more specific oncology terms than MAXO. Prefer NCIT when it materially improves specificity; otherwise stay with the existing MAXO conventions documented in CLAUDE.md.

Use case	Preferred ontology	Notes
`disease_term`	MONDO	Always MONDO-first
Disease-level `mappings`	MONDO + ICD-10-CM + NCIT	All available
`histopathology.finding_term`	NCIT (or HP for canonical patterns)	NCIT has tumor-specific morphology terms
Surgical procedures	NCIT or MAXO — whichever is more specific	NCIT often wins for oncologic procedures
Chemotherapy action	MAXO:0000647 (chemotherapy)	Generic; pair with `therapeutic_agent`
Radiation therapy	MAXO:0000014
Specific drug	CHEBI (small molecule) or NCIT (biologic / drug class)	Use `therapeutic_agent` slot
Biomarkers / gene products	NCIT (clinical biomarker) + HGNC (gene)	`biomarker_term` / `gene_products`
Staging system terms	NCIT	NCIT carries COG, SIOP, AJCC codes

Worked example: Wilms tumor

kb/disorders/Wilms_Tumor.yaml is the reference implementation of this SOP:

disease_term: MONDO:0006058 Wilms tumor (MONDO-first)
mappings: MONDO:0006058 (exactMatch) + ICD10CM:C64.9 (closeMatch) + NCIT:C3267 (exactMatch)
has_subtypes: 12 flat subtypes across 5 axes (anaplasia_status, histological_pattern, laterality, predisposition_context, age_group)
Most subtypes carry mappings.ncit_mappings with skos:closeMatch to the corresponding NCIT class — ontology-grounded, no separate page implied
Histopathology, treatments, and biomarkers follow the table above

When in doubt about a new cancer entry, mirror Wilms tumor's structure.

CLAUDE.md — repo-wide curation conventions (treatment terms, MAXO/NCIT/CHEBI patterns)
.claude/skills/cancer-curator/SKILL.md — mechanics of cancer pathophysiology, histopathology, and therapeutic agent curation
.claude/skills/disease-classification/SKILL.md — deeper guidance on classification axes and the classifications block
#795 — MONDO disposition/susceptibility anchors and skos:closeMatch fallback
#1007 — Subtype gaps between dismech and MONDO
#1198 — This document

Cancer Curation SOP: MONDO/NCIT Usage and Subtype Axes

Curation unit: one entry per mechanism graph

disease_term: MONDO-first

Disease-level mappings: routinely include NCIT