[QC] Two invariants: no cell part_of an acellular structure; epithelial stem cell classification completeness

[cmungall]

Two checks that would have caught defects found while reviewing #3692 and #3694.

1. No cell part_of an acellular anatomical structure

CL ⊑ part_of some <acellular anatomical structure>

should be empty. A cell can be adjacent to, attached to, or in contact with an ECM structure, but not a part of one.

Implemented as a logical axiom in #3707:

DisjointClasses(obo:CL_0000000 ObjectSomeValuesFrom(obo:BFO_0000050 obo:UBERON_0000476))

EL-safe for WHELK (DisjointClasses with an existential reduces to ⊥; the more natural has_part only (not cell) form needs a DL reasoner and is better proposed to CARO/UBERON).

Still worth having as a SPARQL query as well. An unsatisfiable class names the class, not the offending axiom — and for several of the violations below the offending axiom is in an import, so a curator hitting it gets no useful signal. The SPARQL version can report the inherited path.

2. Completeness of epithelial stem cell classification

cl info [ .sub CL:0000036 .and .sub "epithelial cell" ]
CL:1000486 ! basal cell of urothelium
CL:2000033 ! limb basal cell of epidermis
CL:1000349 ! basal cell of epithelium of bronchus
CL:0002187 ! basal cell of epidermis

Four of the eight basal cell descendants. Whatever the right answer turns out to be, a partial result is definitionally a bug: it means some subtypes acquired epithelial classification through a hand-asserted lineage parent (keratinocyte, urothelial cell, basal epithelial cell of tracheobronchial tree) while their siblings did not. Notably CL:0002633 respiratory basal cell is absent while its own child CL:1000349 is present, and CL:1000447 is absent despite being labelled "epithelial cell of stratum germinativum of esophagus".

The set should be empty or complete. Blocked on #3694 — this check fails today and will pass once basal cell is reclassified. This one is CL-local and needs no upstream work.

Part of the epithelial hierarchy cleanup (#3703).

---

@cmungall drove the agent, checked its reasoning, and is accountable for its output.

Generated by Claude Code

[cmungall]

Implemented check 1 as a logical axiom in #3707 (draft), and measured the real violation set. It is larger than the two asserted cases and the extra ones are not CL bugs.

Computing the transitive part_of closure, inherited over is_a, against the 29-member acellular branch gives 7 violating classes:

class	cause
CL:0002169 basal cell of olfactory epithelium	asserted part_of basal lamina — #3704
CL:0002171 globose cell of olfactory epithelium	inherits CL:0002169
CL:0002184 basal proper cell of olfactory epithelium	inherits CL:0002169
CL:4033093 limbal epithelial stem cell of cornea	asserted part_of basal lamina — #3704
CL:0008045 tanycyte of subcommissural organ	upstream
CL:4023181 hypendymal cell	upstream
CL:2000096 fibroblast of the reticular layer of dermis	upstream

The last three have correct CL axioms and inherit a bad part_of from UBERON:

SubClassOf(UBERON:0002139 subcommissural organ,      part_of some UBERON:0011357 Reissner's fiber)
SubClassOf(UBERON:0001993 reticular layer of dermis, part_of some UBERON:0008877 epidermal-dermal junction)   [source: FMA]

An organ is not part of the acellular strand it secretes (UBERON separately records subcommissural organ produces SCO-spondin), and the reticular dermis is not part of the basement membrane zone. Both also put a multicellular anatomical structure inside an acellular one, which is a defect on UBERON's own terms given CARO's DisjointClasses(UBERON:0000476, UBERON:0010000).

Two consequences for this ticket:

Check 1 is worth keeping as a SPARQL query even with the axiom in place. An unsatisfiable class names the class, not the offending axiom, and here the offending axiom is in an import — a curator hitting that gets no useful signal. The SPARQL version can report the inherited path.

Check 2 (.sub CL:0000036 .and .sub 'epithelial cell') needs no such caveat — it is CL-local and unblocks as soon as #3694 lands.

The disjointness axiom itself is EL-safe for WHELK (DisjointClasses with an existential reduces to ⊥; the more natural has_part only (not cell) form would need a DL reasoner and is better proposed to CARO/UBERON, where it would protect every downstream ontology).

---

Generated by Claude Code

[cmungall]

Attribution note for the comment above: @cmungall drove the agent, checked its reasoning, and is accountable for its output.

---

Generated by Claude Code