ADAMTSL5 (Q6ZMM2) — bioinformatics support for the GO annotation review

Reproduce with:

uv run python genes/human/ADAMTSL5/ADAMTSL5-bioinformatics/resolve_withfrom.py
uv run python genes/human/ADAMTSL5/ADAMTSL5-bioinformatics/family_propagation_census.py

resolve_withfrom.py must be run first — the census reads its withfrom_resolution.json.
All numbers below are produced by those two scripts; none is hand-entered.

1. ADAMTSL5 has no metallopeptidase domain — verified three independent ways

The campaign brief flagged "ADAMTSL proteins lack the catalytic metalloprotease domain"
as a lead to establish, not a premise. It is established here:

The predicted failure mode did NOT occur. ADAMTSL5's GOA contains no
peptidase/metallopeptidase term of any kind — no GO:0004222, no GO:0008237, no
GO:0006508. This was a hypothesis, and it is reported here as not confirmed,
not as a finding. PAINT in fact handled the catalysis question correctly (§3).

Domain architecture (UniProt FT): signal peptide 1–42, mature chain 43–481,
TSP type-1 domain 45–97, NTR domain 360–479. At 481 aa ADAMTSL5 is by far the
smallest member of the family (others 951–1935 aa).

Incidental, verified: the psoriasis autoantigen peptide VRSRRCLRL (PMID:26621454)
occupies residues 67–75, inside the TSP type-1 domain.

2. The GO:0031012 IBA is exceptionally well supported — 16/16 donors

WITH/FROM has 17 tokens: 16 protein donors + the tree node PANTHER:PTN000347317.
The GOA field matches the cached PAINT seed list exactly (asserted in code:
goa_tokens - seeds == {"PANTHER:PTN000347317"} and seeds - goa_tokens == set()).

All 16 protein donors carry their own experimental (IDA/HDA) annotation to
GO:0031012 or a descendant. So SOURCE_WEAK_OR_INFERRED / SOURCE_EVIDENCE_WEAK
would be factually contradicted by the measurement; there is no propagation defect to
report on this row.

One token is self-referential (UniProtKB:Q6ZMM2) — a PAINT curator judging the
function core, which is valid.

The donor set is heterogeneous, so the general term is the LCA

Four distinct locations. GRANULARITY_MISMATCH requires the donors to agree; they
do not, so GO:0031012 is the correct LCA and refining it would mean arbitrarily
preferring one donor's compartment. → ACCEPT, no specificity upgrade.

Also checked (ACRV1-style, negative): the IBA does not land above its donors — 14
of 16 donors hold the term itself — so no downward MODIFY is warranted.

Resolver gotchas measured here (each cost a cycle)

• QuickGO's geneProductId rejects MOD ids outright (HTTP 400, "contains invalid
  values") for MGI:MGI:109249, MGI:109249, FB:FBgn0003137, FBgn0003137,
  WB:WBGene00003242, RGD:621241. It indexes UniProtKB only. A donor's own evidence
  can therefore only be queried through its resolved UniProt accession. The first
  version of the script silently printed "NONE" for every MOD token — a silent
  degradation reading as a null result. Fixed, and unqueryable is now its own state.
• UniProt does not index the WBGene id. xref:wormbase-WBGene00003242 returns [];
  UniProt keys WormBase xrefs on the CDS name (C37C3.6). Resolved via the WormBase
  REST (WBGene00003242 → mig-6) then a UniProt symbol search.
• That symbol search must be gene_exact:, not gene:. Fuzzy gene:mig-6 also
  returns mig-10 / mig-5 / mig-14 / mig-18 and puts the wrong entry first — the
  size=1 trap in a new guise. With gene_exact the token resolves uniquely to
  O76840 PPN1_CAEEL papilin (Swiss-Prot, 2167 aa), which holds GO:0005604 by IDA.
  Getting 16/16 via a fuzzy head match would have been the right number for the wrong
  reason.
• 13 of 16 tokens are multi-hit (a Swiss-Prot entry plus TrEMBL fragments). All
  candidates are reported; the reviewed entry is the one queried.
• QuickGO strips the DB prefix from WITH/FROM tokens. withFrom[].connectedXrefs[]
  returns {"db": "FB", "id": "FBgn0003137"} as two separate fields, so comparing the
  bare id set against a GOA TSV WITH/FROM string yields 17 vs 17 tokens reported as
  "not identical" for data that is actually the same. Reassemble as db + ":" + id
  before any set comparison.

Cross-check against the concurrently reviewed sibling ADAMTSL3

Derived from QuickGO directly rather than read off the sibling branch. ADAMTSL3
(P82987) carries a GO:0031012 IBA whose WITH/FROM token set is exactly identical
to ADAMTSL5's — same node, same 17 tokens, verified as a set equality in a single
serialisation. Two independent derivations of one node agree. ADAMTSL3 also shares the
GO:0071953 TAS from PMID:23962539 and the InterPro GO:0030198 IEA, so all three of
this review's non-ACCEPT verdicts have a direct counterpart there.

3. The real PAINT defect: one node, incoherent propagation across the family

All eight human ADAMTSL/papilin proteins are in PTHR13723, and node
PTN000347317 carries four IBD annotations (GO:0031012, GO:0030198,
GO:0004222, GO:0006508). Who actually received them:

What PAINT got right. The catalytic terms reached all four catalytic controls and
none of the seven non-catalytic members — with ADAMTSL2 carrying an explicit
NOT|enables GO:0004222 (IKR at node PTN002673039). The family's loss of catalysis
is modelled correctly. This is the counter-example to the campaign's usual finding.

What is incoherent. From a single node, GO:0031012 reached 5 of 7 and
GO:0030198 reached 3 of 7, in no biologically coherent pattern:

• PAPLN receives neither, although three of its own orthologs are seeds for
  GO:0031012 at that very node — fly Ppn (FB:FBgn0003137), mouse Papln
  (MGI:MGI:2386139) and worm mig-6/ppn-1 (WB:WBGene00003242). Human PAPLN has
  no IBA at all (5 annotations, all IEA/TAS).
• ADAMTSL1 receives neither and likewise has no IBA at all (4 annotations).
• ADAMTSL3 and ADAMTSL5 receive the location term but not the process term, while
  ADAMTSL2/ADAMTSL4/THSD4 receive both.

Consequence for this review, stated as a limit rather than as support: ADAMTSL5's
absence of a GO:0030198 IBA cannot be read as a deliberate curatorial judgement
that the process does not apply. The same gap left ADAMTSL1 and PAPLN with nothing at
all, so it is more likely a propagation failure than a decision. The case against
GO:0030198 for ADAMTSL5 therefore rests on §4 and §5 only.

4. GO:0030198 comes from a family signature that bundles proteases with non-proteases

InterPro:IPR013273 is named, literally, "ADAMTS/ADAMTS-like" (family, 26,580
proteins) and carries exactly one GO mapping: GO:0030198 extracellular matrix organization. The signature therefore lumps the catalytic ADAMTS proteases — for which
ECM organization is a direct consequence of ECM proteolysis — together with the
non-catalytic ADAMTS-like proteins, which have no such activity (§1).

This is the campaign's familiar shape one aspect over: not "a domain's name is not an
activity" (MF), but a family signature too broad to carry a process (BP).

5. The only direct test of ECM organization by ADAMTSL5 was negative

From PMID:23010571 (full text): after showing colocalisation with fibrillin
microfibrils, the authors report

"However, comparison of microfibril density in fBNL cell cultures grown in the
presence of ADAMTSL5 or vector conditioned medium, did not identify a consistent
difference (data not shown)."

and the abstract states colocalisation occurred "but without discernible effect on
microfibril assembly". Direct binding to fibronectin was also not supported.

Weight of this negative, stated honestly: it is a "data not shown" result from a
single exogenous-protein assay. It argues that no role in ECM/microfibril organization
has been demonstrated; it does not refute one. That supports
MARK_AS_OVER_ANNOTATED, and is not strong enough to earn REMOVE.

6. The three GO:0005515 rows are one Y2H screen, on three hub proteins

All three rows come from PMID:32296183 (HuRI). IntAct shows each partner logged under
three sub-methods of the same experiment — two hybrid array + two hybrid prey pooling approach + validated two hybrid, MI-score 0.56. UniProt's NbExp=3 is
therefore one screen counted three ways, replicating the ACRV1 finding on a second
gene.

Distinct-partner counts (derived as an entity set; IntAct records are not partners):

Each partner is topologically incompatible with ADAMTSL5, a signal-peptide
(1–42) secreted ECM glycoprotein: it does not enter the cytoplasm, nucleus, or cornified
envelope. Decided per partner, all three come out the same way — unreplicated
screen noise → MARK_AS_OVER_ANNOTATED.

All three partner accessions resolve to reviewed canonical Swiss-Prot entries with
matching lengths — no TrEMBL/ORFeome substitution of the ACRV1 kind. Reported as a
negative.

The gene's real binding partners, FBN1 and FBN2, come from the same IntAct record
by pull down from PMID:23010571, and are captured by GO:0050436 microfibril binding.

7. GO:0071953 elastic fiber is a bulk review-based assignment

Projection test by reference (entities derived as a set, not from the annotation total):

PMID:23010571 annotates only ADAMTSL5 and its mouse ortholog — gene-specific
curation, not a projection. Negative result, reported.

PMID:23962539 is a review (publication_type: REVIEW, full_text_available: false) used as the TAS source for 62 entities, assigning GO:0071953 to 41 of
them. Its abstract never mentions ADAMTSL5. Notably the same curation from the same
review assigned the more specific GO:0001527 microfibril to 15 other proteins —
including THSD4/ADAMTSL6 — so the specific term was demonstrably available and was not
chosen for ADAMTSL5.

The specific term is missing from GOA but present in UniProt

UniProt's DR block carries GO:0001527; C:microfibril; IDA:UniProtKB, but GOA does
not have it (verified: QuickGO returns exactly 12 annotations for Q6ZMM2, matching the
GOA TSV, and GO:0001527 is not among them). So the gene's best-evidenced and most
specific localisation is absent from GOA while a weaker, broader TAS row is present.

GO:0001527 is current (not obsolete, no secondaryIds) and is a part_of child of
GO:0071953 — so replacing the TAS row with it retains the parent by closure while
gaining precision, and matches the merged ADAMTSL4 review, which independently proposed
GO:0001527 as a NEW term.

An ontology issue this surfaces

GO:0001527 microfibril is part_of GO:0071953 elastic fiber, which asserts that
every microfibril is part of an elastic fiber. That contradicts GO:0001527's own
definition — "Extracellular matrix components occurring independently or along
with elastin" — and contradicts the ciliary zonule, a fibrillin-microfibril structure
essentially devoid of elastin. Recorded in suggested_questions.

8. Checks run that came back negative (recorded so they are not re-run blind)

• Retraction / erratum / expression-of-concern: all 9 PMIDs relied on
  (23010571, 23962539, 32296183, 26621454, 27857980, 28482118, 33197513, 37330172,
  38524140) checked via PubMed publication types and each record's own
  CommentsCorrections (a Publisher Correction is invisible to a pubtype query).
  None flagged. PMID:26621454 has two Comment in entries — commentaries, not
  corrections.
• Partner accession integrity: all three IPI partners are reviewed canonical
  Swiss-Prot entries at their canonical lengths. No dead or truncated accessions.
• IBA-above-donor precision defect (ACRV1 shape): absent — the IBA sits at the term
  14 of 16 donors hold.
• Reference projection for the primary paper: absent — 2 entities.
• Dead-accession check: every WITH/FROM UniProt lookup returns a named entry; the
  script raises rather than reporting a silent zero.