ACP7 (acid phosphatase type 7) — review notes

PAINT no-IBA project review, using the affinage deep-research provider plus UniProt Q6ZNF0
and the GOA TSV.

Uncharacterised, not unstudied

Four annotations, all IEA, and no functional or biochemical data exists for this protein.
The protein itself is real — UniProt records PE 1: Evidence at protein level with proteomics
identification — and there is a primary paper.

The provenance is worth stating precisely, because it applies to every annotation:

Statement Evidence
FUNCTION (metallophosphoesterase, purple acid phosphatase family) ECO:0000250 → P80366
EC 3.1.3.2, CATALYTIC ACTIVITY (RHEA:15017) ECO:0000250 → P80366
COFACTOR Fe cation; COFACTOR Zn²⁺ ECO:0000250 → P80366
SUBCELLULAR LOCATION Secreted ECO:0000305 (curator inference)
SIGNAL 1–26 ECO:0000255 (prediction)

P80366 is PPAF_PHAVU — the Fe(3+)-Zn(2+) purple acid phosphatase of the kidney bean,
Phaseolus vulgaris.
A plant enzyme is a distant source for a human function assignment, and
it is the sole source for this gene's catalytic activity, EC number and both cofactors.

Correction: the kidney bean template is a deliberate published choice

An earlier draft of this review asserted three times that no primary literature exists, reasoning
from an empty affinage record. That was wrong, and the refutation was in the UniProt file the
review was itself citing: reference [4] is PMID:16793224, "Identification and molecular
modeling of a novel, plant-like, human purple acid phosphatase"
— the ECO:0000303 source for
UniProt's own "Purple acid phosphatase long form" AltName.

An empty affinage record is evidence about affinage's coverage, not proof that no literature
exists.
The correct reading is narrower: affinage found no mechanistic discoveries, which fits
a gene whose only paper is a bioinformatic identification plus a structural model rather than a
functional study.

And the paper strengthens the review rather than merely correcting it. It argues ACP7 is the
founding member of a novel plant-like PAP subfamily in animals, and builds a structural model
of the human enzyme on the red kidney bean structure specifically, showing the catalytic
centre is present. So P80366 is not an arbitrary cross-kingdom hop — it is the considered template
for a protein argued to be plant-like. That makes the ECO:0000250 chain far better justified
than the first draft implied.

UniProt's PAN-GO; Q6ZNF0; 0 GO annotations based on evolutionary models still stands, and is
the accurate statement of what is missing: evolutionary-model annotation, and functional data.

Actions

Term Action Reason
GO:0003993 acid phosphatase activity ACCEPT best available inference; fold, family and metal residues all well supported — but a prediction from a plant homolog
GO:0046872 metal ion binding ACCEPT best-supported of the four — residue-level BINDING features for Fe and Zn; the binuclear centre defines the family
GO:0016787 hydrolase activity KEEP_AS_NON_CORE two levels above the specific term from the same prediction
GO:0005576 extracellular region ACCEPT plausible, but ECO:0000305 inferred from an ECO:0000255 predicted signal peptide — a prediction resting on a prediction

Nothing is removed. All four are reasonable inferences from a real, well-supported fold. The
review's contribution is to record what kind of knowledge this is: the gene is annotated
entirely by prediction, and core_functions says so explicitly rather than presenting the
activity as established.

Cache gotcha, hit again

just validate rewrote cache/go/terms.csv and dropped 18 rows added to main by other
merged PRs. Caught by git diff origin/main -- cache/go/terms.csv | grep '^-GO:' before
committing; fixed with git checkout origin/main -- cache/go/terms.csv. Second occurrence in
this campaign (see PR #2227) — worth checking on every gene.