PANTHER/TreeGrafter: a reductase annotated as a dehydrogenase, times 1,387

PANTHER/TreeGrafter: a reductase annotated as a dehydrogenase, times 1,387

Back to the project page. Reproduce with
uv run python projects/CACAO/scripts/upstream_checks.py.

Found while reviewing the CACAO annotations on Myxococcus xanthus
actC, whose three CACAO rows were
accepted or refined. The defective row on that gene was not CACAO's: actC
carries GO:0004029 aldehyde dehydrogenase (NAD+) activity by IEA from
TreeGrafter, and actC has no aldehyde dehydrogenase fold at all.

Tracing it back produced a two-stage error with a single origin.

Stage 1: the seed annotation states the wrong reaction

Q12177 (YLL056C, S. cerevisiae, "Uncharacterized protein YLL056C") carries:

GO:0004029  IDA  PMID:28265724   SGD
GO:0004029  IBA  GO_REF:0000033  GO_Central   WITH/FROM: PTN002449572|SGD:S000003979
GO:0005737  IDA  PMID:28265724   SGD
GO:0005737  IBA  GO_REF:0000033  GO_Central
GO:0008150  ND   GO_REF:0000015  SGD

The cited paper is titled "YLL056C from Saccharomyces cerevisiae encodes a
novel protein with aldehyde reductase activity", and its abstract states:

Enzyme activity assays showed that Yll056cp is not a quinone or ketone
reductase but an NADH-dependent aldehyde reductase, which can reduce at least
seven aldehyde compounds.

GO:0004029 aldehyde dehydrogenase (NAD+) activity is defined as:

Catalysis of the reaction: an aldehyde + H2O + NAD+ = a carboxylate + 2 H+ +
NADH.

These are not two granularities of one claim. The paper's enzyme reduces an
aldehyde to an alcohol; GO:0004029 oxidises an aldehyde to a carboxylic
acid. Different direction, different product, different EC class (1.1.1.x
against 1.2.1.3). The paper also places the protein in the atypical SDR
group, and Q12177 carries Pfam PF01370, the NAD-dependent
epimerase/dehydratase family — not the ALDH superfamily fold.

The reaction the paper reports, written in GO's oxidative direction, is
GO:0004022 alcohol dehydrogenase (NAD+) activity ("an alcohol + NAD+ = an
aldehyde or ketone + NADH + H+") or its parent GO:0018455. The substrate
range given (glycolaldehyde best by Vmax/Kcat, formaldehyde by Km, plus
furfural, butyraldehyde and propylaldehyde) is broad, so a curator may prefer
the parent.

The GO:0005737 cytosol/cytoplasm half of the pair is fine — the abstract does
say the protein "localized to the cytoplasm".

Stage 2: two pipelines amplify it

Route Node Rows
PAINT (IBA, GO_REF:0000033) PTN002449572 63
TreeGrafter (IEA, GO_REF:0000118) PTN005180820 1,324 (+1,324 cytoplasm)

1,387 GO:0004029 rows across 1,277 taxa trace to that one IDA.

The recipients are PTHR48079 — InterPro's integrated name for it is
"NAD(P)-dependent Metabolic Enzyme" (IPR051783), its representative structure
is a "putative NAD-dependent epimerase/dehydratase", and its members carry
PF01370 or PF13460. Spot-checking them returns dTDP-glucose 4,6-dehydratase
(rmlB), GDP-fucose synthase (wcaG), GDP-mannose 4,6-dehydratase (gca),
sterol-4alpha-carboxylate 3-dehydrogenase, wbpK, and many copies of E. coli
YeeZ. These are extended-SDR Rossmann enzymes. None of them is an aldehyde
dehydrogenase.

The seed sits in subfamily PTHR48079:SF9 ("PUTATIVE-RELATED"); the recipients
are overwhelmingly PTHR48079:SF6 ("NAD(P)-BINDING DOMAIN-CONTAINING
PROTEIN-RELATED"). So the assertion is attached at a node ancestral to both
subfamilies and propagates across the family from one member of a different
subfamily.

Why this is not a donor-count argument

This repository's own
IBA guidance warns that a short WITH/FROM donor list is not
weak support, and that is right: PTN002449572 having exactly one donor is not
the complaint. A one-donor node seeded by a well-characterised protein can be
entirely sound.

The complaint is about the content of the seed. The node's single donor
carries a term that states the wrong reaction, so every annotation inheriting
from the node inherits that error, and no amount of phylogenetic care at the
node could have caught it — the node placement may well be correct for whatever
Q12177 actually does. Fixing the propagation therefore means fixing the seed
first.

The contrast with the genuine ALDH clades is worth stating because it shows the
system normally works. PTN000192666 emits 618 IBA rows for GO:0004029 from
34 donors including ALDH1A1 (P00352), ALDH2 (P05091) and ALDH1B1
(P30837); PTN008111916 emits 325 from 12 donors. Those sit in PTHR11699,
the real ALDH family. Of the 56 proteins in GOA with an experimental
GO:0004029 annotation, 29 are in PTHR11699 and exactly one — Q12177 — is
in PTHR48079.

Requested changes

To SGD (the origin, and the only fix that matters). Re-term Q12177's
GO:0004029 IDA to GO:0004022 alcohol dehydrogenase (NAD+) activity, or to
GO:0018455 if the substrate breadth argues for the parent. The cited paper
supports a reduction, not an oxidation, and says so in its title. Leave the
GO:0005737 IDA alone.

To GO Central / PAINT. Once the seed is re-termed, the IBD at
PTN002449572 needs the same correction, which clears its 63 IBA rows.

To PANTHER / UniProt's TreeGrafter pipeline. The GO association behind
PTN005180820 needs the same correction, which clears 1,324 rows. Worth asking
separately whether a node spanning SF6 and SF9 should be emitting a specific
catalytic activity at all, given that the family's own InterPro description
spans "oxidative decarboxylation, reduction, and dehydration reactions" — a
family-level MF this specific is a strong claim for a group that heterogeneous,
independently of whether this particular term is right.

Note on IPR051783's description. InterPro's text for the family is
flagged is_llm: true, is_reviewed_llm: false in the API response. It is
accurate as far as this review can tell, and it is not the source of this error,
but it is unreviewed machine-generated text being served as a family
description, which is worth knowing when citing it.

A note for this project's taxonomy

This is the inverse of the shape the batch-3 screens were built to find. The
screens looked for a student mis-reading a paper; here a professional curator
read the paper correctly and chose a term whose label matches the paper's words
("aldehyde ... dehydrogenase"-shaped) but whose definition is the opposite
reaction. Then two automatic pipelines multiplied it by roughly 1,400.

It belongs with the InterPro2GO GO:0019685 case as the same underlying
failure: a term chosen from its label while its definition says something
else, and in both cases the amplification, not the original slip, is what
makes it worth reporting. The original slips are one row each.

Status