GO:0034045 — corpus slice audit

IN_PROGRESS BIOLOGY_DOMAIN

Species: human, mouse, yeast, SCHPO, worm, DICDI

Subsequent IBA recovery reassessment (2026-09-22)

The decisions below record the earlier audit. PR #3114 subsequently reconsidered ATG14, ULK1, WIPI2 and ATG12 using live ontology responses and additional primary evidence. GO:7770114 phagophore membrane now exists, and GO:0034045 is obsolete. GO:0097632 survives with the name extrinsic component of phagophore membrane and a part_of relation to GO:7770114; GO:0097629 remains valid. Thus the earlier R3 term request has been overtaken by an ontology change, and the earlier ATG14 MODIFY statements are superseded by the current gene review. ULK1 membrane association is supported by the studies of membrane binding and palmitoylation cited in its review (PMID:22885598; PMID:39169022). The current ULK1/WIPI2/ATG12 replacements use GO:7770114 where their evidence supports phagophore membrane. This is not a blanket replacement for every GO:0034045 row: the RAB1B caution and other genes still require their own evidence assessment. See recovery tracking and the current gene YAML for final actions; the older tables below are retained as session provenance.

GO:0034045 — corpus slice audit

Supporting page for Biomolecular Condensates, and the first subproject of
the SL project (SL-0221 Preautophagosomal structure membrane). This is a per-assertion
re-adjudication of every GO:0034045, GO:0097632 and GO:0097629 annotation held in this
repository, carried out against the analysis in GO issue
#29437. It is a small slice —
59 assertions across 21 gene folders, against ~2,151 direct annotations on GO:0034045 — but
it is a slice where every assertion has a written review attached, so it can say something
the annotation counts cannot. (GO:0097629 was added to the scope late; see the R4 gap below.)

Headline: we reproduced the failure mode

The audit ran in two passes. The first pass reached genes through the term itself:
all 23 previously reviewed assertions were ACCEPT. Not one had been questioned. After
re-review, all 31 have moved — 26 to MODIFY and 5 to MARK_AS_OVER_ANNOTATED with the
missing terms proposed. Nothing is left at ACCEPT. The second pass, described below,
reached seven more genes through the issue's §5.4 triage table and reviewed each in full,
adding 28 assertions on the same footing.

That is the strongest corroboration this corpus can offer for the mechanism described in
§2 of the issue analysis: a reviewer meeting this term sees a plausible label and a
definition that is true as written ("a membrane associated with the PAS" — there are
membranes at the PAS), while the false claim sits in a bounding_layer_of axiom that is
never surfaced. Our reviewers had the literature, wrote a justification for each one, and
still accepted every single assertion. Several of the justifications do not even address
localization:

gene term prior reason, verbatim
ATG7 (yeast) GO:0097632 "Accepted because this annotation aligns with established ATG7 E1-like function and autophagy pathway roles."
Rab7 (mouse) GO:0034045 "Consistent with the Rab7 role in autophagy."
RAB7A (human) GO:0034045 "Consistent with RAB7A role in autophagy regulation."

ATG7's cited supporting text was the paper's title. These are pathway-membership
arguments standing in for localization evidence — the same substitution that put the
annotations there originally.

Changes made

All 31 first-pass assertions, across 14 genes, have moved off ACCEPT — 26 to MODIFY,
5 to MARK_AS_OVER_ANNOTATED with new terms proposed. All files revalidate clean. The
second-pass genes are tabulated separately further down.

species gene term evidence reference n destination
mouse Rab7 GO:0034045 IDA PMID:19956673 1 GO:0061908 phagophore
human RAB7A GO:0034045 IEA GO_REF:0000107 1 GO:0061908 phagophore
human ATG5 GO:0034045 IBA, IDA, IEA, ISS PMID:32960676 + refs 4 GO:0061908 phagophore
DICDI atg1 GO:0034045 IBA, ISS, IEA GO_REF:0000033/24/44 3 GO:0000407
SCHPO atg101 GO:0034045 IEA GO_REF:0000044 1 GO:0000407
SCHPO atg2 GO:0034045 IEA GO_REF:0000044 1 GO:0000407
SCHPO atg5 GO:0034045 IBA, IEA GO_REF:0000033/44 2 GO:0000407
SCHPO atg16 GO:0034045 IEA GO_REF:0000044 1 GO:0000407
SCHPO atg38 GO:0034045 IEA GO_REF:0000044 1 GO:0000407
yeast ATG7 GO:0097632 IDA PMID:10233148 1 GO:0000407
human ATG14 GO:0097632 IBA, IDA GO_REF:0000033, PMID:21518905 2 GO:0000407
human ATG14 GO:0034045 EXP ×3, IDA, IEA, TAS PMID:18843052 + refs 6 GO:0061908 phagophore
human ATG14 GO:0097629 IBA, IDA GO_REF:0000033, PMID:21518905 2 GO:1990462 omegasome

Two destinations, chosen by what the cited evidence actually shows:

Knock-on edits: ATG7's core_functions.locations carried GO:0097632 and now carries
GO:0000407; ATG5's carried GO:0034045 and now carries GO:0061908; atg1's carried both
GO:0000407 and GO:0034045, and the duplicate was dropped. propagation_review blocks were
added to the four IBA annotations changed to MODIFY.

The S. pombe set is the cleanest evidence in the audit

Five of the moved annotations are S. pombe IEAs whose sole source is GO_REF:0000044 —
that is, SL-0221 with nothing else behind it. In every case a reviewer had written a
justification quoting experimental evidence about PAS puncta, then accepted a term naming
a PAS membrane, without noticing the substitution. The conflation is not buried in an
axiom file here; it is visible inside our own prose, one line apart.

One correction to the §5.4 triage: Rab7

The triage table lists RAB7 under "no evident support — review for removal", citing
PMID:19956673. Reading the full text, that is not quite right, and the difference matters
for how the migration is framed.

The paper does contain localization data for Rab7, in NIH3T3 and HeLa cells:

Thus, a population of Rab7 is recruited to GFP-Atg5 positive membranes during the early
phase of GcAV formation.

An Atg5-positive early sequestering membrane is a phagophore. So the assertion is not
unsupported — it is mis-targeted: the paper is about GAS-containing autophagosome-like
vacuoles, a xenophagy structure, and never examines the phagophore assembly site. The
abstract also states outright that Rab7 is

an additional component, which is dispensable in canonical autophagosome formation

which is the opposite of a claim about the canonical PAS. The right disposition is
therefore a move to GO:0061908, not deletion.

This distinction is worth carrying into Stage 5. "No evident support" and "supported, but
for a different structure" need different handling: the first is a removal, the second is a
migration that must not be lost. An abstract-level triage cannot separate them — which is
the caveat §8 already makes, borne out on the first case checked.

Note also that this repository holds both ends of the propagation chain: human RAB7A
carries the same term as an IEA projected from mouse Rab7a via GO_REF:0000107
(WITH UniProtKB:P51150). Correcting the source should carry through, and both have been
changed together. This is a two-link instance of the amplification described in §5.1.

A gap in recommendation R4: GO:0097629 has the same defect

R4 proposes obsoleting GO:0097632 extrinsic component of phagophore assembly site membrane,
noting that issue #23424 obsoleted its intrinsic and integral siblings and the parallel
autophagosome set, and that GO:0097632 "was missed". A second term was missed in the same
pass.

set intrinsic integral extrinsic
autophagosome membrane GO:0097636 obsolete GO:0097637 obsolete GO:0097635 obsolete
omegasome membrane GO:0097630 obsolete GO:0097631 obsolete GO:0097629 live
phagophore assembly site membrane GO:0097633 obsolete GO:0097634 obsolete GO:0097632 live

For the autophagosome all three went. For the omegasome and the phagophore assembly site, only
intrinsic and integral went and the extrinsic term survived in both. GO:0097629 carries the
identical protein-topology defect as GO:0097632 and should be obsoleted with it, so R4 should
name both.

ATG14 is the demonstration. It carries GO:0097629 and GO:0097632 as IDAs from the same
paper and the same experiment — PMID:21518905, showing the BATS domain binding curved
PtdIns(3)P-rich autophagic membrane, with puncta overlapping ATG16, LC3 and partially DFCP1.
One observation, split across two topology terms because the membrane it was observed on has
two names in GO. Both are now MODIFY: the omegasome pair to GO:1990462, the
phagophore-assembly-site pair to GO:0000407.

ATG14 is the only gene in this corpus carrying GO:0097629, so this slice says nothing about
how many annotations the term holds at GOA scale — only that the term should not have survived

23424.

ATG14 completes the audit

ATG14's six GO:0034045 annotations were the last ones sitting at ACCEPT, held over from the
first pass. They moved to GO:0061908 phagophore rather than GO:0000407, because the
evidence is about the isolation membrane — "Atg14 is present on autophagic isolation membranes"
(PMID:18843052) — and isolation membrane is a related synonym of GO:0061908, not of the
assembly site. Generalizing to GO:0000407 would also have been redundant: ATG14 already
carries it from an IDA (PMID:20713597).

phagophore membrane is additionally proposed on the ATG14 review, because the BATS-domain
result is specifically a membrane-binding observation and GO:0061908 can record it only as
the structure. Knock-on: ATG14's core_functions.locations carried both defective terms and
now carries GO:1990462 and GO:0061908.

The missing-term cases: propose the term, don't hold the annotation

ATG2A, ATG2B and worm atg-18 were initially left as ACCEPT on the grounds that there
was nowhere correct to move them. That was wrong. A review does not need an existing
destination to record a verdict — proposed_new_terms exists precisely for this, and holding
an annotation at ACCEPT because the ontology is incomplete records the opposite of what the
reviewer actually believes.

All five annotations now read MARK_AS_OVER_ANNOTATED, with the terms that ought to exist
authored on the gene reviews themselves:

gene annotations moved proposed_new_terms authored
human ATG2A 2 (IEA, EXP) phagophore membrane; phagophore rim; endoplasmic reticulum-phagophore membrane contact site
human ATG2B 2 (IEA, EXP) phagophore membrane; endoplasmic reticulum-phagophore membrane contact site
worm atg-18 1 (IBA) phagophore membrane

MARK_AS_OVER_ANNOTATED rather than MODIFY, because MODIFY requires
proposed_replacement_terms carrying real identifiers and there are none to give — inventing
an id would be worse than the problem. The verdict, the reasoning, and the terms that would
resolve it are all recorded; only the identifier is missing, and that is GO's to mint.

Each proposal maps onto a recommendation in the issue: phagophore membrane is R3,
phagophore rim and ER-phagophore membrane contact site are R7. The phagophore rim
proposal records the alternative label "phagophore edge" and cites GO:0097203 phagocytic cup lip as GO's existing precedent for the pattern; the phagophore membrane proposal takes the
collective sheet-plus-rim reading and flags explicitly that the sheet-versus-system question
of §4.2 is unresolved upstream, so an editor is not silently committed by the annotation.

Knock-on: ATG2A's core_functions.locations carried GO:0034045 alongside GO:0044232
organelle membrane contact site and GO:0005789 ER membrane, and ATG2B's carried it alongside
GO:0061908 and GO:0005789. In both cases the defective term was the least informative of
the three and was dropped rather than replaced.

The S. pombe ATG8-conjugation genes were moved to GO:0000407 rather than to a proposed
phagophore membrane because their cited evidence is punctum colocalization, which does not
distinguish the two — the sheet-versus-system decision does not arise on that evidence, so
there is nothing to propose.

Second pass: seven §5.4 genes reviewed in full

The genes above were reached through the term. The seven below were reached through the
issue's §5.4 triage table and reviewed in full — every annotation on each gene, not only
the GO:0034045 ones — which is a different and more demanding exercise: 496 annotations
across seven genes, of which 28 are GO:0034045.

gene annotations GO:0034045 destination
human ULK1 139 4 GO:0061908 phagophore
human RAB1B 75 2 GO:0000407 — see below
human ATG16L1 71 4 GO:0061908 phagophore
human ATG9A 68 5 GO:0061908 phagophore
human ATG12 55 4 GO:0061908 phagophore
human WIPI2 45 7 GO:0061908 phagophore
human STBD1 43 2 none — MARK_AS_OVER_ANNOTATED

That brings the audit to 59 assertions across 21 gene folders, none at ACCEPT.
phagophore membrane is proposed on ATG9A, WIPI2, ATG16L1 and ATG12, joining ATG14, ATG2A,
ATG2B and worm atg-18.

WIPI2 alone carries seven assertions from four evidence routes — EXP ×4, IDA, IBA, and an IEA
carrying UniProtKB-SubCell:SL-0221 — which makes it the heaviest single-gene load in the
audit and a good illustration of how the ambiguity compounds without anyone re-deciding it.

Two genes that do not fit the pattern

RAB1B is the one gene whose assertions do not go to GO:0061908. ATG9A, WIPI2, ATG16L1
and ATG12 are phagophore-membrane proteins. RAB1B is a secretory-pathway GTPase, and
PMID:20545908 locates its requirement upstream — at ER exit sites, with "the autophagic and
secretory pathways intersect at a level preceding the brefeldin A blockage". A membrane term
of any kind overstates that; the site-level GO:0000407 is the right destination, and the
functional claim is already carried by GO:2000785. This matters for the migration plan: a
blanket rewrite of GO:0034045 → phagophore membrane would put RAB1B on a bilayer it was
never shown to be on.

STBD1 confirms the triage table's "no evident support — review for removal" reading,
which is the opposite outcome to the Rab7 correction above. The primary text places STBD1 at
"enlarged perinuclear structures, co-localized with glycogen, the late endosomal/lysosomal
marker LAMP1 and the autophagy protein GABARAPL1" (PMID:20810658) — and GO:0048471 and
GO:0005789 already carry that. There is nothing left for a phagophore term to record, so
both assertions are MARK_AS_OVER_ANNOTATED with no replacement proposed. Not every
defective annotation needs a destination.

ULK1 is the strongest case in the whole audit for the condensate reading over the membrane
one, and it is worth stating separately because it cuts against the R3 proposal. ULK1 has no
membrane-binding domain; it is a soluble kinase whose concentration at the site is part of what
defines the site; and it already carries GO:0000407 for the structure and GO:1903349
omegasome membrane — the more precise membrane term — from the same study that supplies one of
its GO:0034045 assertions. For soluble PAS residents of this kind, the site term may simply be
correct and no membrane term needed.

A reviewer error worth recording: the target in its own WITH/FROM

Five of the seven genes have at least one IBA whose WITH/FROM column lists the annotation
target itself: WIPI2 (UniProtKB:Q9Y4P8 on GO:0034045), ATG16L1 (UniProtKB:Q676U5 on
GO:0034045), ATG12 (UniProtKB:O94817 on GO:0034045), ULK1 (UniProtKB:O75385 on GO:0034045
and on GO:0005737), and STBD1 (UniProtKB:O95210 on GO:0016020).

This pass initially flagged that as circular and marked the five sources
CIRCULAR_OR_REDUNDANT. That was wrong, and the correction matters more than the
observation did. An IBA is not a pairwise similarity transfer. Behind it is a PAINT
curator's IBD: the curator inspected the family tree and the MSA, read the experimental
annotations of all extant members, judged at which node the function arose — sometimes recent,
sometimes as deep as LUCA — and placed the assertion there. A gene with its own experimental
annotation for the term is one of the descendant evidences used to make that call, so it
legitimately appears among the sources of the IBA it later receives. Far from being circular,
it is a marker that experimental grounding exists on the target itself, and the IBA then says
something the experimental annotation does not: that the function is inherited rather than
lineage-specific.

Two related things the same misreading produced, both since corrected in these reviews:

The remaining objections to these annotations are unchanged, because none of them ever
depended on the propagation being weak: they are objections to the destination term
(GO:0034045) or to granularity (bare membrane, bare cytoplasm). The propagation itself
is sound in every one of these five cases, and the propagation_review blocks now say so.
The general guidance is written up in the IBA project.

Three further missing-term cases, from full review rather than from the term

Reviewing whole genes rather than single terms surfaced three more instances of the same
underlying pattern — a curator reaching for the nearest wrong term because the right one does
not exist. None of them involve GO:0034045.

gene annotation what the evidence shows term proposed
ATG16L1 GO:0120095 vacuole-isolation membrane contact site (IDA, PMID:28890335) ER–isolation membrane contacts in mammalian and C. elegans cells; mammals have no vacuole endoplasmic reticulum-phagophore membrane contact site
ATG16L1 GO:0016237 microautophagy (IDA ×2) ATG8 conjugation to single endolysosomal membranes (CASM); no invagination, no cargo uptake protein lipidation at single membranes
RAB1B GO:0090557 establishment of endothelial intestinal barrier (IMP NOT) intestinal epithelial tight junctions establishment of intestinal epithelial barrier

The ATG16L1 contact-site case is the same gap already proposed on ATG2A and ATG2B and modelled
in modules/phagophore_organelle_contact_site.yaml, now reached from a third direction:
GO:0120095 is the ontology's only isolation-membrane contact site term, and it is the yeast
VICS. The organelle membrane contact site branch has ER–vacuole, ER–endosome, ER–lysosome,
ER–plasma membrane, ER–trans-Golgi and ER–lipid droplet children, and no ER–phagophore child.

Two adjacent findings that are not missing-term cases, recorded here because they came from
the same sweep:

Qualifier inconsistency across the conjugation complex

ATG12 annotates GO:0019776 with contributes_to. ATG5 and ATG16L1 use enables for the same
complex-level reaction. All three are subunits of an E3-like conjugate that none of them
performs alone, so contributes_to is right and the other two should follow — and ATG12 has the
best claim of the three to a specific contribution, since the high-affinity ATG3-binding surface
is exclusive to it. Recorded as a question on the ATG12 review rather than acted on, since it is
a consortium-wide convention rather than a gene-level call.

What this slice supports and does not

Supports. That the term is accepted uncritically even under review; that the failure is
compositional rather than evidential (reviewers substitute pathway membership for
localization); that the amplification chain is real and visible at small scale; that every
assertion in the slice turned out to have a destination, either an existing term or one worth
proposing; and — from the GO:0097629 finding — that the #23424 sweep was incomplete in a way
that is checkable from the ontology alone.

Does not support. Anything about the ~2,090 assertions not in this corpus. This slice is
biased toward genes someone chose to review or that the issue itself named. It is a
demonstration that the per-assertion method in Stage 5 works and produces corrections, not a
sample from which the full audit's outcome can be projected.

The two passes differ in a way that bears on that caveat. The first pass was 19 IEA/IBA out of
31, and a single gene (ATG14) supplied 10 of them. The second pass is 15 experimental out of
28 — 10 EXP and 5 IDA — spread across seven genes, none supplying more than seven.
Combined, the audit is 27 experimental assertions out of 59. The objection that this
failure mode is confined to automatic pipelines does not survive that: most of the
second-pass assertions were made by curators reading papers.

Regenerating the inventory

The 59-assertion inventory comes from scanning genes/*/*/*-goa.tsv for the three terms and
joining against the review.action in the corresponding *-ai-review.yaml. The general
condensate-space scan in
the GO and annotation audit uses the same approach over a wider
term list.