ASF1 curation notes

2026-09-02 Update: audit fix — erroneous proposed_replacement_terms entry

While auditing the existing review, found that the GO:0005515 protein binding IPI
annotation (original_reference_id: PMID:11404324, documenting the ASF1–Cac2/CAF-1
interaction) carried a bogus proposed_replacement_terms entry — id GO:0030515 paired
with the label "positive regulation of cation transport" — alongside the correct
GO:0042393 histone binding.

The id and the label did not go together. GO:0030515 is in fact snoRNA binding
("Binding to a small nucleolar RNA"), confirmed against the committed ontology cache
(cache/ontologies/go.tsv → GO:0030515<TAB>snoRNA binding) and against OLS. The string
"positive regulation of cation transport" does not occur anywhere in the ontology cache
either — though that cache is deliberately sparse (go.meta.json: 6820 terms, "Only terms
used in annotations"), so its absence there does not by itself establish that no such GO
label exists anywhere in the ontology. What is established is the part that matters: it is
not the label of GO:0030515. So the entry was a mismatched id+label pair, not a
copy/paste of a real term from another review — precisely the failure mode CLAUDE.md warns
about, where a plausible-looking label conceals a wrong id. Note that
src/ai_gene_review/tools/fix_labels.py deliberately skips proposed_replacement_terms
(they "may contain hallucinated GO IDs"), so no automated label check would have caught
this.

The removal stands either way: snoRNA binding is as unrelated to ASF1's
histone-chaperone / chromatin-assembly biology as ion transport would have been. ASF1 has
no documented snoRNA-binding or ion-transport role in any of the cited literature, the
deep-research reports, or UniProt (P32447). Removed the erroneous term; kept the
GO:0042393 histone binding replacement suggestion, which is well supported by ASF1's
defining H3-H4 histone-binding activity documented throughout this review (e.g.
PMID:15840725).

Left open (flagged as non-blocking by review, needs curator judgment rather than a
mechanical edit): the retained GO:0042393 histone binding replacement does not match
that particular IPI's own evidence, which is an ASF1–Cac2/CAF-1 protein interaction, and
the sibling protein-binding IPIs carry no replacement terms at all; the supporting_text
on that annotation is also the paper title rather than substantive evidence.

No other changes made — the rest of the review (annotation actions, core_functions,
description) is well-supported by the cited evidence and deep-research reports.

2026-09-07 Update: resolved the retained histone-binding replacement, and corrected two misidentified interactors

Follow-up to the review's blocking item on the section above. Two things were wrong, and
they turn out to be the same mistake.

1. The interactors were misidentified. Both the 2026-09-02 note and the review's own
summaries described the WITH/FROM accessions on these IPI rows as CAF-1 subunits
("Cac2", "Cac1"). Checked both against UniProt directly:

• UniProtKB:P32479 is HIR1_YEAST / Hir1 (YBL008W), a subunit of the HIR complex,
  not CAF-1. UniProt's own SUBUNIT annotation cites the very reference on this row: "The
  HIR complex interacts with ASF1 (PubMed:11404324, PubMed:11412995)."
• UniProtKB:Q04003 is SAS4_YEAST / Sas4 (YDR181C), a subunit of the SAS
  (something-about-silencing) complex, not Cac1. UniProt records "Interacts with ASF1",
  consistent with PMID:11731480 —
  the paper title names CAF-I, but the accession actually recorded in WITH/FROM is the
  SAS-I subunit.

Neither the ASF1–Hir1 nor the ASF1–Sas4 interaction is a CAF-1 interaction. Corrected both
summaries (PMID:11404324 and PMID:11731480 rows) accordingly. The genuine ASF1–Cac2
CAF-1 link is real and is discussed elsewhere in the review (IBA row for
replication-coupled nucleosome assembly); it is simply not what these two IPI rows record.

2. The retained GO:0042393 histone binding replacement was therefore unsupportable.
proposed_replacement_terms is a machine-readable instruction attached to a specific
annotation row, so retaining it would have turned an ASF1–Hir1/Sas4 complex-subunit
interaction into a histone-binding IPI. Hir1 and Sas4 are chromatin-regulatory complex
subunits, not histones. Removed the entry. Nothing is lost by doing so: ASF1's histone
binding is independently annotated (IEA from GO_REF:0000002, and IBA), and the sibling
GO:0005515 IPI rows carry no replacement terms either, so the row is now consistent with
them. The reason text on the row stands on its own and now states why no replacement is
proposed.

Also softened the ontology-cache claim in the section above: cache/ontologies/go.tsv is
restricted to terms used in annotations (6820 terms per go.meta.json), so a string's
absence from it cannot show the string is "not a GO label at all". The conclusion that
matters — that it is not the label of GO:0030515 — is unaffected.

Still open (pre-existing, non-blocking): the supporting_text on the PMID:11404324 row
is the paper title rather than substantive evidence. The cached publication is
abstract-only, so a substantive verbatim quote establishing the Hir1 interaction is not
available from the cache; left as-is rather than paraphrasing, since supporting_text must
be verbatim.

2026-09-08 Update: removed unsupported telomere-maintenance replacement; parity + quote fixes

Follow-up to the review's remaining blocking item. The GO:0006282 regulation of DNA repair NAS row (PMID:27222517) was the file's only surviving proposed_replacement_terms,
and it proposed GO:0000723 telomere maintenance — a real, non-obsolete id
(cache/ontologies/go.tsv → GO:0000723<TAB>telomere maintenance) but unsupported by the
row's own evidence. The row's reason and supporting_text are entirely about Rad53
dephosphorylation and DNA-damage-checkpoint recovery; grep -ci telomere over the full
text of publications/PMID_27222517.md (full_text_available: true) returns 0. This is
the same artifact class removed earlier in this PR — a plausible id attached to a row whose
evidence does not support it. Removed the replacement; the checkpoint-recovery biology is
already captured by the sibling GO:2000002 (negative regulation of DNA damage checkpoint)
annotation from the same reference, which is ACCEPTed. Also replaced that row's
supporting_text (previously the paper title) with a substantive verbatim quote from the
full text about the Rad53 dephosphorylation role.

Also (non-blocking parity fix): the PMID:11731480 IPI summary now names both GOA
WITH/FROM accessions — UniProtKB:P40963 (Sas2, confirmed via genes/yeast/SAS2/SAS2-uniprot.txt
AC P40963 / GN Name=SAS2) and UniProtKB:Q04003 (Sas4) — both SAS-I complex subunits,
rather than Sas4 alone.

Note: the 2026-09-02 "Left open" paragraph above (referring to "ASF1–Cac2/CAF-1") is
superseded by the 2026-09-07 interactor correction — those accessions are Hir1 and Sas4,
not CAF-1 subunits.

2026-09-10 Update: removed fabricated GO:0016585 label from free-text reason

The 2026-09-08 review found one more instance of the artifact class this PR was opened to
remove — a wrong GO id carried by a plausible-sounding label — this time in a free-text
reason rather than in proposed_replacement_terms, which is why neither the term
validator (which hard-checks only core_functions) nor the label-fixing tooling surfaced
it.

The GO:0006351 DNA-templated transcription IEA row (GO_REF:0000043) closed its reason
by suggesting three more specific terms: "GO:0006357, GO:0032968, GO:0016585 polymerase II
elongation". The first two are correct, but GO:0016585 is not an elongation term at all:
it is obsolete, and its label is "chromatin remodeling complex"
(cache/ontologies/go.tsv → GO:0016585<TAB>obsolete chromatin remodeling complex;
independently confirmed via OLS, which reports is_obsolete: true and the obsoletion
reason "its definition no longer reflects and cannot be modified to be consistent with the
current state of knowledge"). Beyond being obsolete, it is a cellular-component term
being offered as a more specific alternative for a biological-process annotation.

The fix is a deletion rather than a substitution. GO:0032968 (positive regulation of
transcription elongation by RNA polymerase II) is already named in the same sentence and is
already annotated on this gene with IDA evidence, so the elongation point is covered;
proposing a guessed replacement id would repeat the original mistake.

Also in this pass (non-blocking items from the same review):

• Documented here for the first time the PMID:16554755 softening made on 2026-09-08 but
  omitted from that day's journal entry: the IPI summary previously asserted interactions
  with histones "H3, H4, H2A, H2B", but no H2A or H2B accession appears among the twelve
  GOA WITH/FROM accessions for that reference, so the summary now says "histone proteins"
  without enumerating types. That summary has additionally been rewritten to describe the
  annotation itself rather than narrating what the review declines to assert.
• The GO:0006282 NAS row's reason no longer argues against telomere maintenance. That
  argument was written to justify removing the GO:0000723 replacement term on 2026-09-08;
  with the term gone from the file, the rebuttal had no referent for a reader. The
  substantive point that survives — that the checkpoint-recovery biology is captured by the
  sibling GO:2000002 annotation from the same reference — is retained.
• status flipped IN_PROGRESS → COMPLETE. This is not a judgement call: no annotation
  is PENDING (28 ACCEPT, 18 KEEP_AS_NON_CORE, 1 MARK_AS_OVER_ANNOTATED) and validation is
  warning-free, which is the schema's definition of COMPLETE. The repo's own
  status_manager.compute_status_from_file independently returns COMPLETE for this file.

2026-09-10 — IPI summaries realigned to their own WITH/FROM accessions

Review round of 2026-09-10 flagged two IPI rows whose summaries named interactors that are
not the accessions recorded on those rows — the same defect class as the P32479
"Cac2"→Hir1 and Q04003 "Cac1"→Sas4 corrections of 2026-09-07, and with the same origin:
the paper's title supplied a protein name where the accession should have. Both are
KEEP_AS_NON_CORE on a generic GO:0005515, so the actions are unaffected; only the
identity claims were wrong.

PMID:15755447 — "MCM2 helicase" → Rad53. The row's only WITH/FROM is
UniProtKB:P22216 [ASF1-goa.tsv, ECO:0000353]. ASF1's own UniProt IntAct block names that
accession RAD53 with NbExp=11 [ASF1-uniprot.txt:560], and the block's seven partners
(Hhf2, Hht2, Hir1, Rad53, Rtt106, Sas2, Sas4) include no MCM subunit at all
[ASF1-uniprot.txt:557-563]. The previous reason had built a mechanism on the misreading
("functionally relevant for histone delivery at replication forks"), which pointed the row
at the wrong pathway: ASF1–Rad53 is real, but it is checkpoint biology, already handled by
the GO:2000002 and GO:0006282 rows from PMID:27222517. The paper's title
("…regulation of MCM helicase by Tof1/Mrc1/Csm3…") is the likely source of the error.

PMID:16429126 — "DNA polymerase, MCM, and CAF-1 subunits" matched none of the three
accessions. The rows carry P11484, P22216 and P47171. P11484 is Ssb1, a
ribosome-associated Hsp70 [genes/yeast/SSB1/SSB1-uniprot.txt]; P22216 is Rad53, as
above. P47171 is not identifiable from any record committed to this repository, so the
summary now names the accessions and asserts no identity for it — per CLAUDE.md, an omitted
identification says "not established" where a guessed one would say something false.

Non-blocking items from the same review, addressed in this pass:

• PMID:19536198 — "multiple histone chaperone partners" was one Hsp70. The single
  WITH/FROM on that row is P11484 (Ssb1), so the description overstated both the number
  of partners and their nature. ASF1's histone binding is annotated independently and never
  rested on this row.
• GO:0005634 nucleus IDA — mismatched supporting_text replaced. The quote attached was
  about Rad53 dephosphorylation, which is checkpoint biology rather than localization
  evidence. It now reads "Asf1 associates with the histone H3–H4 heterodimer"
  [publications/PMID_27222517.md:124] — the chromatin complex that this Complex Portal IDA
  actually records. Worth noting for future auditors: grep -niE 'nuclear|nucleus|localiz'
  over that cache returns nothing despite full_text_available: true, so no direct
  localization quote is extractable from this paper at all. ACCEPT stands regardless — the
  Complex Portal curator worked from the full record, and GOA carries two further independent
  nucleus IDAs (PMID:11404324, PMID:22932476). Deleting supported_by outright was tried
  first and rejected: it raises a best-practice warning that flips the computed status back to
  DRAFT, which would have been a worse outcome than the nit it fixed.
• GO:0032968 — "directly demonstrates" softened. publications/PMID_22308335.md is
  abstract-only and grep -ci asf1 returns 0, so that phrasing claimed more than is visible
  here. Per CLAUDE.md the experimental annotation is not second-guessed on that basis;
  ACCEPT is unchanged and only the phrasing was overstated.

2026-09-12 — the GO:0005634 Complex Portal IDA is CPX-1322 RAD53-ASF1, not a chromatin complex

This supersedes the GO:0005634 bullet of 2026-09-10 above, which was wrong. That pass
swapped the row's supporting_text from the Rad53-dephosphorylation sentence to "Asf1
associates with the histone H3–H4 heterodimer", on the stated grounds that the latter named
"the chromatin complex that this Complex Portal IDA actually records". The committed records
say otherwise on both halves of that claim:

• ASF1 belongs to exactly one Complex Portal complex, and it is the Rad53 one.
  ASF1-uniprot.txt:625 reads DR ComplexPortal; CPX-1322; RAD53-ASF1 complex., and
  Complex Portal's only other contributions to this protein are the two Rad53-checkpoint BP
  terms GO:2000002 and GO:0006282 [ASF1-uniprot.txt:664,669] — the same
  PMID:27222517 rows already reviewed in this file. No chromatin complex is recorded.
• The row itself identifies no complex. The GOA line for this annotation
  (ASF1-goa.tsv, GO:0005634 / IDA / PMID:27222517 / assigned by ComplexPortal) has an
  empty WITH/FROM column, so "chromatin complex" was an inference, not a reading.
• The substituted quote is another paper's finding. publications/PMID_27222517.md:124
  reads "Asf1 associates with the histone H3–H4 heterodimer (English et al. 2006)" — the
  sentence is verbatim in the cache, so it passes the substring check, but it credits this
  reference with a claim the text itself attributes elsewhere.

The 2026-09-10 swap therefore traded a quote about the right complex for one about a
different association, and introduced an unsupported assertion in the same free-text reason
channel that had previously hidden GO:0016585. Restored the original
Asf1 facilitates dephosphorylation of Rad53 after DNA double-strand break repair quote
(verbatim, and already the supporting text on the sibling GO:2000002 row), and rewrote the
summary/reason to name CPX-1322 RAD53-ASF1 instead of a chromatin complex. ACCEPT
stands unchanged, as does the note that this cache contains no direct localization statement
and that the nucleus call rests on the curator's full record plus the independent SGD nucleus
IDAs on PMID:11404324 and PMID:22932476.

The general lesson, recorded for future passes on this file: a "mismatched quote" nit is not
automatically fixed by finding a more topical sentence. Where the annotation comes from a
complex-centric source, the complex identity in ASF1-uniprot.txt is the constraint, and an
off-topic quote about the right complex beats an on-topic quote about the wrong one.