PANTHER IBA family review — reviewed S. pombe genes

Family-level review of the Inferred from Biological Ancestor (IBA;
ECO:0000318, GO_REF:0000033) annotations on the 41 S. pombe genes curated in
the pombe review batches. IBAs are produced by GO Consortium PAINT curators
propagating experimental annotations across PANTHER family trees, so the
right place to judge them is the family/subfamily, not the individual gene.

Method (reproducible)

extract_iba_propagation.py reconstructs every IBA propagation from cached repo
data only (nothing hardcoded):

• the GOA with/from field gives the ancestral PANTHER node (PTN…) and the
  seed genes (experimentally annotated relatives that justified the IBA);
• each gene's PANTHER family + subfamily is read from its *-uniprot.txt;
• each UniProt seed is mapped to its subfamily via the cached
  interpro/panther/<FAM>/<FAM>-entries.csv member table;
• a propagation is flagged CROSS_SUBFAMILY when no seed shares our gene's
  subfamily (i.e. the function came only from sibling subfamilies),
  LOCALIZATION for cellular-component terms (which transfer worst across
  trees), GENE_NO_SUBFAMILY / NO_UNIPROT_SEEDS where mapping is incomplete.

Output: iba_propagation.tsv (one row per IBA, with our curation action joined).

IBA annotations analyzed: 160
  with UniProt seeds mappable to subfamilies: 90
  CROSS_SUBFAMILY (seeds only from other subfamilies): 36

The big caveat: CROSS_SUBFAMILY is triage, not a verdict

PANTHER subfamilies are very fine-grained, and true orthologs in different
species routinely land in different SFs. So the flag has a high false-positive
rate for broadly conserved functions. Among the 36 cross-subfamily hits, most
are unambiguously correct and were (correctly) ACCEPTed:

So the family lens confirms our per-gene calls were sound rather than
overturning them — and it did not surface additional genuine errors among
the ACCEPTed IBAs.

Confirmed over-propagations (the genuine errors)

The flag's true positives line up exactly with the three IBAs we removed/flagged
by hand, now substantiated at the family level:

1. pom1 — cytoskeleton (GO:0005856) → REMOVE ✓

• Family PTHR24058 (dual-specificity kinases); pom1 = SF132 (DYRK-family kinase Pom1).
• Node PTN008603465; the only mappable UniProt seed is human DYRK3 (SF35).
• "Cytoskeleton" is a DYRK3-subfamily property, not a Pom1/SF132 function — Pom1
  is a cortical/plasma-membrane kinase. Cross-subfamily transfer of a
  localization term. Over-propagation confirmed.

2. rqh1 — cytoplasm (GO:0005737) → REMOVE ✓

• Family PTHR13710 (RecQ helicases); rqh1 = SF153 (RecQ-like helicase BLM).
• Seeds only from SF105 / SF152 (other RecQ subfamilies).
• Rqh1 is experimentally nuclear; a sibling-subfamily cytoplasm localization
  transferred onto it. Over-propagation confirmed.

3. mid1 — septin ring organization (GO:0031106) → REMOVE (nuanced)

• mid1 has no PANTHER subfamily tag in UniProt (GENE_NO_SUBFAMILY), so the SF
  test could not fire. The with/from node is PTN001853010, seeded by
  Drosophila anillin (FB:FBgn0261385), C. elegans anillin
  (WB:WBGene00013038), and mid1 itself (PomBase:SPCC4B3.15 is mid1's own
  ORFName — not mid2, contrary to a first draft of this note).
• "Septin ring organization" is a genuine ancestral anillin function (the
  metazoan seeds really do organize septins), so at the family level the
  propagation is not spurious. The issue is pombe-specific sub-functionalization:
  the two pombe anillins split the job — mid2 organizes the septin ring,
  while mid1 positions the actomyosin ring — so the ancestral term does not
  describe mid1's retained role.
• The mid1 leaf in with/from traces to mid1's own experimental septin-ring
  annotation, i.e. the IMP from PMID:15385632 — whose full text (cached) we
  checked and found contains 0 mid1 / 6 mid2 mentions. That IMP REMOVE is solid;
  the IBA REMOVE is the weaker, sub-functionalization-based call (KEEP_AS_NON_CORE
  would also be defensible for the IBA alone).

A discrimination the family lens adds: "MAPK cascade"

Two genes carry an IBA to MAPK cascade (GO:0000165) and we gave opposite
actions — the family review shows both are correct:

• wis1 → ACCEPT. Wis1 is the MAP2K (Pbs2 ortholog) of the Sty1 SAPK
  cascade; "MAPK cascade" is its actual biology.
• cdc7 → MARK_AS_OVER_ANNOTATED. Cdc7 is the SIN initiating kinase; the SIN
  is a GTPase-regulated kinase relay, not a MAPK cascade. The term is
  over-propagated from STE-family relatives.

Ral2: loss annotations protect a heterogeneous family boundary

Ral2 belongs to PTHR43503:SF2 (Mds3-related fungal Kelch proteins), while
the broad PTHR43503 family also contains peroxiredoxin-6 subfamilies. PAINT
places explicit IRD loss annotations for peroxidase activity, cytosol, and cell
redox homeostasis on the Ral2/Mds3 node PTN005166285; those ancestral functions
do not propagate into Ral2's current GOA. The one surviving IBA, regulation of
conjugation with cellular fusion, is gained on that same node and is grounded by
Ral2's own experimental mating phenotype. ACCEPT is therefore appropriate.
The NO_UNIPROT_SEEDS;SINGLE_NODE_SEED flags describe the extractor's canonical-
UniProt mapping, not a lack of experimental grounding in the PAINT source set.

Bottom line

• 160 IBAs reviewed at the family level. Two clean over-propagations (pom1
  cytoskeleton, rqh1 cytoplasm — localization terms crossing subfamilies) and
  one sub-functionalization case (mid1 septin ring, where the ancestral
  anillin function partitioned to mid2 in pombe). All three were already caught
  in per-gene curation; the family lens substantiates the localization cases and
  reframes mid1 as sub-functionalization rather than a clean mis-attribution.
• No additional IBA errors were found among the ACCEPTed annotations; the
  remaining cross-subfamily flags are conserved functions (false positives of a
  deliberately sensitive flag).
• The recurring true-positive pattern is localization terms and
  paralog-specific functions crossing subfamily boundaries — the two things
  to scrutinize hardest in any IBA.

Files

• extract_iba_propagation.py — reproducible extractor/flagger.
• iba_propagation.tsv — per-IBA table (node, seeds, subfamilies, our action, flags).