Apoptosis Readout Over-Annotation

This note isolates the APOPTOSIS cases we care about most: annotations where
the paper assayed a late cell-death endpoint, such as TUNEL, Annexin V,
cleaved caspase-3, DEVDase activity, PARP cleavage, or a viability counter,
and the GO row then asserted that the assayed gene product was part of an
apoptotic pathway.

The important distinction is narrower than "all apoptosis cleanup". Generic
protein binding rows on AKT1/MAPK1, broad protein kinase activity imports,
and stimulus-response rows propagated to caspases 3/9 are over-annotations, but
they are not the specific assay-to-apoptosis pattern. The readout pattern is:

perturb gene X -> more or less cell death/caspase/TUNEL signal
               -> annotate X to apoptotic signaling or apoptosis regulation

That chain is safe only when X performs a proximal apoptotic step. It is
usually too strong for survival kinases, transcriptional regulators,
metabolic proteins, autophagy proteins, and broad stress-response factors.

Existing Mined Set

The ASSAY_TO_FUNCTION publication miner has already produced a useful seed
set in projects/ASSAY_TO_FUNCTION/reports/paper_readout_matches.tsv. Filtering
to rows where:

• readout_class = APOPTOSIS_CASPASE
• aligned = True

gave 72 rows whose source paper used a caspase/apoptosis readout and whose GO
term was itself apoptotic or caspase-related.

There were 17 mouse rows in this exact mined set. That action distribution is
the core lesson: apoptosis readouts are usually not false positives, but they
often support a contextual, non-core phenotype rather than a direct apoptosis
function.

High-Signal Reviewed Cases

Pending Or Worth Rechecking

projects/ASSAY_TO_FUNCTION/reports/flagged_candidates.tsv contains a Tier 2
block for ACCEPTed BP calls aligned to APOPTOSIS_CASPASE, with the default
warning:

rubric default is non-core unless gene is in the machinery

Most are probably defensible; the point is to re-ask whether a direct step was
demonstrated rather than trusting a caspase/TUNEL endpoint. The mouse rows from
that block are:

Clearance-Defect Variant

A related but distinct failure: the perturbed gene acts in the engulfing
cell, not the dying one. Knocking out an apoptotic-cell receptor stops corpses
from being cleared, they accumulate, and TUNEL or cleaved-caspase-3 counts go
up. Read naively, that looks like "more apoptosis", and the gene gets an
apoptotic-process term. The correct term is GO:0043652 engulfment of apoptotic
cell, whose usage note keeps it separate from processes in the dying cell.
Orthology transfer (ISS/ISO) then copies the error to other species.

The Megf10 pair above is the first case. Iram et al. 2016 (PMID:27170117)
show impaired astrocyte phagocytosis ex vivo and gain of phagocytosis on Megf10
expression, and the full text (PMC4863057) titles Figure 1 "Megf10 is necessary
for apoptotic cell uptake by astrocytes, and its deficiency results in
accumulation of apoptotic cells in the developing CB".

A second shape of the same error appears in C. elegans. There, engulfment
mutants such as ced-1(e1735) are routinely used as a background that makes
corpses persist so they can be counted. Papers about other genes (CED-8,
PMID:24225442; CSP-1, PMID:23505386) were annotated to ced-1 itself: ced-1 got
GO:1902742 apoptotic process involved in development and GO:1904747
positive regulation of apoptotic process involved in development, although
those papers attribute the extra corpses to the engulfment defect.

C. elegans also shows where the line sits. Engulfing cells genuinely promote
the death of cells that receive weak death signals, and ced-1 expressed in
engulfing cells rescues this (PMID:11449278, PMID:11449279). So for ced-1 the
positive regulation term is right (kept, but sourced to the right papers),
while the process term is not (changed to positive regulation).

Precomposition update (2026-11-10). The positive-regulation replacements and the kept
GO:1904747 rows above now use GO:0043065 positive regulation of apoptotic process.
GO:1904747 is that term plus the developmental context of GO:1902742, and round 6 of
INVOLVED_IN_PRECOMPOSITION drops that context for the
engulfment genes. The sign correction this page argues for (regulation from the engulfing
cell, not participation in the dying cell's program) is unchanged.

The background-genotype rows come in batches. A QuickGO query by reference
(2026-10-05) shows that the CED-8 paper (PMID:24225442) gave GO:1902742
(IMP and IGI) to all seven engulfment genes it used as backgrounds: ced-1,
ced-2, ced-5, ced-6, ced-7, ced-12 and ttr-52. The CSP-1 paper
(PMID:23505386) gave GO:1904747 to ced-1, ced-6 and ced-7 alongside the
genuine death genes (ced-3, ced-4, egl-1, csp-1/2/3). ced-1, ced-6,
ced-7 and ttr-52 are reviewed. The replacement depends on the gene: the
positive-regulation term needs its own evidence that the gene promotes death
(Reddien 2001 and Hoeppner 2001 cover the classic engulfment genes). ttr-52,
identified later, has none, so its rows are removed rather than modified.
All seven genes are now reviewed. For ced-2, ced-5 and ced-12 the
positive-regulation replacement (now GO:0043065) rests on Reddien 2001, whose abstract defines the
engulfment genes it studies as ced-1, ced-2, ced-5, ced-6, ced-7, ced-10 and
ced-12 and states that "mutations in engulfment genes alone allow the
survival and differentiation of some cells that would normally die".

Detection idea (not yet run): genes carrying both an apoptotic-cell
clearance term (GO:0043652, GO:0043277) and an apoptotic-process term from
the same PMID, or an apoptotic-process term on a known engulfment receptor or
bridging molecule (MERTK, AXL, TIMD4, GAS6, MFGE8, ITGB5, ELMO1, DOCK1, GULP1,
CD36, ADGRB1, STAB2, LRP1).

Triage Rules

• Ask what entity performs the apoptotic step. If a caspase, BAX/BAK pore,
  BCL2-family inhibitor, APAF1 apoptosome, FADD/DISC protein, or IAP is the
  acting entity, the apoptosis term may belong there. If the tested gene only
  changes whether those entities fire, default to non-core or over-annotated.
• Ask which cell the gene acts in. If the gene works in the phagocyte
  (an engulfment receptor, bridging ligand or engulfment signalling protein),
  extra TUNEL/caspase-positive cells after loss of function usually mean
  uncleared corpses, not extra death. Annotate clearance, not apoptosis.
  This also applies to rows qualified acts_upstream_of_or_within: when the
  measured change is in clearance, the gene is neither upstream of nor within
  the apoptotic process.
• Treat late markers as convergent. TUNEL, Annexin V, cleaved caspase-3,
  PARP cleavage, DEVDase reporters, mitochondrial depolarization, and viability
  assays are endpoints. They prove that the perturbation moved cells toward or
  away from death; they do not name the direct pathway edge.
• Keep direct transcriptional control distinct from cell death. A TF or
  chromatin regulator can directly activate or repress an apoptotic target gene
  program. That is stronger than a viability effect, but it is still a
  transcriptional role first and should not be collapsed into generic
  GO:0006915.
• Prefer the nearest mechanism. For AKT, PTEN, ERK, beta-catenin, and
  HDACs, use kinase, phosphatase, Wnt, or chromatin terms as the core model;
  keep apoptosis only when a proximal apoptotic branch is shown.
• Do not penalize core apoptosis proteins for using apoptosis assays. For
  effector and initiator caspases, APAF1, BCL2-family proteins, FADD, XIAP,
  and DIABLO, caspase/TUNEL assays are expected orthogonal evidence. The
  concern is only when the gene sits upstream, beside, or downstream of the
  death pathway.

Next Slice

1. Start from projects/ASSAY_TO_FUNCTION/reports/flagged_candidates.tsv
   rows 60-83 and rerun the comparator check against the gene's core function.
2. Expand the filter to apoptosis terms whose cached papers match
   VIABILITY_PROLIFERATION or MITO_MEMBRANE_POTENTIAL, since many
   cell-death papers use viability or mitochondrial potential without a
   caspase-specific string.
3. Query mouse GOA for broad GO:0043065 positive regulation of apoptotic process and GO:0043066 negative regulation of apoptotic process; the
   survival-kinase failure mode will be denser there than under exact
   GO:0006915.