SL-0090 Cytoskeleton

IN_PROGRESS PIPELINE

Species: human, mouse, worm, SCHPO, BETPN, DORPE, DROME

SL-0090 Cytoskeleton → GO:0005856

59 SL-unique annotations reviewed, 16 with a hard issue (27%). This is where the SL and
SPKW failure modes meet: the flagged cases split cleanly into two patterns, one
of which is SPKW's regulatory conflation wearing a cellular-component costume.

Pattern A — under-specification (the SL-0162 pattern)

The majority. cytoskeleton where the evidence supports a named filament system:

gene proposed instead
SCHPO mid1 GO:0110085 mitotic actomyosin contractile ring, GO:0071341 medial cortical node
human ABRAXAS2 microtubule / spindle-pole components
human BAIAP2L1 GO:0015629 actin cytoskeleton
human CFAP61 GO:0005930 axoneme ("already annotated with stronger evidence")
human GDPD2 actin filaments
human ABRA actin-specific term "already independently supported on the same record"
human ACTR10 GO:0005869 dynactin complex

Pattern B — association is not residence

This is the interesting one, and it has no analogue in SL-0162:

human SGCA — "SGCA is not itself a cytoskeletal component but rather a sarcolemmal
membrane protein that indirectly associates with the cytoskeleton through the
dystrophin-glycoprotein complex."

human SGCE — "SGCE is transmembrane protein in DGC that indirectly associates with
cytoskeleton, not a cytoskeletal protein per se. The term 'cytoskeleton' suggests SGCE is
part of cytoskeletal structure itself."

human ACTL7B — "asserts a compartment that no observation supports, derived from a
by-similarity statement"

Here the term is not merely vague — it is wrong in kind. A transmembrane protein that
binds the cytoskeleton is not located in it. This is the cellular-component counterpart of
SPKW's "regulatory conflation" (gene regulates X, annotated to X): gene binds X, annotated
to X
.

The pattern generalises beyond these three. The corpus's SL-0090 population is full of
cytoskeletal regulators and modifiers rather than components — ABL1 (kinase), HDAC6 and
SIRT2 (tubulin deacetylases), INPP5D (phosphatase), FERMT2 and CD2AP and BAIAP2 (adaptors),
GAPDH (moonlighting glycolytic enzyme). Each is plausibly at the cytoskeleton while doing
its job, so none is a clear error; but "cytoskeleton" is carrying an implication of
membership that the evidence does not license. Adjudicating that set is the natural next
batch and would need per-gene reading, not a rule.

Batch reviewed under this subproject

Six annotations moved ACCEPT → MARK_AS_OVER_ANNOTATED, all Pattern A, all with a strictly
more specific term already present from independent evidence:

gene already carries source
human MAPT GO:0045298 tubulin complex; GO:0005874 microtubule IDA
human MAP7 GO:0005875 microtubule associated complex; GO:0005874 microtubule TAS
mouse Tuba1a GO:0035371 microtubule plus-end; GO:0005881 cytoplasmic microtubule IMP, IDA
human ACTG2 GO:0015629 actin cytoskeleton IBA
human WIPF1 GO:0005884 actin filament; GO:0015629 actin cytoskeleton IBA
worm che-3 GO:0005868 cytoplasmic dynein complex; GO:0005930 axoneme IBA

Correction: pattern B was already solved, by existing terms

An earlier version of this page said the pattern B cases "were left alone deliberately"
pending a policy decision about located_in semantics for peripheral association. That was
wrong on both counts.
Re-reading those reviews:

Both had real destinations, both already used them. GO has GO:0016010,
GO:0016011 dystroglycan complex, and GO:0016012 sarcoglycan complex — a protein that
associates with the cytoskeleton through a named complex can be annotated to the complex,
which says exactly what the evidence supports and drops the false implication of membership.

The generalisable lesson is the same one the GO:0034045 audit
learned the hard way: check for a destination before declaring there isn't one. Pattern B
does not need a new relation or a policy ruling in the cases seen here. It needs the complex
term the protein actually belongs to.

Whether pattern B ever needs a relation change remains open — a protein that binds the
cytoskeleton without belonging to any named complex would have nowhere to go — but no such
case appears in this corpus.

Note on redundancy

SL-0090 has the highest redundancy in the corpus — 80% of its SL-unique annotations sit on
genes that already carry a more specific cytoskeletal term. Yet before this batch the issue
rate was identical in both groups (17% redundant, 17% not). That is the single strongest piece
of evidence against the deduplication rule proposed in the SL project's opening
draft.