Molecular Adaptor Activity Annotation Review

SCOPING PIPELINEBIOLOGY_DOMAIN

Species: human

Molecular Adaptor Activity Annotation Review

Bottom line (scoping). GO:0005515 protein binding records that a protein
sticks to something, not what it does; our curation guidelines tell reviewers
to replace it with an informative molecular function, and for a bridging
protein that replacement is GO:0060090 molecular adaptor activity (exact
synonym "binding, bridging") or one of its children. This project will work
out the decision rules for making that protein binding → adaptor call
consistently: which child term to pick, how to record the two partners being
bridged, how to tell an adaptor from a scaffold / sequestering / carrier
protein, and how much weight to give the large body of phylogenetically
propagated adaptor annotations already in GOA. Nothing has been audited yet —
this page captures the term landscape, the GO guidelines, the propagation
picture, and a seeded phase-1 anchor gene set; no annotations have been
adjudicated yet.

Scope

In scope: experimental and propagated annotations to GO:0060090 molecular
adaptor activity and its subtree (chiefly GO:0030674
protein-macromolecule adaptor activity and children), and bare GO:0005515
protein binding rows that the literature shows to be adaptor/bridging
functions and should be modelled as such.

The protein-binding policy
(src/ai_gene_review/validation/protein_binding_policy.py) is the governing
machinery here, not merely an out-of-scope exclusion: its
COMPLIANT_ACTIONS = {MODIFY, REMOVE, UNDECIDED, PENDING} and its docstring
prefer MODIFY to an informative MF over REMOVE, so it is exactly what the
protein binding → adaptor MODIFY in this project runs through. Genuinely
uninformative GO:0005515 rows with no bridging role are the REMOVE case of the
same policy.

Out of scope: the distinct but neighbouring molecular functions below, each with
its own GO-CAM guideline — a reviewer should not collapse them into "adaptor":

Term landscape

Root term: GO:0060090 molecular adaptor activity — "The binding activity
of a molecule that brings together two or more molecules, permitting those
molecules to function in a coordinated way." Exact synonym "binding,
bridging". Adaptors are non-catalytic by definition — with one standing
exception inside the subtree: the SNARE pair GO:0005484 (now
helical-bundle zippering fusogenic activity) and GO:0005483 sit under
GO:0030674 yet describe a fusogenic activity, so they are the case where an
"adaptor" subtree term is not merely a passive bridge. Flag them when they come up.

The protein-adaptor workhorse is GO:0030674 protein-macromolecule adaptor
activity, whose 16 children are where a reviewer usually lands:

Child term id
signaling adaptor activity GO:0035591
cytoskeletal adaptor activity GO:0008093
cargo adaptor activity GO:0140312
autophagy cargo adaptor activity GO:0160247
enzyme-substrate adaptor activity GO:0140767
chromatin-protein adaptor activity GO:0140463
kinetochore adaptor activity GO:0140483
protein-membrane adaptor activity GO:0043495
protein-RNA adaptor activity GO:0140517
guanyl-nucleotide exchange factor adaptor activity GO:0005091
receptor clustering activity GO:0141175
molecular condensate scaffold activity GO:0140693
ubiquitin-like protein reader activity GO:0140035
microtubule site clamp GO:1990644
helical-bundle zippering fusogenic activity (was SNAP receptor activity) GO:0005484
soluble NSF attachment protein activity GO:0005483

Relations were read from QuickGO on 2026-11-07; the table may mix direct
children with grandchildren (e.g. GO:0160247 is under GO:0140312, not a
direct child of GO:0030674), and the sidecar's branch_terms list surfaces 22
core_adaptor terms where the experimental evidence actually sits (including
GO:0035615, GO:1990756) — so when picking a term, prefer the specific
descendant the evidence supports, not only these headline children.

GO:0035591 signaling adaptor activity: "…brings together two or more
molecules in a signaling pathway… Adaptor molecules themselves do not have
catalytic activity."

Other direct children of GO:0060090 outside the protein-macromolecule branch:
GO:0030533 triplet codon-amino acid adaptor activity (tRNA), GO:0106260
DNA-DNA tethering activity, GO:0140369 dbTF DNA-binding stabilizer activity,
GO:0140177 membrane-membrane adaptor activity.

GO annotation guidelines (GO-CAM)

GO maintains dedicated guidelines, available through the Noctua guideline set
(Molecular_adaptor_activity, How_to_annotate_molecular_adaptors). Key
modelling rules:

Propagation / IBA picture

Adaptor activity is heavily propagated by homology, so most existing rows
are computational, not experimental. From the QuickGO annotation-statistics API
(retrieved 2026-11-07; exact queries, URLs and facet counts in
data/propagation-stats.yaml).
The two figures below are over different roots (GO:0030674 vs GO:0060090)
and so are not a single comparable breakdown:

Consequences for review (per the IBA guidance in CLAUDE.md and
IBA_REVIEW):

Phase 1 anchor set

Seeded from QuickGO (2026-11-07): every human gene product carrying at least
one experimental annotation (evidence under ECO:0000269 — IDA, IPI, IMP,
IGI, IEP, EXP) to GO:0060090 or a descendant. 1,289 annotation rows resolve to
666 distinct gene products. This is a worklist, not a set of curation
decisions — the full table, grouped by branch and with per-term evidence and
references, is in genes.yaml.

Genes are tagged with one or more branches so the sub-domains can be scoped
and worked separately. The branches are a pragmatic surface-role grouping, not
strict is_a cuts, and they deliberately cross the ontology: the whole
reader sub-branch nests under GO:0140463 (itself a core_adaptor term), and
GO:0005078 MAP kinase scaffold activity is_a GO:0035591 signaling adaptor
(so it is tagged core_adaptor, not scaffold, despite its "scaffold" name). A
gene may be in more than one branch, so the counts sum above 666:

Branch Genes What it covers
core_adaptor 516 protein-macromolecule / signaling / cargo / enzyme-substrate / ubiquitin-like ligase-substrate adaptors — the primary protein binding → adaptor target
reader 114 histone, RNA and ubiquitin/UFM1 modification "reader" activities (epigenetic/RNA readers; a distinct curation domain)
scaffold 45 GO:0140693 molecular condensate scaffold activity (in-branch; see CONDENSATES)
tether_anchor 17 organelle membrane tether / anchor / docking and chromatin-loop anchoring activities

178 of the 666 already have a genes/human/<symbol>/ folder in the repo
(flagged existing_folder: true in the sidecar) — these are the natural
starting points, though a folder does not mean the adaptor row was reviewed.
Three genes in this worklist carry a COMPLETE review (flagged reviewed: true):
GRB2 (reviewed independently, before this project), and LDLRAP1 and RPTOR
(reviewed here). The reviewed flag — not existing_folder — is the filter for
picking un-reviewed genes; its semantics are defined in the sidecar's source
block.

Proposed phase-1 cut: the core_adaptor branch (516 genes). The reader
branch is large and mechanistically distinct (recognising a modified residue to
recruit downstream machinery) and overlaps the chromatin/epigenetics work, so it
is a candidate for its own phase. Because the branches cut across the ontology
(above), the cut is a work-planning convenience, not an ontological claim; the
GO:0032947 scaffold exclusion is the only true out-of-branch boundary.

Decision rules (draft)

  1. Bare protein binding that is a bridging function → MODIFY with the
    adaptor term in proposed_replacement_terms     (the protein-binding policy's
    preferred action). Only when there is no bridging role does the same policy's
    REMOVE apply. Note that a single binary interaction (an IPI hit) evidences
    binding, not the bridging activity — so refine such a row to a specific
    binding term the partner supports (e.g. protein kinase binding,
    AP-2 adaptor complex binding); assert the adaptor activity only where an
    assay shows the bridging, and do not stack a second, unsupported adaptor-activity
    row on top of one already present by IDA.
  2. Pick the most specific descendant of GO:0030674 the evidence supports
    (the sidecar surfaces 22 core_adaptor terms where the experimental evidence
    sits — e.g. GO:0035615, GO:1990756 — beyond the 16 headline children);
    fall back to GO:0060090 only when nothing more specific fits.
  3. Distinguish adaptor from sequestering and carrier activities, and from the
    one true out-of-branch scaffold term GO:0032947. GO:0140693 molecular
    condensate scaffold activity is inside the adaptor subtree — defer to
    CONDENSATES's rules for it rather than excluding it.
  4. Name the bridged partners (has input) in the review rationale. This is the
    adaptor's inputs, a separate thing from the MF-to-MF relation (provides input for / directly positively regulates) that only a GO-CAM captures; do not
    conflate the two.
  5. A general adaptor term (GO:0030674) may be kept beside a more specific
    child only when it covers a distinct bridging event the child does not
    (e.g. GRB2 keeps GO:0030674 for LAT-microcluster organisation separate from
    its RTK/GEF adaptor children; RPTOR keeps it for Rag-GTPase lysosomal
    recruitment separate from its enzyme-substrate adaptor activity). Otherwise a
    parent beside its own descendant is TOO_GENERAL — state the distinct event
    or drop the parent.
  6. Treat IBA/TreeGrafter adaptor rows as inference: verify the family is
    coherent and the target is not a divergent paralog before ACCEPT.

Plan / open items

Sources