RHEA Gap Case Reviews

Parent project: RHEA.md · Background: RHEA-EC-SPECIFICITY.md

These are reviews of concrete cases of gap class G4/G6: a reviewed
(Swiss-Prot) human enzyme carries an experimentally or curator-asserted RHEA
reaction, that reaction has no rhea2go mapping, and as a result the
protein's GO molecular function is missing or only ultra-high-level. They
were drawn from the reproducible candidate set built by
rhea_gap_finder.py (stream reviewed human enzymes → keep
those whose every annotated RHEA is absent from rhea2go and that carry a
complete EC → 115 human enzymes). Of those, the finder's strict filter (MF
catalytic terms absent or only enzyme-class roots, ignoring cofactor "binding"
terms) auto-flags the root/class cases PHYKPL and SULT6B1; B3GALNT2
and SAMD8 were selected by inspecting the same 115-entry list for enzymes
whose only activity term is a too-general class term or an off-target activity.
Each case was then verified by hand against UniProt, the rhea2go/ec2go
mappings, and a QuickGO term search.

Every identifier below is checkable: the RHEA id and reaction come from the
UniProtKB CC -!- CATALYTIC ACTIVITY line; the "no mapping" claim was confirmed
against the public rhea2go file; the "current GO MF" is the molecular-function
subset of the entry's GOA terms; "specific term exists?" is a QuickGO search.

Summary

* SAMD8 carries several unmapped RHEA reactions; RHEA:36079 is the
ceramide-phosphoethanolamine synthase reaction that defines its function.

These four span the gap taxonomy: a GO granularity gap where no term exists
(PHYKPL, B3GALNT2), a clean propagation gap where the right term already
exists but never reached the protein (SAMD8), and an evidence-limited gap
where the fill is real but the confidence is lower (SULT6B1).

PHYKPL — 5-phosphohydroxy-L-lysine phospho-lyase (Q8IUZ5)

• RHEA:34091 (O-phosphohydroxy-L-lysine = (S)-1-pyrroline-... + NH4(+) + phosphate), EC 4.2.3.134.
• rhea2go: no mapping. ec2go: EC 4.2.3.134 has no ec2go line either.
• Current GO MF: GO:0016829 lyase activity (the lyase root), plus
  pyridoxal phosphate binding and identical protein binding. So the only
  catalytic GO term is the top of the lyase branch — maximally uninformative.
• Evidence: human enzyme experimentally characterised — catalytic activity,
  cofactor (PLP) and kinetics in PubMed:22241472 (UniProt ECO:0000269). This is
  the gene mutated in phosphohydroxylysinuria.

Assessment. Neither GO pipeline can lift this protein off lyase activity:
RHEA has the precise reaction but it is unmapped, and the EC is absent from
ec2go. A QuickGO search finds no GO term for this activity (the nearest
analogue is GO:0050459 ethanolamine-phosphate phospho-lyase activity, a sibling
pyridoxal-phosphate phospho-lyase). The biology is solid and specific; GO simply
lacks the node.

Action.
- MODIFY the existing GO:0016829 lyase activity annotation — it is correct but
far too high-level to be the protein's curated function.
- proposed_new_terms: 5-phosphooxy-L-lysine phospho-lyase activity (a.k.a.
5-phosphohydroxy-L-lysine ammonia-(phosphate)-lyase), placed in the
carbon-oxygen / phospho-lyase grouping that already contains the sibling
GO:0050459 ethanolamine-phosphate phospho-lyase activity; definition tied to
RHEA:34091 and EC 4.2.3.134. This is the right long-term fix and would let the
RHEA pipeline propagate it.

This is the cleanest "ultra-high-level" case in the human set: the difference
between lyase activity and the actual reaction is the entire informativeness of
the annotation.

B3GALNT2 — β-1,3-N-acetylgalactosaminyltransferase 2 (Q8NCR0)

• RHEA:37667 (transfers GalNAc in β1,3 linkage onto a GlcNAc-β-terminated
  glycan of α-dystroglycan), EC 2.4.1.313.
• rhea2go: no mapping. ec2go: EC 2.4.1.313 has no ec2go line.
• Current GO MF: UDP-glycosyltransferase activity and
  acetylgalactosaminyltransferase activity — both generic class terms; neither
  states the β1,3 linkage or the dystroglycan acceptor.
• Evidence: experimentally characterised acceptor specificity in
  PubMed:23929950 (ECO:0000269). Disease relevance: loss of function causes
  muscular dystrophy-dystroglycanopathy MDDGA11 (a defect in the α-dystroglycan
  O-mannosyl glycan elongation, where B3GALNT2 adds the GalNAc-β1,3-GlcNAc
  unit).

Assessment. QuickGO has many specific GalNAc-transferase terms but none
for the β1,3 activity on the GlcNAc-terminated O-mannosyl glycan — the curated
terms are the generic class. RHEA encodes the precise acceptor, but with the
reaction unmapped and the EC outside ec2go, that specificity never reaches GO.

Action.
- proposed_new_terms: N-acetyl-β-glucosaminyl-glycoprotein
β-1,3-N-acetylgalactosaminyltransferase activity (acceptor = the
GlcNAc-β-terminated α-dystroglycan O-mannosyl glycan), child of
GO:0008376 acetylgalactosaminyltransferase activity, anchored to RHEA:37667 /
EC 2.4.1.313.
- Keep the generic acetylgalactosaminyltransferase activity as a parent until
the specific term exists; do not treat UDP-glycosyltransferase activity
as the curated function.

Medically important and mechanistically precise — a strong gap-fill candidate.

SAMD8 (SMSr) — ceramide phosphoethanolamine synthase (Q96LT4)

• RHEA:36079 (an N-acylsphing-4-enine + a 1,2-diacyl-sn-glycero-3-phosphoethanolamine = a ceramide phosphoethanolamine + a 1,2-diacyl-sn-glycerol), EC 2.7.8.-, the defining SMSr reaction (one of a
  family of unmapped CPE-synthase RHEA reactions on this entry: 36079/36080,
  42128/42129, …); plus a PE-PLC hydrolase reaction (EC 3.1.4.62).
• rhea2go: no mapping for the CPE-synthase reaction. ec2go: only the
  generic EC:2.7.8.- → GO:0016780 phosphotransferase activity, for other substituted phosphate groups.
• Current GO MF: hydrolase activity (ultra-high-level), sphingomyelin synthase activity, ceramide cholinephosphotransferase activity. It does
  not carry the term for its principal activity.
• Specific term EXISTS: GO:0002950 ceramide phosphoethanolamine synthase activity is already in GO — but is not annotated to SMSr.
• Evidence: SMSr characterised as the ER ceramide-phosphoethanolamine synthase
  (PubMed:19506037) with later mechanistic work (e.g. PubMed:33621517).

Assessment. This is the clean propagation gap: unlike PHYKPL/B3GALNT2 the
correct GO term already exists, yet the protein carries everything except it —
two SM-synthase-flavoured terms (SMSr is at best a weak SM synthase) and a bare
hydrolase activity. The RHEA reaction that would have delivered GO:0002950 is
unmapped, so the term fell through the cracks even though no new term is needed.

Action.
- NEW (add) GO:0002950 ceramide phosphoethanolamine synthase activity as
the core molecular function — the term exists; this is pure gap-filling.
- MARK_AS_OVER_ANNOTATED / MODIFY the bare GO:0016787 hydrolase activity
(uninformative; the PE-PLC activity, if kept, should use a specific
phospholipase-C-type term, not the root).
- Review sphingomyelin synthase activity: keep only as non-core if supported,
since SMSr's mammalian activity is predominantly CPE synthase, not SM synthase.

The highest-value action of the four because it requires no ontology work —
just apply an existing term the RHEA mapping should already carry.

SULT6B1 — sulfotransferase 6B1 / thyroxine sulfotransferase (Q6IMI4)

• RHEA:26422 (sulfo transfer from PAPS to substrate, producing adenosine
  3',5'-bisphosphate), EC 2.8.2.n2 (provisional).
• rhea2go: no mapping. ec2go: the provisional 2.8.2.n2 is absent; only
  EC:2.8.2.- → GO:0008146 sulfotransferase activity applies.
• Current GO MF: GO:0008146 sulfotransferase activity only — the class term.
• Evidence caveat: the human catalytic-activity annotation is by similarity
  (ECO:0000250|UniProtKB:P0CC03), and the EC is provisional (n2). Substrate
  preference (thyroxine / thyroid hormones) is inferred, not directly shown for
  the human protein in the cited evidence.

Assessment. A real gap (only the class term), but the fill is less certain
than the others. There is no thyroxine sulfotransferase activity term in GO; the
closest specific existing term is GO:0004062 aryl sulfotransferase activity
(SULT6B1 is an aryl/phenolic sulfotransferase). Given the by-similarity evidence,
the responsible move is a cautious refinement plus an explicit question, not a
confident substrate-specific assertion.

Action.
- MODIFY toward GO:0004062 aryl sulfotransferase activity as a more
informative-but-defensible parent, or leave as UNDECIDED pending direct
human-enzyme substrate data — flag in suggested_questions.
- Only propose a thyroxine-specific term once direct activity (not similarity)
is established for the human protein.

A deliberate contrast case: not every RHEA gap is a confident fill — evidence
quality gates how specific the proposed GO term should be.

Cross-case lessons

1. The gap is usually about specificity, not coverage. None of these enzymes
   is left with no MF term — each has a class/root term (lyase activity,
   sulfotransferase activity, generic GalNAc-T, hydrolase activity). The RHEA
   reaction is the only place the specific activity is recorded, and the
   unmapped reaction is why it never reaches GO. This matches the project-level
   finding that G4/G5 are specificity gaps, not coverage gaps.
2. Two fixes, picked by whether the GO term exists. SAMD8 needs only an
   existing term applied (propagation gap). PHYKPL and B3GALNT2 need a new
   term (granularity gap) — these are proposed_new_terms for GO, and good
   candidates for the RHEA team to add to rhea2go once the term exists.
3. Evidence gates specificity. SULT6B1 shows that an unmapped RHEA on a
   by-similarity annotation justifies only a cautious parent-level refinement.
4. These would all be fixed at the source by mapping the reactions in
   rhea2go (where a GO term exists) and by adding the three missing GO MF terms
   (where one does not) — exactly the two levers this project recommends.

Curated mappings

Each case is recorded as a row in the curated SSSOM mapping set
rhea2go.sssom.yaml: SAMD8 as a skos:exactMatch
(RHEA:36079→GO:0002950, ready to add to rhea2go); PHYKPL, B3GALNT2 and
SULT6B1 as skos:broadMatch rows whose comments name the narrower GO term to
request. Validate with just validate-rhea-mappings.

Reproduce

uv run python rhea_gap_finder.py --organism 9606   # regenerate candidate list

To promote any case to a full gene review, run the standard workflow
(just fetch-gene human PHYKPL, edit PHYKPL-ai-review.yaml, just validate …)
and link it back here.