Archaeal GO annotation error patterns

Back to the ARCHAEA project

This catalogue records recurring ways that GO annotation goes wrong for
archaeal gene products. Each entry names the pipeline involved, gives an
example from a completed review, and says what a fix might look like. It
was started from the five-gene pilot (October 2026): Msed_0709, sor, lacS,
mcrA (Methanothermobacter) and cdvB.

The five pilot genes had 20 GOA rows between them: 17 electronic (IEA or
IBA), one ISS, and two EXP (both on Methanothermobacter mcrA).

1. A neofunctionalized paralog inherits its ancestor's annotation

• Pipelines: PANTHER TreeGrafter (GO_REF:0000118), InterPro2GO
  (GO_REF:0000002), UniProt ARBA naming.
• Example: METS5/Msed_0709 (A4YEN2), the malonyl-CoA/succinyl-CoA
  reductase of the 3HP/4HB CO2-fixation cycle. It evolved from
  aspartate-semialdehyde dehydrogenase (ASD) by gene duplication. It
  carried ASD activity (GO:0004073), threonine biosynthesis (GO:0009088)
  and amino acid biosynthetic process (GO:0008652); all three were
  removed. The genuine M. sedula ASD, Msed_1700 (A4YHF3), shares the
  PANTHER subfamily (PTHR46718:SF1) and the InterPro signatures, so both
  genes receive the same GO set.
• The error also runs the other way: ARBA rules ARBA00067001 (name)
  and ARBA00052213 (EC 1.2.1.75 catalytic activity) name the true ASD,
  Msed_1700, as a malonyl-CoA reductase. Both rules have only two
  conditions: FunFam 3.40.50.720:FF:000411 (itself named
  "Aspartate-semialdehyde dehydrogenase", covering both paralogs) and the
  taxon Archaea. Of the 40 archaeal UniProt entries matching the FunFam, 38
  carry the rule's EC 1.2.1.75 naming; of those 38, 17 match MCR-type CDD
  models, 17 match only ASD-type models (probably misnamed true ASDs,
  including Msed_1700 and Sulfurisphaera tokodaii ST1242) and 4 match
  neither. No GO row derives from this naming yet.
  The Msed_1700 review (genes/METS5/Msed_1700/) accepts its ASD terms.
• Possible fix: split the malonyl-CoA reductase clade into its own
  PANTHER subfamily, and add the CDD models as conditions in the ARBA
  rules: cd23940/cd24150 for MCR, cd18130/cd02315 for ASD. NCBIfam
  TIGR00978 (asd_EA) cannot be used, because both paralogs carry it.
• Structured record: rule reviews
  rules/arba/ARBA00067001/ARBA00067001-review.yaml (name) and
  rules/arba/ARBA00052213/ARBA00052213-review.yaml (catalytic activity),
  both action MODIFY, with the per-protein classification in
  rules/arba/ARBA00067001/RESULTS.md: of 38 rule-annotated entries, 17
  are MCR-type, 17 ASD-type and 4 unclassified. No GO row derives from the
  rules yet, but a new malonyl-CoA reductase GO term would carry the error
  into GO through Rhea-to-GO mapping, so the rules should be fixed first.
• Where else to look: other enzymes of the 3HP/4HB cycle and other
  archaeal pathways that recruited enzymes from amino acid metabolism.

2. A plant-seeded PAINT node is propagated to an archaeal enzyme

• Pipeline: PANTHER/PAINT IBA.
• Example: SACS2/lacS (P22498), the GH1 beta-glycosidase. It carried
  galactosyltransferase activity (GO:0008378) by IBA from node
  PTN008498319, whose only experimental seed is Arabidopsis SFR2, a
  galactolipid transferase that has lost hydrolase activity. Of the 58
  GO:0008378 IBA rows citing that node, lacS is the only non-plant one. The
  row was removed.
• Latent risk: the same node has a negated IRD for beta-glucosidase
  activity (GO:0008422), an activity that has been measured directly in
  lacS.
• Possible fix: restrict the SFR2 IBD to the plant clade, or move the
  archaeal GH1 enzymes out of the SFR2-named subfamily (SF209).

3. Eukaryotic compartment terms on archaeal ESCRT-III homologs

• Pipelines: InterPro2GO (GO_REF:0000002), TreeGrafter
  (GO_REF:0000118).
• Example: cdvB itself is not affected, because its InterPro family is
  IPR053654 (CdvB). But InterPro2GO maps IPR005024 (Snf7) to vacuolar
  transport (GO:0007034) on about 41 archaeal ESCRT-III/CdvB homologs,
  including Sulfolobales proteins such as Q972B4 (ST1214) and A4YHE8
  (Msed_1695). TreeGrafter puts multivesicular-body and
  endosome-to-vacuole transport terms (GO:0032511, GO:0043328) on Asgard
  archaeal ESCRT proteins.
• Why it is wrong: archaea have no vacuole or endosome. The archaeal
  process is Cdv-dependent cytokinesis (GO:0061639).
• Possible fix: add a Eukaryota taxon restriction to the IPR005024 to
  GO:0007034 mapping.

4. A subcellular-location description is mapped to the wrong compartment

• Pipeline: UniProt SubCell mapping (GO_REF:0000044).
• Example: SULAC/cdvB (Q4J924) carried nucleoid (GO:0009295). The
  UniProt note says the protein forms bands between segregating
  nucleoids at mid-cell. The controlled value "Cytoplasm, nucleoid" and
  the mapping turned this into a nucleoid location. Modified to cell
  division site (GO:0032153).
• Where else to look: CdvA, CdvC and other archaeal division-site
  proteins described relative to nucleoids.

5. Under-annotation: experimental evidence never reaches GO

• Example: ACIAM/sor (P29082), sulfur oxygenase/reductase. Only five
  proteins in GOA carry GO:0033755, all by IEA. No SOR has an experimental
  GO annotation or any process term, despite biochemistry since 1989 and
  more than ten crystal structures. UniProt records experimentally
  supported iron-binding residues (86, 90, 114), but GOA has no metal- or
  iron-binding row.
• Example: cdvB had no process term at all. GO:0061639 Cdv-dependent
  cytokinesis, the term created for this machinery, has no annotations
  for any organism in GOA.
• Action: the pilot reviews propose NEW annotations for these with
  IDA evidence.

6. Generic or misleading terms where the archaeal biology is known

• Generic methanogenesis. InterPro2GO assigns methanogenesis
  (GO:0015948) to MCR family members even where the organism's only route
  is known. Methanothermobacter strain Delta H grows only on H2 and CO2,
  so its mcrA was modified to methanogenesis, from carbon dioxide
  (GO:0019386).
• Missing GO terms for archaeal pathways. GO has no activity terms for
  malonyl-CoA reductase (NADP+, EC 1.2.1.75) or succinyl-CoA reductase
  (NADP+, EC 1.2.1.76), and no term for the 3HP/4HB cycle. Carbon fixation
  by 3-hydroxypropionate cycle (GO:0043427) describes the Chloroflexus
  bicycle, a different pathway. The Msed_0709 review proposes three new
  terms.
• Look-alike term. Aerobic sulfur oxidation (GO:0070220) is defined
  through the sulfur dioxygenase reaction (EC 1.13.11.18), not the
  sulfur oxygenase/reductase reaction, so the sor review avoided it.
• Default NAD binding. InterPro2GO maps the Rossmann-fold domain
  IPR000534 to NAD binding even for NADPH-specific enzymes such as
  Msed_0709. The row was removed.

7. Evidence attributed to one isoenzyme from a mixed assay

• Example: one of the two EXP rows for coenzyme-B
  sulfoethylthiotransferase activity on Methanothermobacter mcrA cites a
  cell-extract assay (PMID:3122735). Extracts contain both MCR isoenzymes
  (MCR I and MCR II), so the activity cannot be attributed to MCR I alone.
  The row was accepted, because the activity itself is not in doubt, and
  the caveat is recorded in the review.

8. InterPro2GO puts catalytic activity on non-catalytic proteasome alpha subunits

• Pipeline: InterPro2GO (GO_REF:0000002), IPR019982 (archaeal
  proteasome alpha subunit) to threonine-type endopeptidase activity
  (GO:0004298) with enables.
• Example: HALVD/psmA1 (Q9V2V6). The alpha subunits form the gated
  outer ring of the 20S core particle; the catalytic nucleophile is the
  N-terminal threonine of the beta subunit. The InterPro row and three
  endopeptidase rows (2 IDA, 1 ARBA IEA) were modified to structural
  constituent of proteasome (GO:0140756), the same choice as the repo's
  human PSMA1 review, with contributes_to GO:0004298 in core_functions.
• Scale: about 786 archaeal GO:0004298 rows in QuickGO come from this
  mapping.
• Possible fix: drop the mapping or change it to contributes_to.

9. Ubiquitin-dependent terms on organisms without ubiquitin

• Pipeline: InterPro2GO, IPR000426 (proteasome alpha subunit N-terminal
  domain) to ubiquitin-dependent protein catabolic process (GO:0006511).
• Example: HALVD/psmA1. Archaea have no ubiquitin; Haloferax uses
  SAMP conjugation, and its link to proteasomal degradation is indirect.
  The row was removed.
• Scale: about 888 of the 910 archaeal GO:0006511 rows in QuickGO come
  from this mapping.
• Possible fix: add a Eukaryota taxon restriction to the mapping.
• Implemented as CR-0003 (2026-10-31). Evaluated on all 610 archaeal
  proteomes with an IPR000426 member: the ubiquitin relay is absent from 501
  complete proteomes, so the rule removes 733 of the 888 archaeal rows (83%).
  It keeps the term in two Asgard archaea and Candidatus Methanoplasma
  termitum, which encode a UBC E2, a ubiquitin-like protein and a RING
  protein, and in 106 proteomes too incomplete to call the relay absent
  (rules/curated/context/RESULTS.md).

10. Microbial rhodopsin pumps annotated as channels

• Pipeline: InterPro2GO, IPR018229 (rhodopsin retinal-binding site) to
  monoatomic ion channel activity (GO:0005216); proteorhodopsin also gets
  light-activated channel activity (GO:0010461) through IPR017402.
• Example: HALSA/bop (P02945), bacteriorhodopsin, a light-driven proton
  pump. The channel row was removed. The same row is on every microbial
  rhodopsin pump checked: halorhodopsins P0DMH7, P15647 and P33742,
  sensory rhodopsin II P42196 and proteorhodopsin Q9F7P4.
• Why it is wrong: among microbial rhodopsins only the algal
  channelrhodopsins are channels.
• Possible fix: remove the channel mapping from IPR018229 (it is a
  retinal-binding-site signature shared by pumps, sensors and channels).

11. The Chloroflexus 3-hydroxypropionate cycle term on the archaeal 3HP/4HB cycle

• Pipelines: UniProt ARBA (GO_REF:0000117), plus one IDA row.
• Problem: GO:0043427 carbon fixation by 3-hydroxypropionate cycle is
  defined as the pathway "by which two molecules of carbon dioxide are fixed
  to form glyoxylate", with malyl-CoA cleaved to acetyl-CoA and glyoxylate.
  That is the Chloroflexus bicycle. The crenarchaeal 3HP/4HB cycle regenerates
  acetyl-CoA through 4-hydroxybutyrate and has no malyl-CoA lyase.
• Scale: 729 of the 766 GO:0043427 annotations in QuickGO (2026-10-31) are
  on archaeal proteins, 727 of them from ARBA.
• Examples: METS5/Msed_1456 (IDA and ARBA rows modified); Msed_0709,
  Msed_2001, Msed_1424 and Msed_0406 use GO:0015977 carbon fixation instead.
• Possible fix: create the proposed term "carbon fixation by
  3-hydroxypropionate/4-hydroxybutyrate cycle" and repoint the ARBA rules to it.

12. A Swiss-Prot name and EC number for the wrong step of a pathway

• Example: METS5/Msed_1424 (A4YGN0) is named succinate-semialdehyde
  dehydrogenase (acetylating), EC 1.2.1.76, with RHEA:26450. The purified
  native enzyme is the NADPH succinic semialdehyde reductase, the next step
  (PMID:19684143), and EC 1.2.1.76 belongs to Msed_0709 from the same paper.
  The protein is a zinc alcohol dehydrogenase (IPR002328), not an aldehyde
  dehydrogenase.
• Risk: no GO row derives from it yet, but a succinyl-CoA reductase GO
  term mapped from RHEA:26450 would carry the error into GO.

13. EC-to-GO chain-length mismatch

• Example: METS5/Msed_0406. EC 6.2.1.2 (butanoate--CoA ligase) maps to
  GO:0031956 medium-chain fatty acid-CoA ligase activity, which GO defines for
  C6 to C12; the enzyme acts on C2 to C5 acids. Modified to GO:0031955
  short-chain fatty acid-CoA ligase activity, including the IDA row.

14. TreeGrafter and SubCell terms from the wrong reaction or a weak prediction

• Synthase for synthetase: PANTHER node PTN008683229 (PTHR43605) gives
  Msed_0406, an AMP-forming acyl-CoA ligase, GO:0004321 fatty-acyl-CoA
  synthase activity (the fatty acid synthase reaction) and fatty acid
  biosynthesis, in an archaeon whose membrane lipids are isoprenoid ethers.
• Membrane from one predicted helix: Msed_0406 has a UniProt
  transmembrane prediction (ECO:0000255) inside its N-terminal domain, but was
  purified from the soluble fraction and has crystal structures. This is a
  sub-pattern of pattern 4.
• Family-founder process terms: Msed_2001 inherits fatty acid
  beta-oxidation from the ECHS1-named crotonase subfamily, and Msed_1456
  acetyl-CoA biosynthesis from the acetyl-CoA synthetase family (pattern 1).

15. An experimental row with the wrong cofactor variant of the activity

• Example: METS5/Msed_1426 (A4YGN2), acryloyl-CoA reductase. A UniProt IDA
  row assigns GO:0043958 acryloyl-CoA reductase (NADH) activity, but the cited
  paper (PMID:19429610) reports that the enzyme "did not act on either NADH or
  crotonyl-CoA". Corrected to GO:0043957, the NADPH term, which an EXP row
  already carried. Worth reporting to UniProt.

16. Whole-complex activity on a single subunit

• Example: METS5/Msed_1375, the carboxyltransferase subunit of the
  acetyl-CoA/propionyl-CoA carboxylase, received propionyl-CoA carboxylase
  activity from TreeGrafter and ligase activity from InterPro2GO; the ATP-using
  ligation belongs to the biotin carboxylase subunit. The PANTHER node also
  misses the enzyme's equal activity on acetyl-CoA. A wider form of pattern 8.
• Example: METS5/Msed_2055, the separately encoded B12-binding subunit of
  methylmalonyl-CoA mutase. InterPro2GO maps the MCM B12-binding domain
  (IPR006159) to isomerase activity with enables; in split-subunit mutases the
  protein matching it binds the cofactor but does not catalyse alone (active
  enzyme needs both subunits, PMID:22752162).

Related findings from round 2

• Under-annotation reaches the best-studied archaeal proteins too.
  bop is the only gene product in GOA annotated to light-driven active
  monoatomic ion transmembrane transporter activity (GO:0015454), with no
  ion named; no archaeal protein in GOA has any retinal-binding term. Two
  of its three experimental rows come from one 1973 abstract. psmA1's six
  IDA rows come from two papers by one lab. See pattern 5.
• ESCRT-III vacuolar transport (pattern 3) hits paralogs arbitrarily.
  In M. sedula, only Msed_1695 (CdvB2) carries the row, because only it
  also matches Pfam PF03357 / IPR005024; its close paralog Msed_2179
  (CdvB1) escapes. The 41 InterPro2GO rows span about 36 archaeal taxa.
• Unsupported process terms on ammonia monooxygenase. The merged NITRP
  amoA review dropped an anammox term (anaerobic respiration, using
  ammonium as electron donor, GO:0019331) from an aerobic ammonia
  oxidizer, and found that GO:0019409 (aerobic respiration, using ammonia
  as electron donor) is defined through hydrazine intermediates, which
  does not fit aerobic nitrification.

Structured records

The upstream-fixable patterns are recorded in machine-readable form, ready to
send to the resource that owns them:

Pattern	Record	Verdict	Replacement rule
1 (ARBA naming)	rules/arba/ARBA00067001/ARBA00067001-review.yaml, rules/arba/ARBA00052213/ARBA00052213-review.yaml	MODIFY: require MCR-type CDD models	CR-0004
3 (ESCRT-III vacuolar transport)	projects/INTERPRO/interpro2go.sssom.yaml, IPR005024 to GO:0007034	broadMatch	CR-0001 (Eukaryota), CR-0002 (archaeal Cdv cytokinesis)
6 (default NAD binding)	same file, IPR000534 to GO:0051287	broadMatch, narrow	none yet
8 (catalytic activity on proteasome alpha)	same file, IPR019982 to GO:0004298	REMOVE	not needed
9 (ubiquitin terms in archaea)	same file, IPR000426 to GO:0006511	broadMatch	CR-0003 (module context, evaluated per proteome)
10 (rhodopsin pumps as channels)	same file, IPR018229 to GO:0005216 and IPR017402 to GO:0010461	REMOVE	not needed

The replacement rules live in rules/curated.
A taxon or module condition is used only where the lineage or the missing
component is what separates right from wrong members; where the term is wrong
for every member, the record is a removal and no rule is needed.

Patterns 2 (plant-seeded PAINT node on lacS) and 4 (SubCell nucleoid
mapping) are not yet in a structured record: the first belongs with PANTHER
IBA reviews (see IBA_REVIEW), the second with UniProt's
SubCell-to-GO mapping, which has no review file in this repository yet.