UniPathway Unique Terms Project

Overview

This project reviews GO annotations derived from UniPathway vocabulary mapping
(GO_REF:0000041) where UniPathway provides the only non-redundant support for a
gene-term pair. It follows the same principle used for the SPKW project: a source
is only considered uniquely informative if the annotation is not already supported
by another source for the exact term, or by another source for a more specific
descendant term.

UniPathway-derived annotations are mostly biological-process pathway assertions.
Unlike SPKW, the strongest signal is not broad eukaryotic process conflation; it is
a mix of useful pathway gap-filling, broad parent-process assertions, and a small
number of source-pathway mapping errors where a pathway bucket is too coarse for a
specific gene product.

GO_REF:0000041 describes UniPathway as an archived, no-longer-maintained pathway
vocabulary. That makes this a legacy-source audit: useful annotations should be
retained, but broad or stale mappings should not be treated as automatically
current.

Key Findings

• Closure filtering matters: in human GOA, 1,129 UniPathway annotations reduce
  to 247 true unique annotations after exact and descendant-term support from other
  sources is removed.
• Most unique annotations are plausible metabolic or protein-modification pathway
  context, not obvious over-annotation.
• The largest human unique term is protein ubiquitination (GO:0016567),
  driven by UniPathway:UPA00143. This bucket contains clear positives such as
  E3 ligases and CRL adaptors, but also at least one incorrect mapping for ISG15
  conjugation. However, decisions as to pathway boundaries are unclear: UniPathway
  treats E1 activation, E2 transfer, and E3-mediated substrate recognition/transfer
  as children of the same pathway bucket, while GO curation still has to decide
  whether each gene product is directly participating in ubiquitination or only
  regulating, binding, or responding to ubiquitinated proteins.
• Metabolic pathway terms are usually accurate when the gene catalyzes a step in
  the pathway, especially in bacteria and yeast.
• Some UniPathway terms are correct but too broad for the final curated
  representation, for example parent lipid-metabolism terms where the gene is
  better captured by a more specific enzymatic product or remodeling pathway.
• The microbial signal is much larger than the vertebrate signal. Aggregate
  bacterial, fungal, archaeal, and viral scans show that UniPathway is often a
  unique source for pathway-level microbial metabolism.
• Targeted term-group review is necessary. Nitrogen-cycle terms show a useful
  positive signal for denitrification and urease-linked urea catabolism, but also
  probable mapping-risk cases such as nrfA genes annotated to nitrate
  assimilation.

Methods

This project used local GOA DuckDB databases under ~/repos/go-db/db/.
UniPathway annotations were counted from non-negated GO_REF:0000041 rows.

A row was treated as TRUE unique only when no non-UniPathway source already
supported the same gene-term pair, and no non-UniPathway source supported a more
specific descendant term for the same gene. This prevents parent pathway terms
from looking novel when another source already supports a more specific process.

The nitrogen-cycle audit used the GO:0071941 nitrogen cycle metabolic process
closure. Exemplar rows were then reviewed through the gene-review workflow:
fetch gene data, edit the review YAML, validate, render the gene page, and link
the reviewed result from this project page.

For the pilot rows below, the YAML reviews now include explicit per-annotation
reasons and supported_by evidence. Where direct species-level experiments were
not available, the review states the inference basis: enzyme name and EC,
UniProt pathway/reaction text, InterPro/NCBIfam/PANTHER family placement, locus
context, or broader family literature.

Deep Research Highlights

The deep-research pass was most useful for checking whether broad family-level
summaries matched the specific propagated node, subfamily, or independent
enzyme evidence for each gene:

• NorR-family research supports norR1 and norR2 as NO-responsive sigma-54
  transcription activators, not denitrification enzymes. This supports keeping
  regulatory activity terms while marking the UniPathway denitrification process
  rows as over-annotated.
• The NosZ/Cox2 CuA family report explains why nosZ and Ferp_0128 can carry
  misleading cytochrome-c oxidase annotations. The family has shared CuA-related
  ancestry, but the reviewed proteins sit in nitrous-oxide reductase context, so
  the reviews remove cytochrome-c oxidase activity and keep N2O reduction.
• The cytochrome c-552/NrfA family report showed why a broad family display
  label should not be treated as the propagation unit. PANTHER is useful when
  the tree, subfamily, and IBA propagation identify the relevant clade; for nrfA,
  the decisive evidence is the gene-specific EC/HAMAP/InterPro nitrite-reductase
  assignment, while the UniPathway nitrate-assimilation row is removed because it
  describes the wrong biological-process endpoint.
• Multicopper-oxidase family research supports copper nitrite reductase calls
  only when the nitrite-reductase subfamily evidence is present. That reinforces
  nirK positives while leaving the singleton fungal AJ80_06654
  denitrification-pathway row undecided.
• Urease-family research supports ureC1 and ureC2 as direct urea-catabolism
  positives, but adds a useful caveat: most ureases are nickel enzymes, yet rare
  iron ureases mean metal-binding annotations should be tied to gene-specific or
  subfamily evidence.
• For BRADI_1g66227v3, Falcon found no direct paper on the Brachypodium gene.
  Plant UXS literature strongly supports UDP-glucuronate decarboxylase activity
  and UDP-D-xylose biosynthesis, but Golgi localization remains a UniProt
  prediction rather than a gene-specific experiment.

Cross-Organism Scan

Broader Species and Clade Patterns

Targeted Term Group: Nitrogen Cycle

The nitrogen-cycle audit used the GO:0071941 nitrogen cycle metabolic process
closure. In the scanned databases, UniPathway contributes rows for
denitrification pathway, nitrate assimilation, urea catabolic process, and
urea cycle. It did not contribute rows for nitrogen fixation or nitrogenase
activity terms in the scanned datasets, so this source is not currently filling
nitrogenase annotation gaps.

Nitrogen-Cycle Pilot Assessments

Plant Pathway Pilot Assessments

Top Human TRUE UniPathway-Unique Terms

Protein Ubiquitination Boundary Review

UniPathway itself helps explain why this term is hard. UPA00143 is not just a
generic label; it has three child pathways for the ubiquitination cascade:

That source boundary supports including E1, E2, and E3 machinery in
GO:0016567, and it also explains why substrate-recognition proteins can be
legitimate positives. It does not settle every GO curation boundary, because GO
annotations are made to individual gene products rather than to the pathway as a
whole.

The key boundary problem is therefore how broadly GO:0016567 should be read
when projected from the UniPathway parent:

Under this boundary, GO:0016567 remains useful as a biological-process
grouping even when we add molecular-function adaptor terms. The BP term says
the gene product is directly involved in the ubiquitination process; the MF term
says what it does in that process. For non-catalytic substrate receptors, the
preferred core MF is GO:1990756 ubiquitin-like ligase-substrate adaptor
activity, not a generic binding term and not an E3 catalytic activity term.

This means protein ubiquitination is not redundant with existing regulation
annotations such as positive regulation of proteasomal ubiquitin-dependent
protein catabolic process. The regulation terms describe the outcome or pathway
effect. GO:1990756 describes the substrate-adaptor activity. GO:0016567
describes direct participation in the ubiquitin-conjugation process.

Human GO:0016567 Triage

The 124 human TRUE UniPathway-unique GO:0016567 rows divide into the following
curation buckets:

Re-review Outcomes

Reviewed Exemplars

Pattern Analysis

1. Metabolic Enzyme in Pathway - Usually ACCEPT

Examples: COX5B in oxidative phosphorylation, catA in beta-ketoadipate pathway,
algE in alginate biosynthesis.

These are the strongest UniPathway successes. The source supplies a pathway-level
biological-process annotation that may be missing from InterPro, PAINT, or direct
manual curation, while staying close to the enzyme or complex role.

2. Protein Modification Buckets - Inspect Mechanistic Role

Examples: SOCS4, SOCS5, KCTD11, ZSWIM8, ELOB, ELOC, UBA7.

UPA00143 is valuable for E3 ligases, direct E3-complex modules, and
substrate-adaptor proteins, but it needs the boundary rules above. The key
curation split is catalytic ubiquitin-transfer activity versus substrate-adaptor
activity versus non-ubiquitin UBL specificity.

3. Broad Lipid Parent Terms - Often KEEP_AS_NON_CORE or MARK_AS_OVER_ANNOTATED

Examples: PM20D1, LPCAT1.

UniPathway often captures the broad lipid class correctly, but the broad term may
not be the best curated endpoint when specific lipid remodeling, acyltransferase,
or N-acyl-amino-acid chemistry is available.

4. Bacterial Pathway Coverage - Strong Positive Signal

The PSEPK pilot resembles the positive-control side of the SPKW bacterial work.
Top TRUE unique terms include beta-ketoadipate pathway, oxidative phosphorylation,
thiamine diphosphate biosynthesis, tetrahydrofolate biosynthesis, histidine
catabolism, and LPS/heptose biosynthesis. These generally map directly to operon
or enzyme-pathway membership and look curatorially useful.

5. Term-Group Audits - Better Than Source-Wide Sampling

The nitrogen-cycle scan shows why UniPathway should be reviewed by biologically
coherent term groups, not only by source-wide top terms. Denitrification is a
high-confidence microbial target, while nitrate assimilation is mostly redundant
and its remaining true-unique cases include likely non-assimilatory nrfA rows.
Urea-cycle terms are high volume but need careful scope interpretation in microbes.

6. Pathway Regulators - Do Not Treat as Pathway Enzymes

Examples: norR1, norR2.

UniPathway can correctly record that a regulator is associated with a pathway, but
the GO biological-process annotation may become too direct if the row is projected
as involved_in denitrification pathway. NorR1 and NorR2 regulate nitric oxide
reductase expression; they should not be curated as denitrification-pathway
enzymes.

Curation Recommendations

1. Keep closure filtering as the default. Naive UniPathway-only queries overstate
   novelty by counting rows already supported by exact or more specific terms from
   other sources.
2. Prioritize UPA00143 review. It is the dominant human unique source and has
   both good positives and clear UBL-specific mapping errors.
3. Treat bacterial metabolic pathway rows as high-value candidates. They often
   add real pathway context absent from other GOA sources.
4. Downgrade broad parent lipid terms when specific processes exist. Use
   KEEP_AS_NON_CORE, MARK_AS_OVER_ANNOTATED, or MODIFY depending on whether
   the broad term is merely contextual or actually misleading.
5. Check modifier identity before accepting protein-modification terms. Do not
   assume that a UniPathway protein-modification bucket is specific enough for GO.
6. Use targeted term-group audits for microbial processes. For nitrogen-cycle
   terms, denitrification and urease-linked urea catabolism are higher-value review
   targets than nitrate assimilation, which is mostly redundant or potentially
   mis-mapped in the true-unique tail.
7. Separate pathway enzymes from pathway regulators. For UniPathway rows on
   transcriptional regulators, inspect whether the source is asserting direct
   pathway membership or only regulatory context.

Follow-Up Targets

Project Status

• Started: 2026-05-21
• Pilot databases reviewed: human, mouse, rat, Xenopus, zebrafish, yeast,
  fission yeast, worm, ANOGA, Brachypodium, wheat, PSEPK, aggregate bacteria,
  Pseudomonadota, Clostridium, fungi, archaea, and virus
• Targeted term group reviewed: nitrogen-cycle closure under GO:0071941
• Full exemplar reviews referenced: SOCS4, SOCS5, KCTD11, ZSWIM8, ELOB,
  ELOC, BRCA1, RAD18, SYVN1, PEX2, PEX10, PEX12, UBA7, COX5B, GK5,
  PM20D1, LPCAT1, algE, catA, nirK1, nirK2, nosZ, ureC1, ureC2, nrfA,
  AJ80_06654, norR1, norR2, Ferp_0128, BRADI_1g22147v3, LOC100829928,
  BRADI_1g66227v3
• Current conclusion: UniPathway is a net positive source for unique pathway
  annotations, with a lower apparent over-annotation rate than problematic SPKW
  eukaryotic BP mappings. The main risk is broad pathway-bucket propagation, not
  wholesale process conflation.

Technical Appendix

The scan uses a closure-filtered uniqueness query. The important curation point is
that exact duplicate support and more-specific descendant support are both removed
before a UniPathway row is called uniquely informative.

WITH upa AS (
    SELECT internal_id, db_object_id, db_object_symbol, ontology_class_ref, aspect
    FROM gaf_association
    WHERE supporting_references = 'GO_REF:0000041'
      AND is_negation = false
),
true_unique AS (
    SELECT u.*
    FROM upa u
    WHERE NOT EXISTS (
        SELECT 1
        FROM gaf_association o
        WHERE o.db_object_id = u.db_object_id
          AND o.ontology_class_ref = u.ontology_class_ref
          AND o.supporting_references != 'GO_REF:0000041'
          AND o.is_negation = false
    )
    AND NOT EXISTS (
        SELECT 1
        FROM isa_partof_closure ipc
        JOIN gaf_association o
          ON o.db_object_id = u.db_object_id
         AND o.ontology_class_ref = ipc.subject
         AND o.supporting_references != 'GO_REF:0000041'
         AND o.is_negation = false
        WHERE ipc.object = u.ontology_class_ref
    )
)
SELECT count(*) AS annotations,
       count(DISTINCT db_object_id) AS genes,
       count(DISTINCT ontology_class_ref) AS terms
FROM true_unique;