UniPathway unique terms

Auditing GO_REF:0000041, a legacy pathway-vocabulary mapping, where it is the only source

AI Gene Review · projects/UNIPATHWAY · 2026

Bottom line

  • Closure filtering removes rows already supported at the same or a more specific term: human UniPathway rows drop from 1,129 to 247 truly unique.
  • 32 exemplar rows across 9 organisms reviewed: 24 ACCEPT. UniPathway is a net positive pathway gap filler, strongest in microbes.
  • Errors are specific, not systemic: UBA7 (ISG15, not ubiquitin) MODIFY; nrfA (not nitrate assimilation) REMOVE; NorR regulators over-annotated to denitrification.

Why audit UniPathway

  • GO_REF:0000041 maps UniPathway pathways to GO BP terms. UniPathway is archived and unmaintained, yet its rows still sit in GOA.
  • Question: is it a source to trust, clean, or retire?
  • Same approach as the SPKW project: a row only counts as uniquely informative if no other source supports the gene at that term or a descendant.
  • Review by biologically coherent term groups (ubiquitination, nitrogen cycle, plant cell wall), not random sampling.

Most UniPathway rows are already covered

The hard case: protein ubiquitination

Nitrogen cycle: enzymes vs regulators vs wrong endpoint

Gene (organism) UniPathway term Action
nirK1, nirK2, nosZ (R. palustris) denitrification pathway ACCEPT
Ferp_0128 (Ferroglobus) denitrification pathway ACCEPT
ureC1, ureC2 (N. viennensis) urea catabolic process ACCEPT
norR1, norR2 (Cupriavidus) denitrification pathway MARK_AS_OVER_ANNOTATED
nrfA (Desulfotalea) nitrate assimilation REMOVE
AJ80_06654 (fungus, singleton) denitrification pathway UNDECIDED

nrfA is an ammonia-forming cytochrome c nitrite reductase (GO:0042279), a dissimilatory enzyme. NorR is an NO-responsive sigma-54 activator, not a pathway enzyme.

Outcome across all exemplars

Patterns

  1. Metabolic enzyme in its pathway → ACCEPT (COX5B, catA, algE, Brachypodium PAL and UXS).
  2. Modification buckets need the mechanism: E3s and substrate adaptors yes; UBL enzymes and DUBs no.
  3. Broad lipid parents are correct but not core (GK5, PM20D1 non-core; LPCAT1 over-annotated).
  4. Regulators are not pathway enzymes (norR1, norR2).
  5. Microbial signal dwarfs vertebrate: 165,344 TRUE-unique bacterial rows vs 247 in human.

Status and next steps

  • Scans done: 13 single-species databases + 6 clade aggregates; 32 exemplar reviews in genes/.
  • Not yet done: full human UPA00143 audit (124 genes), retinol/cholesterol CYP subset, bacterial denitrification set (615 rows, 530 taxa), nrfA nitrate-assimilation tail (14 rows).
  • Recommendation: keep closure filtering as default; review microbial rows by term group.

Read more: projects/UNIPATHWAY.md · genes/human/UBA7/ · genes/DESPS/nrfA/