Dark-Gene Batch Reviews: Curation Highlights

Three batched review campaigns targeted deliberately understudied genes — the population
where function knowledge gaps concentrate — rather than the well-annotated core:

Campaign	Organism	Genes	Selection basis
CAEEL	C. elegans (worm)	50	Unreviewed members of the flagship CAEEL pathway projects (cilia/IFT, mitophagy, proteostasis, UPR, surveillance immunity, P-granule)
YEAST	S. cerevisiae (yeast)	50	Named-but-dark genes from the SGD GAF: standard gene name, zero experimental-evidence annotations, few informative GO terms
POMBE	S. pombe (SCHPO)	20	Named-but-dark genes from the PomBase GAF, same darkness filter

Each gene received a full annotation review plus a literature-grounded knowledge_gaps
section (the point of the exercise — see the parent project).
Beyond the gap records themselves, the reviews surfaced recurring, generalizable curation
findings. Those patterns are collected here because they are the concrete argument for why
dark genes need human-grade review rather than electronic propagation.

1. Identity correction before curation

Dark genes are disproportionately mislabeled — a "standard name" or a family hint can point at
the wrong biology. Grounding every review in the actual UniProt/PomBase record (not the symbol
or a prior assumption) caught several outright misassignments before they could drive wrong
annotations:

• sel0 (pombe, SPAC20G4.05c) — not a Sel1-repeat / SEL1L ERAD protein as the name suggests,
  but Selenoprotein O, a mitochondrial protein AMPylase / adenylyltransferase (SELO family,
  pseudokinase fold). Entire review reframed around AMPylation.
• mtl3 (pombe, SPBC215.13) — not the Mtr4-like MTREC RNA helicase (that is mtl1), but a
  Mid2-like GPI-anchored plasma-membrane cell-wall stress sensor with no catalytic domain.
• spa1 (pombe, SPBC577.14c) — not the mammalian Rap-GAP SIPA1/SPA1 despite the shared
  symbol, but ornithine decarboxylase antizyme (an ODC inhibitor).
• knh4 (pombe, SPBC1E8.05) — not a Knr4/Smi1 protein, but a Kre9/Knh1-family β-glucan
  cell-wall glycoprotein.
• MNN14 (yeast, YJR061W) — not a GT15/MNN1 α-1,3-mannosyltransferase, but a LicD-family
  mannosyl-phosphate transferase (activity later added from primary literature; see §4).
• ppgn-1 (worm) — the flagship-project framing implied an immunity effector; the record
  shows the paraplegin / SPG7 m-AAA protease (a reproducible motif scan confirmed intact
  catalytic residues).
• MCO14 (yeast) — fetch-gene failed on the symbol; resolved via SGD to YHL018W, a
  PCD/DCoH-family protein. (Yeast/pombe standard names frequently are not in UniProt's gene
  line — the pipeline now falls back to the systematic name + --uniprot-id.)

2. Pseudoenzyme detection

A conserved catalytic fold is routinely over-annotated with the ancestral catalytic activity
even when the catalytic residues have degenerated. Inline domain analysis (reading the UniProt
features and checking the specific catalytic motif) flagged several:

• KDX1 (yeast) — protein-kinase fold but activation-loop DFG→NFG and catalytic-loop
  HRD→HCD; a catalytically inactive pseudokinase. Nine propagated catalytic-kinase MF/BP
  terms marked as over-annotations; its real role is a non-catalytic Swi4/Rlm1 scaffold.
• OCA6 (yeast) — PTP/DSP fold but the invariant CX5R catalytic arginine is replaced (→Ile);
  a likely pseudophosphatase; the "protein tyrosine phosphatase activity" IEA flagged.
• GPM3 (yeast) — retains the phosphoglycerate-mutase histidines yet is experimentally
  inactive as a mutase; glycolysis/mutase terms demoted with propagation_review.
• wago-4 (worm) — an Argonaute lacking the catalytic tetrad; the endonuclease/"slicer" term
  removed on biological grounds, miRNA binding corrected to siRNA binding (binds 22G-RNAs).

3. Family over-propagation removed or demoted

The dominant false-positive class for dark genes is IBA/ISS transfer from a characterized
paralog or distant ortholog whose function does not hold for the target:

• fis-1 (worm) — contrary to the FIS1 family's fission-receptor reputation, worm FIS-1 is
  not required for mitochondrial or peroxisomal fission (LOF is silent); family fission
  terms kept non-core, and its supported role (mitophagy coupling) added as NEW.
• ILT1 (yeast) — PQ-loop transporter annotated as a vacuolar basic-amino-acid transporter
  by propagation from Ypq1/2; ILT1 is experimentally plasma-membrane and in a distinct
  uncharacterized subfamily — vacuolar terms removed.
• LEE1 (yeast) — makorin-family; ubiquitin-ligase activity propagated from RING-containing
  metazoan makorins, but LEE1 has no RING domain (confirmed by direct InterPro query) —
  ligase terms marked over-annotated; only its CCCH zinc-finger zinc-binding kept.
• NVJ3 (yeast) — PI3P-binding transferred from paralog Mdm1, but NVJ3 shares only the PXA
  domain and lacks the PX domain that carries the activity — removed.
• AAD3 (yeast) — aryl-alcohol dehydrogenase activity propagated from a lignin-degrader
  enzyme; the AAD deletants are phenotype-null and budding yeast is not a lignin degrader —
  demoted to the honest superfamily-level oxidoreductase term.
• THI22 (yeast) — HMP-P kinase activity: the one direct assay found no activity for
  THI22, so the propagated kinase term was marked over-annotated (residues intact → not a
  pseudoenzyme, but activity unproven).
• gpa-12 (worm) — Gα subunit carrying propagated Gs/Gi adenylate-cyclase-modulating and
  D5-dopamine-receptor-binding terms wrong for the G12/13 subfamily — removed.

In every case the annotation was flagged as electronic over-propagation with a structured
propagation_review, not by overruling an experimental annotation.

4. Evidence-grounded upgrades

Dark-gene review is not only subtractive; deep reading also finds well-supported functions
missing from GOA:

• MNN14 (yeast) — a falcon-surfaced, PubMed-verified paper (Kang et al. 2021) shows
  recombinant Mnn14 transfers mannose-phosphate from GDP-mannose to high-mannose N-glycans;
  mannosylphosphate transferase activity (GO:0000031) added as an evidence-backed NEW MF.
• spa1 (pombe) — ornithine-decarboxylase-inhibitor activity confirmed directly for the
  fission-yeast protein (recombinant Spa1 inhibits S. pombe ODC) and kept as the core function.

5. Fabrication guardrails held

The supporting_text-must-be-verbatim rule and independent fact-checking caught deep-research
errors before they entered a review:

• tin-44 (worm) — a falcon report's "RNAi extended lifespan 11.1%" was a hallucination; the
  cached full text actually groups tin-44 among lifespan-shortening import knockdowns. Removed.
• NIT1 (yeast, YIL164C) — a falcon report conflated NIT1 with its dGSH-amidase paralog
  NIT2/YJL126W; the misattributed activity claim was fact-checked out (catalytic-residue numbering
  mismatch) and the reference flagged MISCITED.
• prg-2 (worm) — correctly failed rather than being fabricated: it is a WormBase
  pseudogene with no UniProt entry and zero GO annotations; substituted with another gene.
• LPX2 (yeast) — a rebase accident degraded a protected publication cache to abstract-only,
  which had let the review claim the Ploier 2013 full text was "inaccessible." Restoring the
  full text showed that paper tested Ykl050c and found no in-vivo lipolytic activity —
  correcting the review and strengthening its MF-unknown conclusion.

Why this matters

Across 120 dark genes, the value was rarely a brand-new function statement; it was (a) getting the
protein's identity right, (b) refusing to inherit a paralog's activity through a degenerate fold
or a wrong-subfamily transfer, and (c) writing down, with provenance, exactly what remains unknown.
That is the raw material the Function Knowledge Gaps register is
built from.