The problem: an annotation desert

• Cephalopods have the largest invertebrate nervous systems, adaptive camouflage, and remarkable cognition
• Yet only 27 experimental GO annotations exist across all cephalopod species
• That is 0.008% of 333,921 total Cephalopoda annotations
• Only 11 gene products across 7 species have any experimental GO evidence

Almost everything we "know" about cephalopod genes comes from electronic inference, not curation.

How thin is the evidence?

GO annotation status for Cephalopoda (taxon:6605), QuickGO May 2026:

~99% is electronic transfer; the experimental core is vanishingly small.

Key biology: why cephalopods are special

• Protocadherin expansion: >160 genes in octopus (vs ~70 in mammals); neural-enriched
• C2H2 zinc finger expansion: vertebrate-convergent gene family amplification
• Extensive A-to-I RNA editing: >57,000 recoding sites; temperature-responsive
• Taxonomically restricted genes: hundreds of cephalopod-specific genes in skin, suckers, nervous system
• Horizontal gene transfer: 12 bacterial HGT genes in O. bimaculoides, including immune defense

These features make naive homology-based annotation systematically unreliable.

The approach

• Apply the ai-gene-review framework to experimentally characterized cephalopod genes
• Synthesize literature deep research + computational predictions + bioinformatics
• Review each existing annotation against GO guidelines: ACCEPT / MODIFY / REMOVE / over-annotated
• Propose new annotations where strong evidence exists but GO is silent
• Integrate scRNA-seq cell-type context (MF + cell type + cell role → inferred BP)
• Organism dirs follow UniProt mnemonics: OCTVU, OCTBM, DORPE, SEPOF, ...

Status: 16 genes reviewed across 7 species

(8 more: OPR, FMRFa, khc, CTR2, TDO, SERT, reflectin, ADAR2)

Finding 1: systematic IEA errors exposed

• Vasopressin-receptor mis-annotation is pervasive: CTR1, CTR2, OPR all carried GO:0005000 (vasopressin receptor activity) from IEA — none bind vasopressin. Correct term: GO:0008188 (neuropeptide receptor activity)
• ARBA false positives for non-model organisms: kinesin tagged "locomotion", "nuclear migration", "isomerase activity"; AP180 tagged "locomotion", "Golgi apparatus"
• tRNA vs RNA deaminase confusion: ADAR2 mis-annotated with tRNA-specific adenosine deaminase activity (ADAT function), via TreeGrafter

Finding 2: ancestral-function over-annotation

S-crystallin (OCTS1, P27013) — a textbook case:

• Retains GO:0004364 (glutathione transferase activity) from sequence homology
• But has 1/700 the catalytic efficiency of real GST
• Binds GSH 43-fold tighter than GST-sigma — for structural stabilization (raises Tm by ~7 C), not catalysis
• PDB 5B7C; assembles into a colloidal gel for the lens

Sequence homology preserved; the function diverged. The annotation should not.

Finding 3: novel biology captured for the first time

• Dermal photoreception: rhodopsin in skin chromatophores mediates LACE (light-activated chromatophore expansion). Best term: GO:0030265 (PLC-activating opsin-mediated signaling) — child of both GPCR signaling and phototransduction
• Contact chemosensation: CRT1, a cephalopod-specific channel (no homologs outside Cephalopoda), evolved from nAChRs via retrotransposition (26 intronless genes on one chromosome). Bacterial metabolites on eggs activate CRT1 to drive brooding behavior — a microbiome-receptor-behavior circuit (Sepela 2025, Cell)

Finding 4: RNA editing tunes protein function

• Kinesin (khc, DORPE) velocities are tissue-specific via RNA editing:
  • optic lobe variants ~520 nm/s; unedited ~587 nm/s; stellate ganglion ~620-674 nm/s
  • 37 editing sites in the motor domain alone
  • the most direct demonstration of RNA editing tuning protein properties in any organism
• Complexin (cpx) has a dual role: N-terminal domain (residues 1-26) promotes fusion while the central helix clamps spontaneous release — first shown at the squid giant synapse

Finding 5: filling the desert

• Well-characterized genes with near-zero GO coverage:
  • FMRFa (2 annotations), SLC6A4/SERT (0), reflectin (0)
• The project proposed 19 annotations across these 3 genes alone
• Deep research surfaced biology missing from initial reviews:
  • DDO has a visual-system role: D-aspartate at 2.30 umol/g in retina (PMID:15491279)
  • Symplectin may be bifunctional: conserved pantetheinase triad (E60-K163-C196); unique cation specificity K+ > Rb+ > Na+ (not Ca2+)
  • OPR is an antidiuretic-hormone receptor (octopressin lowers hemolymph osmolarity)

scRNA-seq x function: inferring biological process

Logic: gene G has MF F + is a marker for cell type C + C has role R → G involved in BP = F in R context

Yield is small: a Swiss-Prot bottleneck (~35 useful entries) and cross-species mapping limit reach.

Why curation here is hard

1. Near-zero experimental GO — only 27 annotations; most curation is necessarily novel
2. Fragmented UniProt — E. scolopes has 0 Swiss-Prot entries despite being a major model
3. No nomenclature committee — gene names inconsistent across species
4. Multi-species biology — same gene studied in different species; per-species vs consolidated reviews
5. Large genomes — 2.7-5.7 Gb, 42-71% repeats; gene counts vary 2-fold between methods
6. Limited reverse genetics — only 2 CRISPR papers exist (D. pealeii TDO; E. berryi TDO+IDO)

Conclusions & future directions

• AI gene review converts an annotation desert into curated, evidence-backed function statements
• It both fixes systematic IEA/ARBA errors and captures genuinely novel cephalopod biology (dermal photoreception, contact chemosensation, RNA-edited motors)
• Status: 16 genes / 7 species reviewed; project IN_PROGRESS (FLAGSHIP)
• Next:
  • Request Swiss-Prot review of O. bimaculoides TrEMBL markers (dat, vacht, nos, tyrbh)
  • Build GO-CAM models for the visual and chromatophore systems with cell-type context
  • Leverage expanding E. berryi CRISPR knockouts for functional inference