Arabidopsis thaliana corpus

Bottom line: the repository holds 282 Arabidopsis thaliana gene reviews
covering 8,746 annotation rows, but they were built by many separate projects
and nobody has yet looked at them as one body of work. This project does that.
It summarizes the corpus several ways (by action, by evidence type, by term,
by module, by family, by GO-CAM) and then uses those views to pick follow-up
curation. Excluding protein binding, the reviews accept 56% of rows, keep 26%
as non-core, modify 8%, mark 4% over-annotated and remove only 3%, and they add
185 NEW rows. protein binding alone is 1,954 rows, 22% of the corpus, and
accounts for 83% of all removals. The main gaps are bookkeeping and coverage,
not biology: 22 reviews are fully actioned but still say INITIALIZED, 7 have
no status at all, 58 rows in 23 genes are UNDECIDED, 71 reviews belong to no
module, only 9 PANTHER family reviews touch an Arabidopsis gene, and the GO-CAM
index has one Arabidopsis model, none of whose genes we have reviewed.

All numbers on this page come from
ARATH/scripts/arath_corpus_stats.py,
which writes ARATH/ARATH-corpus-stats.yaml
(per-gene rows, module and family membership, term rankings). Re-run it after
any curation round and update the numbers here; the snapshot below is from
2026-10-14.

Use UniProt code ARATH. Folders are named by gene symbol where one exists, and
otherwise by AGI locus (e.g. AT4G17750 for HSFA1A) or UniProt accession.

Corpus at a glance

Review status

None of the 8,746 rows is still PENDING, so the 22 INITIALIZED reviews
have in fact been worked through; their status field was never updated. The 7
reviews with no status are the cytokinin two-component set.

View 1: actions

Read the first column with care: most of its removals and modifications are the
project-wide rule that protein binding says nothing about function. The middle
column is the better measure of how often the reviews disagreed with GOA about
biology, and there the removal rate is under 3%.

View 2: evidence type against action

Evidence codes are grouped as experimental (EXP, IDA, IPI, IMP, IGI,
IEP and the high-throughput codes), phylogenetic (IBA, IBD, IKR, IRD),
electronic (IEA), similarity (ISS, ISO, ISA, ISM, RCA, IGC) and
author/curator (TAS, NAS, IC, ND).

Two things stand out. Experimental rows carry most removals because IPI
protein binding is experimental. And phylogenetic rows (almost all IBA) are
the most often accepted class (73%), which fits the repo's position that an IBA carries a PAINT
curator's considered phylogenetic judgment. Similarity rows are removed about
as often as experimental ones (15% against 16%), but for a different reason:
ISM organelle-targeting predictions (next section).

View 3: recurring term-level corrections

• protein binding (GO:0005515): 1,954 rows; 863 removed, 744 modified to
  a specific binding or adaptor term, 249 marked over-annotated.
• Organelle localizations from prediction or proteomics: 26 of 95
  chloroplast (GO:0009507) rows were removed. Of the chloroplast and
  mitochondrion (GO:0005739) rows that were removed or marked over-annotated,
  35 are ISM (targeting-peptide prediction) and 17 are HDA (organelle proteomics).
  This pattern is plant-specific and worth checking as a class (see plan).
• Transcription-factor terms: DNA-binding transcription factor activity
  (GO:0003700; 77 of 244 rows modified), DNA binding (52 of 101) and
  regulation of DNA-templated transcription (23 of 130) are routinely
  sharpened to activator, repressor or sequence-specific children.
• Kinase terms: protein kinase activity and protein serine/threonine kinase activity are modified in 8 rows each, mostly to receptor or MAPK
  kinase children.
• Most common NEW terms: transcription corepressor activity (6 rows),
  ubiquitin-like ligase-substrate adaptor activity (5), ESC/E(Z) complex (5),
  DNA-binding transcription activator activity (4), cellular response to
  strigolactone (4). Full rankings are in the stats file.

View 4: modules

Modules are the main way the corpus is organised by pathway. 29 modules contain
at least one reviewed Arabidopsis gene:

11 further modules mention Arabidopsis but contain no gene we have reviewed.
Most are cross-kingdom metabolic modules where Arabidopsis is one of several
representative organisms:

• Plant cellulose biosynthesis module
• Endoplasmic reticulum unfolded protein response
• Eukaryotic (THI4/THI5-type) de novo thiamine diphosphate biosynthesis
• Folate one-carbon carrier-state interconversion
• L-methionine biosynthesis (from homoserine)
• Oxygenic photosynthesis module
• Pyrenoid-based CO2-concentrating mechanism
• Siroheme biosynthesis from uroporphyrinogen III
• De novo tetrahydrofolate and folylpolyglutamate biosynthesis
• Glycerolipid / triacylglycerol biosynthesis (glycerol-3-phosphate and DHAP routes)
• Plant xylan (glucuronoxylan) biosynthesis module

71 reviews belong to no module: ABCG36, ABI1, ABI5, AGO1, APO1, AT1G06680, AT1G32330, AT1G74310, AT2G26150, AT3G02990, AT4G17750, AT4G38370, AT5G02500, AT5G03720, AT5G16820, AT5G52640, At1g67980, BGLU26, BUB3.1, CASPL1B1, CASPL1B2, CASPL1D1, CASPL1D2, CASPL4C1, CBF1, CER1, CHL1, CIPK24, DCL1, DRB1, DREB2A, DRS1, EFR, ERF094, F4JLB7, FAD2, FBL22, FER, FLS2, FRK1, FTSH12, GIP1, HEN1, HKT1, HSP17.6A, HSP17.7, ICS1, LCR1, LYM2, MLO2, MYB122, MYB34, MYB51, NCED3, NEK3, NPR1, OST1, PYR1, Q1G3K8, RCA, RDR6, RPM1, RPS2, SIZ1, SOS1, SPC24, SYP121, TT8, UVR8, WRI1, WRKY70.

Several of these fall into obvious groups that have no module yet: ABA
perception and signaling (PYR1, ABI1, OST1, ABI5, NCED3), small-RNA biogenesis
and silencing (DCL1, DRB1, HEN1, AGO1, RDR6), pattern-triggered immunity
(FLS2, EFR, FRK1, FER), salicylic acid and systemic resistance (ICS1, NPR1,
WRKY70, RPM1, RPS2), salt tolerance (SOS1, CIPK24, HKT1), indolic glucosinolate
regulation (MYB34, MYB51, MYB122) and the heat-stress set from
ARATH_HEAT_STRESS.

Since this snapshot, modules have been drafted for ABA signaling, miRNA
biogenesis, pattern-triggered immunity, salicylic acid / NPR1, the SOS
pathway and heat stress. Of the groups above, only indolic glucosinolate
regulation still has no module. The counts in this section will change when
the stats script is next re-run.

Pathways not yet curated

Arabidopsis pathways not yet curated checks 107
Arabidopsis pathways against the modules. The pathway list comes from KEGG's
A. thaliana maps plus plant processes that KEGG has no map for. 34 pathways
have a module with a reviewed Arabidopsis gene. The other 73 do not: 35 have
no module, 21 have only a cross-kingdom module that never mentions
Arabidopsis, and 17 have a module with no reviewed Arabidopsis gene. The
largest gap is plant-specific metabolism: hormone biosynthesis (all but GA and
SA), starch and sucrose, flavonoids and anthocyanins,
cuticle and plastid lipids, phytosterols, and nitrate and ammonium
assimilation.

View 5: families

Two of these (PTHR47821, PTHR48004) cover the ProtNLM benchmark proteins;
the rest arose from single-gene reviews. None of the large plant
transcription-factor or receptor-kinase families that dominate the corpus
(MADS-box, ARF, AUX/IAA, type-B ARR, DELLA, LRR-RLK) has a family review yet.

View 6: GO-CAMs

gocams/index.tsv has one Arabidopsis model,
67e5e74400000148 "Arabidopsis MRG domain proteins bridge two histone
modifications", with ASHH2, ASHH3, MRG2 and SCL27. None of these four has a
review here. We have not yet checked whether this reflects the number of
production Arabidopsis GO-CAMs or only what has been cached.

Related projects

Focused Arabidopsis projects:
ARATH_HEAT_STRESS,
ARATH_SINGLE_CELL_FUNGAL,
CASPL_FAMILY,
PLANT_FUNGAL_INTERACTIONS,
PHOTOSYNTHESIS,
PLANT_BIOENERGY.
Another 20 cross-cutting projects (obsoletions, evaluations, evidence-code
studies) list ARATH among their species; the stats file names them all.

STATUS

Last updated: 2026-10-20

Phase 1: summarize the corpus

• [x] Reproducible corpus statistics script and YAML snapshot
• [x] Action, evidence-class, term, module, family and GO-CAM views (this page)
• [ ] Per-gene table (status, rows, actions, module) rendered as a sub-page
• [ ] Narrative synthesis by biological area (hormone signaling, flowering,
  development, immunity, specialised metabolism, abiotic stress)
• [ ] Compare against TAIR/GOA: what share of all Arabidopsis GOA rows and of
  experimentally annotated Arabidopsis genes the corpus covers

Phase 2: consistency review of the standard annotations

• [ ] Set status on the 22 fully actioned INITIALIZED reviews after a spot
  check: AP3, AT4G17750, AT5G16820, AT5G52640, BIK1, CIPK24, CLV3, DREB2A, EFR, FER, FKF1, GAI, HKT1, LHY, MUTE, PI, PIF3, PIF4, RCA, SPCH, TOC1, UVR8
• [ ] Add status to the 7 reviews that lack it: AHK2, AHK3, AHK4, AHP1, ARR1, ARR12, ARR5
• [ ] Audit the 68 DRAFT reviews and promote the finished ones
• [ ] Resolve the 58 UNDECIDED rows: ABI1 (12), AT5G52640 (2), BUB3.1 (1), CASPL1D1 (5), CHL1 (1), DRB1 (3), DRS1 (3), F4JLB7 (4), FBL22 (3), FLS2 (1), FRK1 (1), FTSH12 (3), GSTF11 (1), GSTF6 (2), HEN1 (2), MKK4 (1), NPR1 (1), NSP1 (5), PAS2 (2), RCA (1), SUN2 (1), SUR1 (1), WIP1 (2)
• [ ] Check ISM/HDA chloroplast and mitochondrion calls as a class, so the same
  kind of evidence gets the same action across genes
• [ ] Check that sibling genes in the same module received consistent actions
  for shared terms (e.g. the three GID1s, the SEPALLATAs, the AHKs)
• [ ] Add deep research for the 85 reviews that have none
• [ ] File the draft reports in upstream-reports.md
  (GO-CAM HAM1/SCL27; TAIR SUR1, F4JLB7/RIC7, GSTF6) and record the tickets
• [x] Bring OST1 in line with SnRK2.2/SnRK2.3: all three now MODIFY protein
  kinase activity (GO:0004672) and kinase activity (GO:0016301) to GO:0004674
• [ ] Re-run the corpus statistics after the 2026-10-19 additions

Phase 3: modules

• [x] Draft modules for ABA signaling, miRNA biogenesis, pattern-triggered
  immunity, salicylic acid / NPR1, SOS salt tolerance and heat-stress response
• [ ] Draft a module for indolic glucosinolate biosynthesis and its MYB regulation
• [x] Map Arabidopsis pathways to modules and list the uncurated ones
  (ARATH-pathway-gaps)
• [ ] Draft modules for plant-specific pathways that have none, starting with
  those that already have reviewed genes: ABA biosynthesis (NCED3),
  flavonoids/anthocyanins (TT8), cuticle lipids (CER1), UV-B (UVR8), cold
  acclimation (CBF1)
• [ ] Draft the remaining hormone biosynthesis modules (auxin, cytokinin,
  brassinosteroid, jasmonate, ethylene, strigolactone)
• [x] Draft a photorespiration module
  (photorespiration_c2_cycle)
• [x] Review the Arabidopsis representatives of the photorespiration module
  (18 genes: PGLP1, PLGG1, GLO1, GLO2, CAT2, GGAT1, AGT1, HPR, HPR2, GLYK,
  GLN2, GLU1, SHM1, GLDP1, GLDP2, GDCH, GDCST, mtLPD2)
• [x] Review mtLPD1 (At1g48030), the main leaf glycine decarboxylase L protein,
  and GGAT2 (At1g70580)
• [ ] Consider reviewer suggestions for the photorespiration module: BASS6 as a
  second glycolate exporter, the shade-induced cytosolic GLYK isoform, viable
  Arabidopsis gln2 mutants, and the H-protein MF term in
  glycine_cleavage_system (lipoic acid binding vs molecular carrier activity)
• [ ] Draft starch and sucrose modules
• [ ] Review Arabidopsis genes for modules that mention Arabidopsis but have no
  reviewed gene (photosynthesis, cellulose and xylan first)
• [ ] Add Arabidopsis representatives to the 21 cross-kingdom modules that
  never mention it (TCA cycle, pentose phosphate, MEP pathway and others)
• [ ] Compare the pathway map with AraCyc/PMN
• [ ] Move the 29 populated modules from DRAFT toward review

Phase 4: families

• [ ] Family reviews for the large plant families in the corpus (MADS-box,
  ARF, AUX/IAA, type-B ARR, DELLA/GRAS, LRR-RLK), checking IBA propagation
  within each. Family ids to be looked up in interpro/panther/, not assumed.

Phase 5: GO-CAMs

• [ ] Find the production Arabidopsis GO-CAMs and cache those that touch
  reviewed genes
• [ ] Review the cached MRG-domain model and decide whether its four genes
  need reviews

NOTES

2026-10-14

Project created. Wrote the stats script and took the first snapshot. Counted
rows include NEW rows, since those live in existing_annotations. Module
membership is found by matching module UniProt accessions to review ids, plus
explicit genes/ARATH/<GENE>/ paths; family membership by genes/ARATH/ paths
in PANTHER family reviews.

2026-10-19

Followed up the handoff from the previous session.

• New reviews: CBL4 (SOS3), CBL10 (SCaBP8), SLAC1, SnRK2.2 (SRK2D), SnRK2.3
  (SRK2I) and SERRATE (SE). No deep-research provider worked in this session
  (Edison authentication failed and the OpenAI key was rejected), so each review
  rests on hand-written notes in its -notes.md. None has a deep-research file.
  All validate; all are DRAFT.
• Paralogs were aligned on shared terms. CBL4 and CBL10 both use GO:0043539 for
  CIPK24 activation. SnRK2.2 and SnRK2.3 both MODIFY GO:0004672 to GO:0004674
  and GO:0010029 to GO:0010187.
• GAI, RGA and GID1A: GOA was re-fetched, and their IBA rows (7, 7 and 4) were
  reviewed against the PTHR31636 and PTHR23024 family reviews. GID1A lost two
  ARBA rows; GO:0009939 was re-proposed as NEW because a core function cites it.
• AGO1: the RNA endonuclease activity IDA row is now MODIFY to GO:0090624.
• GID1A notes: the source status in the GID1B cross-reference now refers to
  PTN009058710, not AADAC.
• Draft upstream reports are in upstream-reports.md.
  None has been filed.

Added the pathway gap analysis: a hand-curated pathway-to-module map
(ARATH/ARATH-pathway-map.yaml), a script that checks it against modules/
and genes/ARATH/ (ARATH/scripts/arath_pathway_gaps.py), its output
(ARATH/ARATH-pathway-gaps.yaml) and a write-up
(ARATH-pathway-gaps). The corpus stats file was
not regenerated in this change, so the other numbers on this page are still
from the 2026-10-14 snapshot.

2026-10-20

Drafted the photorespiration_c2_cycle module (ten parts from 2-phosphoglycolate
phosphatase to D-glycerate 3-kinase, plus ammonia refixation; glycine
decarboxylase is included by reference to glycine_cleavage_system).
Representatives are Arabidopsis proteins checked against UniProt; none has a
gene review yet, so photorespiration now counts as "module without reviewed
ARATH genes" in the gap analysis.

Reviewed the 18 Arabidopsis members of the photorespiration module; the
reviews were written by parallel agents, then checked and committed in batches. Deep research
could not be run in this environment (no provider API keys), so each gene's
notes file is built from cached publications and UniProt. Two corrections came
out of the reviews: the module now names mtLPD1 (At1g48030) rather than mtLPD2
as the main leaf L protein, and proteomics-only location rows follow the
organelle audit (marked over-annotated, not removed). Photorespiration now
counts as curated in the gap analysis (34 of 107 pathways).

Reviewed mtLPD1 and GGAT2. mtLPD1 evidence (leaf GDC holds about four times more
L1 than L2, PMID:32311173; mutants show mtLPD1 predominant in photorespiration,
PMID:41990722) is now cited in the photorespiration module. GGAT2's
photorespiration row rests on enzyme activity, leaf peroxisome peptides and
expression only; no ggat2 mutant has been described.

2026-10-24

• OST1: the four general kinase MF rows now MODIFY to GO:0004674, matching SnRK2.2 and SnRK2.3.
• OpenScientist checked the two single-study potassium channel inhibitor calls. Neither finding
  has been independently replicated. For CBL10/AKT1 the in-planta data are overexpression-only,
  so AKT1 inhibition was dropped from CBL10's core functions. For SLAC1/KAT1, slac1
  loss-of-function shows reduced rather than increased inward K+ current (PMID:23126621). Both
  keep their MODIFY rows with caveats. SLAC1's KAT1 row uses GO:0070320 (inward rectifier
  potassium channel inhibitor activity), matching CBL10's AKT1 row; its AKT2 row stays on
  GO:0019870 because AKT2 also has non-rectifying gating modes (PMID:16316977). Reports are in
  each gene's -hypotheses/ folder.