Unreviewed phosphorylation demo candidates

48 protein-coding genes: 31 mouse, 13 rat, 2 fly, and 2 zebrafish.
Selected on 2026-09-28 PDT (2026-09-29 UTC). All have a current positive
involved_in annotation to the target process, no positive protein-kinase
activity annotation in the current fetched GOA, and no gene or prediction
review file in this checkout when checked by symbol, UniProt aliases, primary
accession, and secondary accessions. Even partial review stubs are excluded.
This is a list of review candidates, not completed reviews or predetermined
REMOVE decisions. No review stubs have been created.

Parent project ·
Candidate table with identifiers and rationales ·
Plain species/gene list ·
62 exact source annotation rows

ARBA follow-up: Generic-rule audit checked all 90
conditions of ARBA00027234 and found 370 matching Swiss-Prot proteins, all
carrying protein-kinase annotations. No current QuickGO rows explicitly named
this rule or the four tested phosphorylation-descendant rules. No defensible
ARBA-derived additions were made; the demo list remains 48 genes.

Demo use

The first 12 entries are suggested starting points, mixing wrong-substrate,
substrate, phosphatase, signaling, and cross-species propagation examples.
There are 39 high-priority candidates and 9 medium-priority discussion cases.
Priority describes the strength/usefulness of the triage hypothesis, not a
completed evidence assessment. The medium group contains scaffolds and
complex-associated proteins for which direct participation could justify an
annotation even without intrinsic kinase activity.

The list contains 35 genes not previously individually triaged in the old
project's cross-species tables and 13 already discussed there without gene
review YAMLs. prior_project_triage makes that distinction explicit. Orthologs
are separate species/gene targets, not independent mechanistic examples.
The main list uses protein-coding genes with verified UniProt accessions so it
can be used with the usual gene-fetch workflow.

Why there are no human candidates

The rerun of the local human protein-phosphorylation-minus-protein-kinase
query returned 42 objects. Every gene-symbol hit already had a review; the
remaining object was the Cyclin A1–CDK2 complex, not an unreviewed gene.
A live QuickGO check also exposed four accession/alias-only hits (O15302,
Q15453, Q96GZ3, Q9H4D1); UniProt identifies these as kinase proteins, so they
were excluded. Human LIMD1 and TOLLIP from the generic-phosphorylation query
are also already reviewed. This is the result of this screen, not a claim
that every possible human phosphorylation error has been exhausted.

Method and safeguards

1. Reran the project set difference in local go-db snapshots: goa_human,
   goa_mouse, mgi, rgd, fb, zfin, and tair. The broader follow-up
   also checked wb, sgd, and pombase for additional candidates.
2. Focused on GO:0006468 protein phosphorylation and is-a/part-of
   descendants, subtracting positive GO:0004672 protein kinase activity
   and descendants. Required biological-process aspect; part-of traversal
   can otherwise include molecular-function terms. Explicitly excluded
   activation/regulation terms and NOT assertions on both sides.
3. Required the positive relation to be exactly involved_in. The large
   MGI candidate expansion mostly used acts_upstream_of_or_within; those
   are not claims of direct participation and were excluded. GO documents
   the distinction in its annotation relations guide.
4. Added five generic GO:0016310 phosphorylation cases (mouse/rat Limd1 and
   Tollip, zebrafish tollip), using subtraction of all kinase activity for
   this discovery arm. Added one carbohydrate-phosphorylation case,
   rat/Epm2a, a glycogen phosphatase. Thus 42 genes have protein-specific
   phosphorylation rows, 5 generic phosphorylation rows, and 1 carbohydrate
   phosphorylation row. Legitimate small-molecule kinases were not selected
   merely because they lack protein-kinase activity.
5. Refetched all annotations for selected UniProt products from QuickGO,
   checked current ontology descendants, and inspected current UniProt
   identity/function records. All 48 lack positive protein-kinase activity
   rows. This is supporting triage evidence, not proof of absence of catalysis.
6. Excluded known kinase subunits Phka1/Phka2/Phkb, the Hspa9
   autophosphorylation case, the TP53RK-binding protein, unresolved Q6PIU9,
   true kinases missed by names, and noncoding RNA hits. The prior Arabidopsis
   examples TOPP4/CDC25/RIN4/PI4KA1 currently use the broader upstream/within
   relation in QuickGO, so they are not in this direct-participation demo set.
7. Matched repository reviews using symbols and aliases within species and
   primary/secondary UniProt accessions across the repository. No shortlisted
   gene has an existing review file. This check is scoped to the checkout,
   not unmerged work in other branches or unpublished reviews.

The SQL reproduces discovery against a local snapshot.
Source databases have different dates, so the local query is a candidate
generator; the live row check determines eligibility. The dated
snapshot retains current ontology responses, UniProt identity
and function records, all fetched molecular-function rows, the selected
process rows, and total annotation counts. annotations.tsv preserves
qualifiers, evidence, reference, WITH/FROM, assigning group, date, and annotation
extensions. Multiple GOA rows and duplicate protein records are not counted as
multiple genes (e.g. mouse Pick1).

Ranked list

Codes follow the parent project: S substrate, W wrong substrate,
O opposite reaction, R regulator/signaling, C complex attribution,
U ubiquitin-system mechanism. They are hypotheses to investigate.
Each UniProt link supports protein identity/function; GO IDs and references
identify the actual candidate assertion. Complete row-level provenance is in
annotations.tsv, including orthology donors for ISO/ISS annotations.

Mouse — 31

Rat — 13

Fruit fly — 2

Zebrafish — 2

Useful literature entry points

These spot-checks establish why several newly selected mouse genes are good
questions for a demo; they are not full annotation reviews.

• Klhl3: PMID:24821705
  identifies impaired WNK1/WNK4 ubiquitination/degradation and consequent
  activation of the WNK–OSR1/SPAK–NCC phosphorylation cascade. The question is
  whether the ligase adaptor belongs to regulation of that process.
• Ppme1: PMID:31213650
  explains how PME-1 inactivates PP2A through demethylation, increasing IRF3
  phosphorylation. The mechanism is control of a phosphatase.
• Pkd1: PMID:20181743
  measures flow-induced HDAC5 phosphorylation downstream of polycystin signaling.
• Sqstm1: PMID:32385399
  identifies p38 as the ATF2 kinase and p62 as an ATF2-binding scaffold. This is
  intentionally a medium-priority case: determine whether scaffolding contributes
  directly to the reaction or acts elsewhere in the signaling/transcriptional response.
• Usp25: PMID:23042150
  reports that USP25 deficiency increases I-kappaB/JNK phosphorylation and
  identifies TRAF5/TRAF6 deubiquitination as the mechanism.

Do not reject experimental rows from titles or abstracts alone. Read the
relevant full text during the demo; use UNDECIDED when the necessary evidence
cannot be accessed. Being noncatalytic is not, by itself, a reason to remove a
process annotation from a direct participant or scaffold.

Recheck before the demo

From the repository root:

# Offline: checks the dated evidence and detects reviews added since selection.
uv run --script projects/PHOSPHORYLATION_REFACTOR/demo/verify_candidates.py

# Live: also refreshes GOA, ontology descendants, and UniProt identities.
uv run --script projects/PHOSPHORYLATION_REFACTOR/demo/verify_candidates.py --live

# Discover more mouse candidates from a local go-db snapshot.
duckdb -readonly ~/repos/go-db/db/mgi.ddb -csv \
  < projects/PHOSPHORYLATION_REFACTOR/demo/find_candidates.sql

The verifier exits nonzero for an existing review, a duplicate, a taxon
mismatch, positive protein-kinase activity, a missing candidate process row,
or a change in the listed terms/evidence/references. It does not make curation
decisions. It leaves the dated snapshot and gene files untouched.