2026-10-10. CDK13 (Q14004) is a nuclear cyclin-dependent protein kinase with a long arginine/serine-rich region. This review preserves all 43 normally seeded annotations, their original source/evidence/qualifier/with-from fields, and both product records. The decisions are 28 ACCEPT, 10 KEEP_AS_NON_CORE, four MODIFY and one UNDECIDED. No NEW annotation is proposed. References to row numbers below are zero-based positions in the preserved seed.
The standing ClinGen project instruction governs supported generic protein interactions: retain them as non-core unless a more informative molecular function is justified. The three retained generic interactions are p32/C1QBP and two source-specific CDC37 associations. The four cyclin K interactions are refined to cyclin binding, because the partner class and the target's role are independently established. No partner's kinase, chaperone or RNA-degrading activity is assigned to CDK13.
PMID:20952539(https://pubmed.ncbi.nlm.nih.gov/20952539/) includes human CDK13 purified from HeLa nuclear material and its GST-CTD kinase assay in Figure 4C. Selected indexed full Results, Figure 7 and purification Methods were inspected independently, even though the unchanged repository cache is abstract-only. The CDK12-focused title does not exclude the CDK13 experiment. The paper's restrained in-vivo conclusions are retained; CDK12-specific results do not establish CDK13 DNA-repair activity.
PMID:22012619(https://pubmed.ncbi.nlm.nih.gov/22012619/) separates human cyclin K/CDK12 and cyclin K/CDK13 complexes in reciprocal experiments. The genuine normal full-text refresh was read for the selected complex/kinase Results. The initial normal abstract output remains archived. Only the new full cache, absent from the fixed main baseline, is proposed for publication.
PMID:26748711(https://pubmed.ncbi.nlm.nih.gov/26748711/) provides the load-bearing CDK13/cyclin K structure, CTD phosphorylation and human knockdown/rescue evidence. The authentic publisher PDF was inspected for selected Results, Figures 1, 4, 5 and 7 and construct/rescue Methods. Recombinant human protein domains expressed in insect cells are distinguished from full-length complexes recovered from HCT116 cells. CDK13 phosphorylates CTD Ser2/Ser5; recognition of a pre-existing phospho-Ser7 substrate does not demonstrate Ser7 catalysis. Existing transcription/elongation annotations are retained on phosphoregulation and transcript-output evidence, without inventing a measured elongation velocity. GO:0008024 describes the transcriptional cyclin/CDK complex class; the core uses the specific cyclin K-CDK13 complex.
The second catalytic core is protein serine kinase activity on RNA-processing proteins. PMID:37079685(https://pubmed.ncbi.nlm.nih.gov/37079685/) was read in the genuine full XML-derived cache, especially Figure 4D–I and corresponding Results. Human melanoma CDK13 interacts with nuclear ZC3H14; CDK13/cyclin T1 directly phosphorylates ZC3H14 Ser475 in a wild-type versus S475A substrate assay. Phosphomimetic recruitment experiments connect this site to PAXT-associated RNA-surveillance machinery. CDK13 performs phosphotransfer, not RNA degradation. No new RNA-decay process is inferred merely from perturbation or rescue. Zebrafish and mouse experiments in the paper retain their separate species scope.
The human CDC2L5 abstract in PMID:16721827(https://pubmed.ncbi.nlm.nih.gov/16721827/) independently establishes nuclear/speckle localization and changes in splice-site selection. PMID:18480452(https://pubmed.ncbi.nlm.nih.gov/18480452/) explicitly connects CDK13 to ASF/SF2 phosphorylation and splicing; only its authentic abstract was recovered, so uninspected kinase controls are not claimed. The directly read ZC3H14 assay supplies an independent catalytic example rather than treating a splicing phenotype as proof of every substrate relationship. Neither a splicing influence nor localization assigns intrinsic RNA cleavage to CDK13.
CDK13-source-evidence.json preserves finite target records, source URLs and raw-response hashes. The IntAct extraction retains exact accessions, method, target/partner taxonomy, host, features, source publication and negative-interaction flag. Repeated release entries do not count as independent replication.
Row 6, RNA binding, is retained at screen resolution. The original PMID:22681889(https://pubmed.ncbi.nlm.nih.gov/22681889/) journal full text was not directly recovered. The first author's authentic institutional thesis, The mRNA-bound Proteome of a human cell-line and in early Drosophila embryogenesis, explicitly identifies the published HEK293 project in its contributions statement. Table S1 (PDF page 145) contains the exact CDK13 class II hit: enrichment in two of three UV/oligo(dT) capture experiments. Table S3 (page 150) distinguishes positive construct expression, failed immunoprecipitation and a blank CLIP result. A failed IP does not establish a negative RNA-binding test. Sequence specificity and a narrower RNA-target function remain unresolved. The evidence file records these manually transcribed cells and the downloaded PDF hash; it is not a fabricated publication cache.
Row 18 is contextual proteomic detection. Authentic PMC XML for PMID:22664934(https://pubmed.ncbi.nlm.nih.gov/22664934/), Table II, contains the exact CDK13_HUMAN entry, verified against Q14004. The normal full-text markdown omitted this table. The study used group-pooled Schirmer tear samples from 25 breast-cancer patients and 25 controls. Detection supports a non-core extracellular location, not active secretion or physiological extracellular kinase activity. For rows 19/42, the genuine Reactome event and input/output sets explicitly contain CDK13 in the ficolin-rich granule lumen and extracellular context. These are curated event/proteomic locations, not independent catalytic experiments.
The ortholog used in the relevant automated and ISS assertions is mouse Cdk13 Q69ZA1. Its experimental splicing, cyclin and nuclear-complex annotations cite PMID:17261272(https://pubmed.ncbi.nlm.nih.gov/17261272/). The authentic abstract and donor annotations were checked; the HEK293T assay host does not make a mouse construct human. Independent human CDK13 evidence supports the retained broad functions. PAINT assertions are inherited-node judgments; target Q14004 appearing among experimental descendants is legitimate provenance, not circular evidence. No uninspected ancestral-node placement or PANTHER identifier is invented.
Row 33 remains a non-core hematopoietic context. PMID:1731328(https://pubmed.ncbi.nlm.nih.gov/1731328/) cloned human CHED, but the abstract explicitly places the antisense megakaryocyte-development assay in murine bone-marrow cultures. The review preserves the original assertion and this organism boundary; it does not describe human marrow experiments.
Row 40 retains the cell-cycle regulation inference as non-core. The actual ontology path is GO:0004693 is_a GO:0097472 part_of GO:0051726, consistent with GO_REF:0000108. Independently, PMID:41680470(https://pubmed.ncbi.nlm.nih.gov/41680470/) Figure 2d assays CDK13-specific knockdown and overexpression in human 786-O cells. The independently viewed publisher histograms show increased G1/decreased G2 fractions after knockdown and the opposite direction under overexpression. These are endpoint phase distributions, not measured transition rates. The displayed S fraction decreases after knockdown, contrary to the Results sentence saying both G1 and S increased; that sentence is not adopted. Figure 4e/f measures EdU incorporation, not a phase-specific rescue. No dual-inhibitor selectivity, CDK12 activity or CDK13-derived circular-RNA function is transferred to this protein.
Row 26, regulation of signal transduction, remains UNDECIDED. The full PMID:26748711 analysis reports signaling-category enrichment among affected transcripts. The inspected assays do not resolve a particular signaling rate or extent regulated by CDK13 itself. This evidence-specific limit is not a claim that the source curator used the wrong gene. The demonstrated kinase and transcriptional functions remain retained.
Normal fetch completed before manual review. Falcon reached the bounded time limit; the explicit Perplexity-lite attempt returned insufficient quota. No provider report was produced or fabricated. This file is the manual research record. Exact-main publication and PANTHER cache bytes are retained, separately from fresh normal variants. The current source-evidence file records primary-table and machine-record extracts; the isolated research packet additionally preserves source downloads, access receipts, immutable normal seed and independent core, screen and cell-cycle peer reports.
A selected full-text read is reported as such and does not claim that every figure or supplement was inspected. Full-text availability flags reflect actual external access as well as the normal cache. Abstract-only sources retain that limitation. All short quotations are confined to the annotation YAML; this notes file adds no verbatim source excerpts. The per-source quotation audit includes repeated excerpts and is an authoring check, not a repository validation rule.
This entry supersedes the initial decision count and unresolved signaling assessment above: 28 ACCEPT, 10 KEEP_AS_NON_CORE, four MODIFY and one MARK_AS_OVER_ANNOTATED. All 43 source objects, both product records and both molecular cores remain unchanged.
The full selected PMID:26748711(https://pubmed.ncbi.nlm.nih.gov/26748711/) Results make the measurement boundary concrete. Human HCT116 knockdown, expression microarrays, selected qRT-PCR and RNAi-resistant rescue establish transcript-output effects. Figure 7C categorizes affected genes by GO enrichment; it does not test transmission of a signal or a defined signaling response. Row 26 therefore marks the specific signaling-process assertion as over-annotated. This does not dismiss CDK13's established transcriptional regulation or claim universal absence of a signaling role.
The three externally inspected but abstract-only cached references (PMIDs 20952539, 26748711 and 22547058) now omit the ambiguous availability boolean. Their reference assessments explicitly distinguish repository-cache depth from actual external access, and short external evidence anchors use supporting_text_fulltext. The earlier blanket statement about availability flags is superseded by this explicit distinction. The schema's field description is general, while the validator's warning discusses uncached text; neither justifies pretending that actually accessed primary sections were inaccessible. External URLs, sections, download/response hashes and access limits are included in the existing evidence artifact. The protected publication caches were not edited.
Row 40 remains non-core on the directly tested endpoint phase-distribution changes after CDK13 loss or gain in human 786-O cells. PMID:41680470(https://pubmed.ncbi.nlm.nih.gov/41680470/) contains an internal S-phase discrepancy between its Results sentence and the displayed representative Figure 2d histogram. It is now recorded in a findings[].finding_review and the evidence artifact, including the actual PDF page, hash and panel values. The annotation reason does not choose an S direction and does not infer transit rates or a phase-specific rescue. This is a bounded internal inconsistency, not rejection of the paper's entire cell-cycle result.
The suggested Table S1 sign correction was checked against the actual Baltz thesis page 145. The source really prints positive 4.52 alongside negative L1/L2 values. Light-labeled UV capture produces negative heavy-to-light log ratios; the heavy-labeled swap reverses the direction. The artifact now names the value as source-reported and explains the ratio convention. It preserves the published number instead of replacing it with an arithmetic average of signed raw ratios. The target-level RNA-binding assessment is unchanged; that screening reference is categorized as supporting rather than central literature.
The optional C1QBP isoform-field addition is not made: the source assertion is gene-level, and the independently known tested-product detail stays in the reason and notes. The two CDC37 source interactions remain non-core at their recorded association resolution; no new binary interface or folding activity is assigned. The already supported protein-serine-kinase core remains grounded in the full ZC3H14 experiment; additional substrate and complex annotations are not needed to address this follow-up. The biological description was tightened by removing a curation-style contrast with RNA degradation.
All newly authored quotations occur only once in the YAML, including the disputed finding's source sentence. The aggregate audit includes both supporting_text and supporting_text_fulltext; this notes entry and the evidence interpretation add no verbatim source prose.