Oncofusion

start / graph / ana-utp18-is-not-in-the-mll-complex-it-is-wdr5-miscurated

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q74-interactome-gap

UTP18 is not a KMT2A partner - it is WDR5, mis-curated into IntAct from the founding paper, and it manufactured the only KMT2A-bait gap in the interactome recount

also called UTP18 · WDR50 · Q9Y5J1 · WDR5 curation error · EBI-592993 · EBI-592758

A public interaction database lists a nucleolus protein called UTP18 as part of the MLL machine, but the 2004 paper it says this came from never mentions UTP18 at all. Line by line, the database's list of that paper's proteins is the paper's own list with WDR5 - a protein this project already has an entry for - replaced by UTP18.

The same thing, for a scientist

The 2026-09-21 interactome recount named **17 bait-supported IntAct partners of the six core proteins with no entry in this graph**, and exactly one of them was fished by **KMT2A itself**: UTP18 (Q9Y5J1), from **PMID 15199122** - Yokoyama et al. 2004, the founding paper for this graph's WRAD and menin-complex position, cited in 29 entries here [E1][E5]. That single row made KMT2A look like it had an unwritten directed partner. **It does not.** The full text of PMID 15199122 contains no occurrence of `UTP18`, `WDR50`, `CGI-48`, `U3`, `snoRNA`, `nucleolar` or `nucleolus` [E2]. The paper states that its purification resolved **eight polypeptides**, identified by mass spectrometry as MLL(N), MLL(C), **HCF-2, ASH2L1 and -2, menin, RBBP5 and WDR5** [E2]. IntAct's curation of the same paper contains **eight participants - KMT2A, HCFC1, RBBP5, ASH2L (twice), MEN1, HCFC2 and UTP18 - and no WDR5 at all** [E3]. The substitution reproduces in the second experiment curated from that paper: the Fig. 2C gel-filtration panel blots MLL(C), ASH2L, **WDR5**, RBBP5, HCF-2 and menin, and IntAct's `molecular sieving` experiment EBI-592758 lists KMT2A, ASH2L (twice), **UTP18**, RBBP5, HCFC2 and MEN1 [E2][E3]. Independently, **IntAct holds no WDR5-KMT2A row from PMID 15199122 whatsoever** - its three WDR5-KMT2A rows are Tang 2013, Hauri 2016 and BioPlex 3.0 [E4]. UTP18's UniProt synonym is **WDR50** and its recommended name is *WD repeat-containing protein 50*; WDR5 is *WD repeat-containing protein 5* [E6]. **The conclusion is that UTP18 should not be written as a KMT2A interactor, and that WDR5 - which already has an entry - is the protein the paper reported.** The corrected KMT2A-bait gap is zero, and the recount's clean denominator of 54 partners with 17 unwritten should read **16 unwritten**.

Go deeper
The arithmetic, which is what makes this a finding rather than a suspicion

Absence of a symbol from a full text is suggestive. The decisive part is that the two curated participant lists reconstruct the paper's two figures exactly, with one name substituted in each.

paperIntAct
Fig. 1B/1C, anti-MLL(N) immunopurification, 8 bandsMLL(N), MLL(C), HCF-2, ASH2L1, ASH2L2, menin, RBBP5, WDR5KMT2A, HCFC1, HCFC2, ASH2L, ASH2L, MEN1, RBBP5, UTP18 (EBI-592993, anti-bait coimmunoprecipitation, 8 participants)
Fig. 2C, Superose 6 gel filtration, 6 antibodiesMLL(C), ASH2L, WDR5, RBBP5, HCF-2, meninKMT2A, ASH2L, ASH2L, UTP18, RBBP5, HCFC2, MEN1 (EBI-592758, molecular sieving)

The MLL(N)/MLL(C) pair collapses to one KMT2A accession and HCF-1 - detected by western in Fig. 2A - accounts for HCFC1. Every other slot matches. UTP18 occupies WDR5's slot in both experiments and WDR5 appears in neither.

Both IntAct experiments carry psi-mi:"MI:1060"(spoke expansion), meaning these are not measured binary contacts at all but IntAct's binarisation of a complex purification against the bait. So the row is a spoke-expanded inference about a protein that is not in the paper.

Why WDR5 and WDR50 are the pair that would collide

UniProt Q9Y5J1's primary symbol is UTP18, its only listed gene synonym is WDR50, and its recommended name is U3 small nucleolar RNA-associated protein 18 homolog with the alternative name WD repeat-containing protein 50. WDR5 is P61964, WD repeat-containing protein 5.

The curation was deposited 2005-06-03, when WDR50 was a current alias and before the UTP18 nomenclature settled. Yokoyama et al. write "WDR5 (also called BIG-3)", so the paper itself offers two strings for the same protein. Both proteins are WD40 beta-propellers; neither the family, the name, nor the domain content separates them by eye.

This is a new member of the class the brief's rules 62-69 describe, with one difference worth recording: it is not a lookup that returned the wrong record, it is a curated row that has been wrong in a public database for twenty-one years. No accession check catches it, because the accession IntAct stores is a correct accession for UTP18. Only reading the source paper catches it, which is exactly what brief rule 91 asks for and what no automated interactome recount can do.

What this costs and what it corrects

It is the reason KMT2A appeared in the gap list at all. analyses/directed-interactome-coverage-recheck/ reports KMT2A fishing exactly one unwritten bait-supported partner, and that partner is UTP18. With the row removed, KMT2A's directed interactome has no unwritten partner, and the coverage line becomes 38/53 rather than 37/54 on the same generous join.

It also hid a partner that is already written. gene-wdr5 has existed in this graph since the first interactome beat. The correct row - WDR5 as a KMT2A bait-supported partner from the founding paper - would have joined immediately. So the defect both invented a gap and suppressed a hit, and the net effect on the coverage percentage is small while the effect on which protein a reader thinks is in the MLL complex is total.

The general lesson is narrower than "databases have errors". This project already knows a curated interaction is not a measurement. What it did not have written down is that a curated interaction can be about a protein that does not appear in the cited paper, and that the cheapest detector is to count the participants and match them against the paper's own figure legend. Two figures, eight names, ten minutes.

What I am NOT claiming

I have not contacted IntAct, and this entry is an inference from three observations - the symbol's absence from the full text, the positional match of the participant lists, and the absence of any WDR5 row from that publication - not a statement from the curators.

The alternative hypothesis is that UTP18 was genuinely identified in the Fig. 1C large-scale Coomassie LC-MS/MS and reported only in a figure annotation that the PMC HTML rendering does not carry as text. That hypothesis has to also explain where WDR5 went, since WDR5 is named in the running text as one of the eight and is absent from IntAct's curation of the paper. It does not, so I record the substitution as the explanation and the alternative as the caveat.

I also do not claim UTP18 is a contaminant of MLL purifications in general. It is a real nucleolar protein with 78 distinct partners across 25 publications in IntAct, including 38 from BioPlex 3.0. None of them is KMT2A, MEN1, AFF1, AFF4, PSIP1 or DOT1L. UTP18's absence from this graph is correct.

Facts and figures
scriptnone - manual adjudication, queries recorded in evidence URLs
statusrun
result summaryUTP18 (Q9Y5J1) is not a KMT2A interactor. IntAct's two curated experiments from PMID 15199122 reproduce the paper's Fig. 1 and Fig. 2C participant lists with UTP18 substituted for WDR5, and hold no WDR5 row from that paper. KMT2A's unwritten-bait-supported-partner count is therefore 0, not 1, and the recount's 17 should read 16.
verdictno entry written for UTP18 AS A KMT2A PARTNER ON THE STRENGTH OF THAT INTACT ROW - that particular partner record does not exist. SCOPED 2026-09-21: this verdict is about the IntAct row from PMID 15199122 and nothing else. UTP18 the protein is real, and Skucha 2018 recovers it in 3 of 7 MLL-fusion baits; gene-utp18 now exists as a separate, independently sourced entry. Do not quote this sentence as a claim about UTP18 in general
correction owed outside my file setanalyses/directed-interactome-coverage-recheck/results.json and docs/coverage.md record 17 unwritten bait-supported partners including UTP18 under bait KMT2A. Both should record 16 and note the KMT2A row as a database curation error. scripts/check-pi-dimensions.py quotes the cached 37/54; that becomes 38/53 if WDR5 is credited, or 37/53 if the row is simply dropped.
utp18 intact profile 2026 09 21
core_baits_other_than_the…0.000core_baits_other_than_the_disputed_kmt2a_row: 0distinct_publications25.0distinct_publications: 25rows_from_bioplex_3_038.0rows_from_bioplex_3_0: 38distinct_partners78.0distinct_partners: 78total_rows91.0total_rows: 91
inputs
[
  "IntAct PSICQUIC MITAB 2.7, queries pubid:15199122, id:Q9Y5J1 AND id:Q03164, id:P61964 AND id:Q03164, retrieved 2026-09-21",
  "PMC HTML full text of PMC480881, 68,159 characters after tag-strip then unescape",
  "UniProt REST gene_exact lookups for UTP18 and WDR5, 2026-09-21",
  "analyses/directed-interactome-coverage-recheck/results.json"
]
intact rows adjudicated
[
  {
    "interaction_ac": "intact:EBI-592993",
    "method": "psi-mi:\"MI:0006\"(anti bait coimmunoprecipitation)",
    "expansion": "psi-mi:\"MI:1060\"(spoke expansion)",
    "host": "taxid:9606 Homo sapiens K562 myeloid leukemia cell",
    "participants_in_intact": [
      "KMT2A",
      "HCFC1",
      "UTP18",
      "RBBP5",
      "ASH2L",
      "ASH2L",
      "MEN1",
      "HCFC2"
    ],
    "participants_in_the_paper": [
      "MLL(N)",
      "MLL(C)",
      "HCF-2",
      "ASH2L1",
      "ASH2L2",
      "menin",
      "RBBP5",
      "WDR5"
    ],
    "created": "2005-06-03",
    "updated": "2026-01-11"
  },
  {
    "interaction_ac": "intact:EBI-592758",
    "method": "psi-mi:\"MI:0071\"(molecular sieving)",
    "expansion": "psi-mi:\"MI:1060\"(spoke expansion)",
    "participants_in_intact": [
      "HCFC2",
      "KMT2A",
      "MEN1",
      "ASH2L",
      "ASH2L",
      "RBBP5",
      "UTP18"
    ],
    "antibodies_blotted_in_paper_fig_2c": [
      "MLL(C)",
      "ASH2L",
      "WDR5",
      "RBBP5",
      "HCF-2",
      "menin"
    ]
  }
]
accessions
{
  "UTP18": {
    "accession": "Q9Y5J1",
    "length_aa": 556,
    "gene_synonym": "WDR50"
  },
  "WDR5": {
    "accession": "P61964",
    "graph_entry": "gene-wdr5"
  }
}
utp18 intact profile 2026 09 21 — the numbers behind the chart
{
  "total_rows": 91,
  "distinct_partners": 78,
  "distinct_publications": 25,
  "rows_from_bioplex_3_0": 38,
  "core_baits_other_than_the_disputed_kmt2a_row": 0
}
counted absence
{
  "query": "(UTP18[tiab] OR WDR50[tiab]) AND (MLL[tiab] OR KMT2A[tiab] OR leukemia[tiab])",
  "records": 0,
  "querytranslation_note": "PubMed silently dropped the WDR50 operand from the translation because it matches nothing; the executed query was \"UTP18\"[Title/Abstract] AND (\"MLL\"[Title/Abstract] OR \"KMT2A\"[Title/Abstract] OR \"leukemia\"[Title/Abstract]). Operand control: UTP18[tiab] alone returns 17.",
  "date": "2026-09-21",
  "bound": "zero in titles and abstracts; supporting figures not searched"
}
How this could bite us back
  • A reader who takes this entry as licence to distrust IntAct generally would be wrong, and would lose a resource this project depends on. The finding is one row, argued from one paper, and the same curation's HCFC2 row is correct.
  • If the alternative hypothesis is right - UTP18 really was identified in the Fig. 1C mass spectrometry and reported only in a figure graphic - then declining to write the entry leaves a genuine KMT2A partner unrecorded. The way to close that is to obtain the original figure or ask the authors, not to write the entry on the strength of a database row.
Still open
  1. Should this be reported to IntAct? The row has been live since 2005 and propagates to every downstream resource that mirrors IntAct, including the recount that dispatched this beat. A one-paragraph curator note is the highest-leverage output of this entry and nobody in this project has ever filed one.
  2. How many other rows in this graph's interactome evidence base name a protein that does not appear in the cited paper? The detector is cheap - count the curated participants and match them against the paper's figure legend - and it has never been run on the other load-bearing publications.
  3. Does the same substitution appear in BioGRID, STRING or Reactome's import of PMID 15199122? If it does, the error has a single upstream source and one correction fixes all of them; if it does not, IntAct's row is isolated and the case is even stronger.
Every source behind this entry (6)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    IntAct records a KMT2A-UTP18 association attributed to PMID 15199122, curated as anti-bait coimmunoprecipitation in K562 with KMT2A as bait, spoke-expanded, interaction EBI-592993; this is the only unwritten bait-supported partner attributed to KMT2A in the 2026-09-21 interactome recount.

    What this evidence cannot show

    This evidence item asserts only what the database says, which is not in dispute. The rest of the entry argues that what it says is wrong. The MITAB expansion column reads psi-mi:"MI:1060"(spoke expansion), so even taken at face value the row is not a measured binary contact.

    Evidence that argues against this (1)
    • IntAct is a curated, IMEx-standard resource whose rows are read back by the entire field, and the prior on any individual row being wrong is low. That prior is the reason this took twenty-one years to notice and the reason the claim here is argued from the paper rather than asserted. Yokoyama A et al. 2004
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    The full text of PMID 15199122 contains no occurrence of UTP18, WDR50, CGI-48, U3, snoRNA, nucleolar or nucleolus, and states that the eight polypeptides resolved from the MLL purification were MLL(N), MLL(C), HCF-2, ASH2L1 and -2, menin, RBBP5 and WDR5.

    Leukemia proto-oncoprotein MLL forms a SET1-like histone methyltransferase complex with menin to regulate Hox gene expression. · Yokoyama A, Wang Z, Wysocka J, Sanyal M, Aufiero DJ, Kitabayashi I, Herr W, Cleary ML · Mol Cell Biol 2004
    PMID 15199122DOI
    What this evidence cannot show

    Read through the PMC HTML route (PMC480881, 68,159 characters after tag-stripping then entity-unescaping) on 2026-09-21, because Europe PMC fullTextXML returns HTTP 500 and efetch db=pmc returns a document with no body element for this paper - the route documented in the standing brief. An HTML rendering does not carry text baked into figure images, so a protein named only inside a figure graphic would be invisible to this search. That is the alternative hypothesis this entry records and rejects on the grounds that it cannot account for WDR5's absence from the curation.

    Evidence that argues against this (1)
    • A term-absence search over one rendering of one paper is bounded by that rendering. Supplementary material was not retrieved: PMID 15199122 is a 2004 MCB paper and no supplementary protein table was reached, so the bound is 'absent from the article body as served by PMC HTML', not 'absent from the study'. Huttlin EL et al. 2017
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    IntAct's curation of PMID 15199122 comprises 13 MITAB rows across two experiments and eight distinct participants - KMT2A, MEN1, RBBP5, ASH2L, HCFC1, HCFC2 and UTP18 - with no WDR5 participant in either experiment.

    What this evidence cannot show

    PSICQUIC returns IntAct's holdings only. Another IMEx partner could in principle hold a WDR5 row curated from this paper that IntAct does not mirror; that would not affect the argument, which is about what IntAct's own participant list for this publication contains. The creation date on both experiments is 2005-06-03 and the last-update date is 2026-01-11, so the row is live and has survived a recent re-release.

    Evidence that argues against this (1)
    • This graph's gene-hcfc2 entry was written from the same curation and is correct - HCF-2 genuinely is in the paper. So the curation of PMID 15199122 is not globally unreliable, and a reader should not use this entry to discount the other rows from it. Oncofusion knowledge graph, gene-hcfc2 2026
  4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    IntAct holds three WDR5-KMT2A interaction rows and none of them is attributed to PMID 15199122; they come from Tang et al. 2013 (anti-bait coimmunoprecipitation), Hauri et al. 2016 (tandem affinity purification) and Huttlin et al. 2021 (anti-tag coimmunoprecipitation).

    What this evidence cannot show

    The absence of the founding paper from WDR5's own IntAct record is the load-bearing half of this item and it is a database absence, not a literature absence. WDR5's membership of the MLL complex is not in doubt on the strength of the paper; only its presence in IntAct's curation of that paper is.

    Evidence that argues against this (1)
    • A curated database omitting a well-known interaction from a specific paper is common and by itself unremarkable - IntAct does not curate every figure of every paper. It becomes evidential here only in conjunction with UTP18 occupying the exact slot WDR5 would have filled in both experiments. Huttlin EL et al. 2021
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis

    The 2026-09-21 directed-interactome recount reports 54 human bait-supported partners of the six core proteins with 17 lacking an entry, and names UTP18 as the only such partner fished by KMT2A.

    The interactome dimension is reopened, and the number that closed it was never re-derivable · Oncofusion knowledge graph, docs/coverage.md and analyses/directed-interactome-coverage-recheck/ 2026
    What this evidence cannot show

    An internal analysis of this project, not an external source. Its own caveats apply, including that the join is deliberately generous and that the 54-partner denominator is a snapshot of a live database. This entry does not dispute the recount's method - the recount is correct about what IntAct says; it is IntAct that is wrong.

    Evidence that argues against this (1)
    • Correcting 17 to 16 does not rescue the cycle-34 conclusion that closed the interactome beat for seven cycles. Sixteen unwritten bait-supported partners is still not zero, and the reproducibility defect rule 90 describes - the count written without its members - is untouched by this correction. Rolland T et al. 2014
  6. E6established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    UTP18 is UniProt Q9Y5J1, 556 aa, primary gene name UTP18, sole gene synonym WDR50, recommended name U3 small nucleolar RNA-associated protein 18 homolog with alternative name WD repeat-containing protein 50; WDR5 is P61964, WD repeat-containing protein 5.

    What this evidence cannot show

    Resolved on 2026-09-21 with gene_exact:UTP18 AND organism_id:9606 AND reviewed:true, fields accession,id,gene_primary,gene_synonym,protein_name,length; exactly one reviewed record returned and the primary gene name was read back. gene_exact matches synonyms, so the primary-name read-back is the check that matters here.

    Evidence that argues against this (1)
    • Nomenclature in 2005, when this row was curated, differed from today's: WDR50 was then a usable symbol for Q9Y5J1. A reader should not read the substitution as carelessness; the two names were closer in 2005 than the current symbols make them look. Ewing RM et al. 2007
Connected to