Oncofusion

start / graph / concept-mll2-and-mll4-each-name-two-genes

Explainerstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q44-kmt2-family

MLL2 and MLL4 each name two different genes, and the ambiguity is in the papers, not in the databases

also called MLL2 ambiguity · MLL4 ambiguity · KMT2B versus KMT2D naming · rule 65 for the KMT2 family

Two genes in this family, KMT2B and KMT2D, have both been called MLL2 and both been called MLL4, by different laboratories in different decades. You cannot tell them apart from the name. You can tell them apart from their size: one is 2,715 amino acids and the other is 5,537.

The same thing, for a scientist

Rules 62, 63 and 65 in this project's standing brief describe query tools returning the wrong record. **This case is worse: the source document is ambiguous.** UniProt records fetched 2026-09-20 show KMT2B (Q9UMN6, 2,715 aa) carrying the synonyms MLL2 *and* MLL4 with the protein name 'Myeloid/lymphoid or mixed-lineage leukemia protein **4**', and KMT2D (O14686, 5,537 aa) carrying the synonyms MLL2 *and* MLL4 with the protein name 'protein **2**' - the names swapped relative to the aliases [E1]. PubMed's own query translation maps the string `MLL2` to the KMT2D supplementary concept and `MLL4` to the KMT2B one [E2], which is the current convention and the *opposite* of how several foundational papers used the terms. Reading the protein name back - the defence rule 65 prescribes - returns something that looks right for either gene. **Only the length, the accession, or a phenotype separates them.** This entry records every instance this beat met and how each was decided, including the two it could not decide.

Go deeper
The resolution table

Every MLL2 or MLL4 this beat encountered, with the method used and the outcome. The method column is the point of the table.

PMIDwritten asresolves tohow decided
23508102MLL2KMT2Bthe paper's own accession table lists E9PQG7 / Q9UMN6 MLL1/2 and O14686-3 / Q8NEZ4-3 MLL3/4 3
28609655MLL2KMT2Bthe paper calls it MLL1's closest paralog; KMT2D at 5,537 aa is not 4
30315824MLL2KMT2Bsame authors, same mouse alleles as 28609655
36635503MLL2KMT2Bstated in the text: "Menin is unique to KMT2A (MLL1) or KMT2B (MLL2) containing complexes" 5
34440566MLL2KMT2Bthe title reads MLL2 (KMT2B) and the review states 2,715 aa
17021013MLL2KMT2Bgrouping: the paper contrasts menin-containing MLL1/MLL2 against ASC-2-containing MLL3/MLL4 6
19060922Mll2KMT2B"the closest relative of MLL"
27992417MLL4KMT2B"KMT2B (also known as MLL4)" in the abstract
20711175MLL2KMT2Dphenotype: Kabuki syndrome type 1 is KMT2D; KMT2B's germline phenotype is dystonia 7
30028819MLL2KMT2Dstated only in the full-text discussion as "MLL2 (KMT2D)"; the title and abstract are undecidable 8
31128216MLL2(4)KMT2Dthe review prints both aliases beside the KMT2D symbol
14992727MLL2UNRESOLVEDfounding menin paper; no accession in the abstract, no PMCID, full text not obtained. Left open.
(Strick 2006, in genome-aff1-breakpoint-region-architecture)MLL2UNRESOLVEDalready flagged by that entry; this beat did not resolve it either

Eight resolve to KMT2B and three to KMT2D, which is itself the finding: in the leukaemia literature that matters to this project, MLL2 usually means KMT2B, while in the developmental-syndrome and pan-cancer literature it usually means KMT2D. The two literatures barely cite each other, which is why the collision has survived.

The four methods, ranked
  1. Accession in the paper. Decisive and rare. Only one paper in the table above supplied one, in a supplementary-style table of quantified proteins 3.
  2. Length stated in the text. Decisive when present: 2,715 against 5,537 cannot be confused. A review title that also states the length is the ideal case 1.
  3. Phenotype. Decisive for germline work: Kabuki syndrome type 1 is KMT2D, early-onset dystonia is KMT2B 7. Useless for somatic or biochemical work.
  4. Grouping and relatedness. Weaker but usually sufficient: a paper that writes 'MLL1/MLL2' as the menin branch and 'MLL3/MLL4' as the other branch has told you MLL2 is KMT2B, because KMT2D is one of the MLL3/4 pair. A paper that calls it 'the closest paralogue of MLL1' has told you the same thing by size 46.

What does not work: reading the protein name back, trusting the database synonym list, or trusting PubMed's supplementary-concept mapping to reflect what an older paper meant. All three return a confident answer of the right shape.

When none works, record it unresolved. Two cases in the table are left open, and that is the correct outcome. An entry that says 'this paper's MLL2 could not be resolved' is more useful than one that guessed right, because the next reader knows which claims to re-check.

The machine-checkable defence

Assert both the accession and the length in data on every entry touching this family. verify-accessions.py fetches the accession and compares the length, so a KMT2B/KMT2D swap - which is a 2,822-residue error - fails the check immediately, while a swap that only changed the symbol would pass every other test in this project.

This beat's seven gene entries each carry uniprot, length_aa, gene_primary_verified and an accession_resolution string describing the query and why exactly one row was kept. That is the machinery rule 65 asks for, applied to the family where it is most needed.

One additional near-collision, recorded because it is the kind of thing that produces an eleventh identifier incident: NCOA6 is Q14686 and KMT2D is O14686, one character apart, and the two proteins are subunits of the same MLL3/MLL4 complex, so they co-occur in exactly the sentences where a transposition would look unremarkable. Lengths 2,063 and 5,537 separate them 1.

Facts and figures
kindnaming hazard
adjacent accession hazardNCOA6 Q14686 versus KMT2D O14686 - one character apart, same complex, lengths 2063 and 5537
machine checkassert both uniprot and length_aa in data on every family entry so verify-accessions.py can catch a 2,822-residue swap
colliding symbols
[
  "MLL2",
  "MLL4"
]
genes
[
  {
    "symbol": "KMT2B",
    "uniprot": "Q9UMN6",
    "length_aa": 2715,
    "protein_name_contains": "mixed-lineage leukemia protein 4",
    "germline_phenotype": "early-onset dystonia"
  },
  {
    "symbol": "KMT2D",
    "uniprot": "O14686",
    "length_aa": 5537,
    "protein_name_contains": "mixed-lineage leukemia protein 2",
    "germline_phenotype": "Kabuki syndrome type 1"
  }
]
resolution methods ranked
[
  "accession stated in the paper",
  "protein length stated in the paper",
  "germline phenotype (Kabuki vs dystonia) - germline studies only",
  "the paper's own family grouping or a 'closest paralogue' statement"
]
methods that fail
[
  "reading the protein name back (both contain a colliding number)",
  "the database synonym list (both genes carry both aliases)",
  "PubMed supplementary-concept mapping (current convention applied to older usage)"
]
resolved instances
{
  "to_KMT2B": [
    "23508102",
    "28609655",
    "30315824",
    "36635503",
    "34440566",
    "17021013",
    "19060922",
    "27992417"
  ],
  "to_KMT2D": [
    "20711175",
    "30028819",
    "31128216"
  ],
  "unresolved": [
    "14992727",
    "Strick 2006 comparator cited in genome-aff1-breakpoint-region-architecture"
  ]
}
How this could bite us back
  • A silent swap here does not produce an obviously wrong protein - it produces the other member of the same family in the same pathway, which is the failure mode rule 63 was written for and which nothing about the result looks wrong under.
  • The two genes point in different directions therapeutically. KMT2B is a menin-inhibitor off-target in normal tissue; KMT2D is a tumour suppressor whose loss is a proposed resistance mechanism. Confusing them turns a toxicity question into an efficacy question.
  • Length is the only robust check and it only works if someone asserts it. An entry that stores a symbol and a claim, with no accession and no length, is unfalsifiable by any script this project runs.
Still open
  1. What does Hughes 2004's MLL2 denote? It is the founding menin-complex paper and it is cited across this field. No PMCID, no accession in the abstract; one route-5 DOI fetch at Elsevier would probably settle it and this beat did not attempt it.
  2. How many entries in this graph carry an MLL2 or MLL4 whose gene was never determined? The token scan found 85 entries naming a non-KMT2A family member, but it did not separate resolved from unresolved usages. That separation is a half-day of reading and would convert a hazard into a list.
  3. Does any automated extraction in this project resolve MLL2? If a script or an LLM pass ever maps the string MLL2 to a gene, it is making this decision silently, and nothing currently records which way.
Every source behind this entry (8)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    UniProt reviewed human records fetched 2026-09-20: KMT2B is Q9UMN6, 2,715 aa, synonyms HRX2/KIAA0304/MLL2/MLL4/TRX2/WBP7, protein name 'Myeloid/lymphoid or mixed-lineage leukemia protein 4'; KMT2D is O14686, 5,537 aa, synonyms ALR/MLL2/MLL4, protein name 'Myeloid/lymphoid or mixed-lineage leukemia protein 2'; KMT2C is Q8NEZ4, 4,911 aa, synonym MLL3 only; NCOA6 is Q14686, 2,063 aa - one character from KMT2D's accession and a subunit of the same complex.

    What this evidence cannot show

    Each gene_exact query returned exactly one reviewed human row whose gene_primary matched the queried symbol. Synonym lists are curatorial and change; the lengths do not. This record describes the database as of one day and says nothing about what any given paper meant.

    Evidence that argues against this (1)
    • A database snapshot cannot arbitrate historical usage. UniProt's current protein names assign 'protein 4' to KMT2B and 'protein 2' to KMT2D, which is the opposite of how the Chen, Slany and van Nuland leukaemia papers use MLL2 - so following the database would mis-resolve the very literature this project depends on. Chen Y et al. 2017
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    PubMed's querytranslation, inspected 2026-09-20, expands the string MLL2 to "kmt2d protein human"[Supplementary Concept] and the string MLL4 to "kmt2b protein human"[Supplementary Concept], so a query written as (KMT2B OR MLL2 OR MLL4) retrieves a mixture of KMT2B and KMT2D records and its count cannot be attributed to either gene.

    What this evidence cannot show

    Supplementary-concept mappings are NLM curatorial decisions and can change without notice. The mapping is applied to the string at query time and is not applied retroactively to how an indexed paper used the term, so a 2009 paper meaning KMT2B by 'MLL2' is retrievable under the KMT2D concept.

    Evidence that argues against this (1)
    • The mapping is not wrong - it follows the current HGNC-aligned convention in which MLL2 is a KMT2D alias. The failure is that the convention post-dates much of the literature, so a correct mapping retrieves papers that meant the other gene. Klonou A et al. 2021
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    The quantitative SET1/MLL complex survey lists its baits against UniProt accessions, giving 'E9PQG7 / Q9UMN6 MLL1/2' and 'O14686-3 / Q8NEZ4-3 MLL3/4' - so in that paper MLL2 is Q9UMN6 (KMT2B), MLL3 is Q8NEZ4 (KMT2C) and MLL4 is O14686 (KMT2D), decided from accessions rather than names.

    Quantitative dissection and stoichiometry determination of the human SET1/MLL histone methyltransferase complexes. · van Nuland R, Smits AH, Pallaki P, Jansen PW, Vermeulen M, Timmers HT · Mol Cell Biol 2013
    PMID 23508102DOI
    What this evidence cannot show

    Full text read via the PMC HTML route (PMC3647974) on 2026-09-20; the accession string was read from the flattened HTML of an abundance table, so the pairing of accession to label depends on that table's layout surviving tag-stripping. The accessions themselves are unambiguous and were cross-checked against the UniProt records fetched the same day.

    Evidence that argues against this (1)
    • Reading an accession out of a tag-stripped HTML table is exactly the kind of extraction this project's brief warns can silently lose or reorder content. The conclusion is corroborated independently by the paper grouping MLL1/2 as the menin-containing complexes, which only works if MLL2 is KMT2B. Lee S et al. 2006
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The paper that established the wild-type paralogue dependency in MLL-rearranged leukaemia identifies its subject as 'the closest paralog' of Mll1 and calls it Mll2 - which is KMT2B, since KMT2D at 5,537 aa is not MLL1's closest paralogue.

    MLL2, Not MLL1, Plays a Major Role in Sustaining MLL-Rearranged Acute Myeloid Leukemia. · Chen Y, Anastassiadis K, Kranz A, et al. · Cancer Cell 2017
    PMID 28609655DOI
    What this evidence cannot show

    Abstract fetched and read 2026-09-20; resolution is by the phrase 'closest paralog' plus mouse orthology, not by an accession in the paper. Full text (PMC5598468) not obtained, and the methods would name the floxed allele, which would settle it outright. This is a strong inference, not a citation of a statement.

    Evidence that argues against this (1)
    • Resolution by relatedness is the weakest of the four methods and this is the paper on which ten entries in this graph rest a correction. It would be worth one full-text fetch to read the allele name and convert a strong inference into a stated fact. Mendoza-Castrejon J et al. 2026
  5. E5established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    The Menin-KMT2A/B bivalency study states in its introduction that 'Menin (MEN1) is unique to KMT2A (MLL1) or KMT2B (MLL2) containing complexes', supplying both mappings explicitly in one sentence.

    Targeting Menin disrupts the KMT2A/B and polycomb balance to paradoxically activate bivalent genes. · Sparbier CE, Gillespie A, Gomez J, et al. · Nat Cell Biol 2023
    PMID 36635503DOI
    What this evidence cannot show

    Full text read via Europe PMC fullTextXML (PMC7614190) on 2026-09-20. A statement in one paper's introduction is a statement about that paper's usage; it does not bind any other paper's.

    Evidence that argues against this (1)
    • The same paper's methods list its antibody as 'anti-MLL2/KMT2B', so even within one document both names are in use - which is the practice that keeps the collision alive rather than resolving it. Klonou A et al. 2021
  6. E6moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    A 2006 specificity study groups the family as menin-containing MLL1 and MLL2 complexes against ASC-2-containing MLL3 and MLL4 complexes, and reports that RARbeta2 expression is intact in MEFs devoid of menin - so its MLL2 is the menin-branch paralogue, KMT2B, resolved by the grouping the paper itself uses.

    Coactivator as a target gene specificity determinant for histone H3 lysine 4 methyltransferases. · Lee S, Lee DK, Dou Y, et al. · Proc Natl Acad Sci U S A 2006
    PMID 17021013DOI
    What this evidence cannot show

    Abstract fetched and read 2026-09-20; full text (PMC1622834) not obtained. Resolution is by grouping, method 4 of four and the weakest. The paper gives no accession and no length. Graded moderate for that reason.

    Evidence that argues against this (1)
    • Resolution by grouping assumes the paper's MLL3/MLL4 pair is KMT2C/KMT2D, which is itself a convention rather than a statement. It is a safe assumption because MLL3 is unambiguous, but the chain has two links and neither is an accession. Fagan RJ et al. 2019
  7. E7established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Exome sequencing identified MLL2 mutations as a major cause of Kabuki syndrome; that MLL2 is KMT2D, resolved by phenotype, because KMT2D causes Kabuki syndrome type 1 while heterozygous KMT2B variants cause complex early-onset dystonia in 27 unrelated individuals.

    Exome sequencing identifies MLL2 mutations as a cause of Kabuki syndrome. · Ng SB, Bigham AW, Buckingham KJ, et al. · Nat Genet 2010
    PMID 20711175DOI
    What this evidence cannot show

    Abstract fetched and read 2026-09-20. Phenotype-based resolution works only for germline studies and only where the two candidate genes have distinguishable syndromes, which here they do. Kabuki syndrome type 2 is KDM6A, a different gene in the same complex, so the phenotype identifies the complex more tightly than it identifies the gene - it is sufficient to separate KMT2D from KMT2B and not from KDM6A.

    Evidence that argues against this (1)
    • The dystonia paper writes 'KMT2B (also known as MLL4)', using the alias this beat elsewhere resolves to KMT2B - so the germline literature contains both aliases pointing at both genes, and only the syndrome, not the symbol, separates them. Meyer E et al. 2017
  8. E8moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    A paediatric B-ALL clinical-sequencing case report titled around an 'MLL2 truncation' is undecidable from its title and abstract and resolves only in its full-text discussion, which writes 'MLL2 (KMT2D)'.

    PMID 30028819DOI
    What this evidence cannot show

    Full text read via efetch db=pmc (PMC6339603) on 2026-09-20, 20,274 stripped characters. A single case report; cited here for the naming lesson only. **The clinical significance of this instance is the point**: a candidate resistance driver was reported in a child under a name that identifies two genes, and only a reader who reached the discussion could know which.

    Evidence that argues against this (1)
    • One case report is not evidence that the ambiguity causes clinical harm, and the authors did in fact disambiguate, just not where a database curator or an abstract-only reader would see it. The systematic risk is in automated extraction rather than in careful reading. Larson JK et al. 2023
Connected to