Oncofusion

start / graph / ana-kmt2a-deferred-partner-enumeration

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q-partners-deferred

The deferred KMT2A partners: what the counts actually are, what was written, what was rejected, and three numbers the previous enumeration got wrong

also called deferred partner enumeration · partners-deferred rejections

Before writing anything, this beat counted how much real evidence exists for each KMT2A fusion partner that an earlier round had put off. Nine entries were written, several partners were rejected with the numbers behind each rejection, and three counts in the earlier list turned out to be wrong.

The same thing, for a scientist

Systematic enumeration behind the partners-deferred beat, with two independent denominators for every partner: patient counts read directly from Table 1 and the AML/ALL prose of the 2023 DCAL recombinome full text (PMC10169636, n=3401), and PubMed [Title/Abstract] and [All Fields] record counts from raw E-utilities esearch, **every result set title-checked** [E1][E2]. Nine entries written: gene-gphn, fusion-kmt2a-gphn, gene-mllt11, fusion-kmt2a-mllt11, fusion-kmt2a-eps15, gene-gas7, fusion-kmt2a-gas7, fusion-kmt2a-mllt6, gene-myo1f, fusion-kmt2a-myo1f. Two gene entries deliberately NOT written because equivalents already exist (`protein-eps15-domains`, `gene-mllt6`). **The single most important result is a counted zero**: no functional experiment has ever been performed on KMT2A::MLLT11, the 10th most common KMT2A fusion — the fusion-name query returns 10 records, all descriptive, and intersecting it with mice/mouse/retroviral/colony/replating returns 0 [E2]. Three corrections to the inherited enumeration are recorded with sources [E3][E4]. The largest deliberate gap is **USP2**: rank 9 with 30 patients, 11 records of which about 8 are directly on the fusion, and a 2019 Leukemia paper reporting a **second breakpoint cluster region in KMT2A** for recurrent MLL-USP2 fusions — a structural claim about KMT2A itself that this graph does not carry [E2].

Go deeper
Method, and why both denominators were re-derived rather than inherited

Every patient count in this beat was read from the Europe PMC full-text XML of PMC10169636 directly, with the Table 1 column structure verified from the table header (# | direct TPG | Infant(ALL,AML,other) | Pediatric(ALL,AML,other) | Adult(ALL,AML,other) | not-age-classified(ALL,AML,other) | Total). The previous enumeration's numbers were not taken on trust, and the re-derivation is what surfaced the corrections below 1.

PubMed counts were raw E-utilities esearch, not the MCP search tool, which ANDs every keyword and returns zero on topics with large literatures. Every result set that drove a decision was title-checked, per the project's standing rule that short or word-like aliases inflate counts while retired symbols deflate them. Two things follow for this beat specifically:

  • The retired-symbol trap applies and was handled. AF1q returns 58 records against MLLT11's 45, and the KMT2A intersection is dominated by AF1q; AF1p returns 8 against EPS15's 609; AF17 dominates the MLLT6 intersection. All searches were run on both symbols.
  • The inflation trap mostly does not bite here, and where it does it was measured. AF1q, AF1p and AF17 are not common words. The one genuinely inflated count is MAML2: the raw intersection returns 20 records, of which title-checking leaves roughly 12-13 about KMT2A::MAML2 and the remainder about MAML2 in mucoepidermoid carcinoma and thymoma without a KMT2A fusion. The checked number is reported below, not the raw one.
  • A new trap, not previously recorded: typographic variants. The recombinome's AML prose prints MYO1F as "5x MYOF1", which is not an HGNC symbol and collides with the real, unrelated gene MYOF (myoferlin). A symbol-keyed search over that sentence scores the mention wrongly. This is neither the retired-symbol nor the short-alias failure mode; it is a third one.
Written as entries
partnerrecombinome n (rank)distributionPubMed intersection (checked)why written
EPS1555 (8)22 infant ALL, 2 infant AML, 3 infant other, 8 paed ALL, 5 paed AML, 4 adult ALL, 7 adult AML, 1 adult other, 3 NA15biggest patient count in the beat; infant-ALL-weighted; one of the two coiled-coil pillars
MLLT1128 (10)1 infant ALL, 13 infant AML, 10 paed AML, 4 adult AML20rank 10 and zero functional literature; the sharpest gap in the landscape
MLLT627 (11)5 infant ALL, 3 paed ALL, 5 paed AML, 12 adult AML, 1 NA AML, 1 NA other31the bridge case: leucine zipper AND DotCom subunit
MYO1F5 (22)4 infant AML, 1 paed AML2recurrent in infant AML; partner explicitly lacking both transactivation and dimerisation domains
GAS73 (28)1 infant AML, 1 paed ALL, 1 NA other5the second coiled-coil pillar; best mouse model in the class
GPHN0 (not in the frequency table)2 published cases: 1 de novo AMoL, 1 therapy-related AUL3the 15-residue counterexample the coordinator cited five times with no entry behind it

Gene entries were deliberately not written for EPS15 or MLLT6: protein-eps15-domains and gene-mllt6 already exist and carry the protein-level content, and duplicating them would violate the one-entry-per-thing rule in the opposite direction. The fusion entries link to those existing nodes.

Deferred with real literature — these are the gaps this beat leaves

Each is a partner with enough evidence to deserve an entry, not written for stated reasons. Counts are title-checked PubMed intersections and recombinome Table 1 totals 12.

  • USP2 — 30 patients (rank 9), 11 records of which ~8 directly on the fusion. THE BIGGEST GAP THIS BEAT LEAVES. It is a larger partner than MLLT11 and MLLT6 and the literature on it is recent and substantive: a 2019 Leukemia report that the human KMT2A gene exhibits a second breakpoint cluster region for recurrent MLL-USP2 fusions; cryptic and atypical KMT2A-USP2 and KMT2A-USP8 rearrangements found by mate-pair sequencing in infant and childhood leukaemia; a whole-transcriptome-identified KMT2A-USP2 in infant AML; and clinical reports of blinatumomab use in KMT2A::USP2 B-ALL. The second-breakpoint-cluster claim is about KMT2A itself, not just about USP2, and this graph does not carry it. Not written here only because it is not a deferred partner from the previous beat and the beat's scope was already at nine entries.
  • MAML2 — 4 patients, ~12-13 checked records (20 raw). Strongly associated with inv(11)(q21q23) and with therapy-related T-lineage ALL after treatment for another cancer, plus a recurrent KMT2A-MAML2 in aggressive thymoma histologies — a KMT2A fusion in a solid tumour, which is unusual and relevant to the graph's non-leukaemia sections. Also one report of a KMT2A-MAML2 that emerged and regressed during neuroblastoma therapy without leukaemia over 12.8 years, which is directly relevant to the project's preleukaemic-clone material.
  • ARHGEF12/LARG — 3 patients, 13 records. A Rho guanine nucleotide exchange factor partner; count not title-checked in detail, so the 13 is a raw number and should be treated as an upper bound.
  • LASP1 — 7 records. Shares t(11;17)(q23;q21) with MLLT6 and is the impersonation partner for that band; cited in fusion-kmt2a-mllt6 but not given its own entry.
  • NEBL/nebulette — 3 patients, 5 records. Shares the 10p11-12 ambiguity with ABI1 and MLLT10.
  • KNL1/CASC5/AF15q14 — 6 patients, 6 records (all three symbols searched).
  • FLNA/filamin — 4 patients, 6 records.
  • PRPF19 — 3 patients, 2 records, but 3 of 3 cases are spliced fusions, which is the highest spliced proportion of any partner in the recombinome.

Rejected outright, with counts: TNRC18 — 6 patients, 0 records (a partner known only from the compendium, the same shape as SEPTIN3). CLTC — 3 patients, 0 records on the intersection with either the symbol or 'clathrin heavy chain'.

Three corrections to the previous enumeration, with sources

Reported rather than edited, per the standing rule; the file ana-kmt2a-nonelongation-partner-systematic-enumeration is not mine.

1. GPHN is two patients, not one, and only one of them is therapy-related. That entry's data.deferred_with_real_literature records GPHN as "the cleanest minimal-oligomerisation-domain result in the field, from a single therapy-related patient". There are two independently published index cases: a de novo acute monoblastic leukaemia with t(11;14)(q23;q24) in which GPHN fused in frame to MLL exon 9 (PMID 11289145, Kuwada et al., Cancer Res 2001 — read in abstract), and a therapy-related acute undifferentiated leukaemia in a child three years after VP16-containing chemotherapy (PMID 11579461, Eguchi et al., Genes Chromosomes Cancer 2001, DOI 10.1002/gcc.1185 — read in abstract) 3. The recombinome's own historical partner table lists GPHN with disease annotation "AML, t-AML", i.e. both 1. It should say: two independent index cases, one de novo monoblastic and one therapy-related undifferentiated.

2. MYO1F has two PubMed records, not one, and the second one matters. That entry records MYO1F as "5 patients in the recombinome; 1 PubMed record on the intersection... the count is real but there is nothing to read". The intersection (MLL OR KMT2A) AND MYO1F [Title/Abstract] returns 2 records on the current index, both title-checked: the 2005 founding report (PMID 15897884) and PMID 21620472, Leuk Res 2011, DOI 10.1016/j.leukres.2011.04.022, titled "The t(11;19)(q23;p13) fusing MLL with MYO1F is recurrent in infant acute myeloid leukemias" 24. It should say: 2 records, one of which establishes recurrence in infant AML — which is why MYO1F was promoted from deferral to an entry here rather than left as "nothing to read".

3. The MLLT11 record count understates the literature by a factor of four. That entry records "MLLT11 = 7, AF1q = 14" and a headline of 14. On the current index, (MLL OR KMT2A) AND (MLLT11 OR AF1q) [Title/Abstract] returns 20, and the gene-level query AF1q OR MLLT11 returns 88 2. The difference does not change that entry's conclusion — it deferred MLLT11 correctly — but it matters because the 88 records look like a healthy literature and are almost entirely about wild-type AF1q in solid tumours and Mllt11 in mouse neurodevelopment. This is a third alias failure mode: the count is neither deflated by a retired symbol nor inflated by a common word, but is about a different molecule — the free protein rather than the fusion.

A fourth item, not a correction but a finding in a source: the recombinome is internally inconsistent about EPS15, whose Table 1 row totals 55 while the spliced-fusion paragraph refers to "EPS15 (12 out of 51 cases)" 1.

The counted zeros, and what they do and do not bound

Three absence claims underpin entries in this beat. Each is stated with its query so a later reader can re-run it.

  1. No functional experiment on KMT2A::MLLT11. "MLL-AF1q" OR "MLL/AF1q" OR "MLL-MLLT11" OR "KMT2A-MLLT11" [All Fields] = 10 records, all titles read, all descriptive (case reports, cytogenetic surveys, multiplex RT-PCR panels). Same query AND (mice OR mouse OR retrovir* OR colony OR replating) = 0. (MLL OR KMT2A) AND (MLLT11 OR AF1q) AND (transform* OR immortali* OR leukemogen*) = 4, none a functional study 2.
  2. No mechanistic work on KMT2A::GPHN since 2004. (MLL OR KMT2A) AND (gephyrin OR GPHN) = 3 records total, all titles read: two 2001 index-case reports and the 2004 functional paper 2.
  3. No functional work on KMT2A::MYO1F. (MLL OR KMT2A) AND MYO1F = 2, both descriptive 2.

What these bound: the indexed literature. What they do not bound: a construct in a supplementary table of a screen about something else; an unpublished negative result; a preprint PubMed does not index; or — for MYO1F specifically — an experiment done on material labelled MLL-ENL or MLL-ELL without transcript typing, since those fusions are cytogenetically indistinguishable from it at t(11;19)(q23;p13).

The project's own best demonstration of the limits is that PubMed counting and patient compendia have non-overlapping blind spots. In this beat: GPHN has 3 papers and 0 recombinome patients; TNRC18 has 6 recombinome patients and 0 papers. Neither method alone enumerates the partner set.

Dependency: every partner in this beat, lineage-split

Because the coordinator established that the KMT2A-rearranged DepMap panel is myeloid-enriched (8 AML lines, 5 B-ALL) and that a myeloid-identity gene therefore scores as KMT2A-selective for free, no combined dependency number is quoted anywhere in this beat without its lineage split. The split was re-run for this beat's partners using the coordinator's own script logic 5.

genegenome-wide rank / 17,728KMT2A myeloid (SD)KMT2A B-lineage (SD)verdict
EPS151155-1.24-0.08myeloid-half only; small absolute effect; opposite lineage to the fusion's own epidemiology
MLLT118022+0.30-0.58no dependency in either half
GAS712299+0.89-0.89no dependency; symmetric split on a zero effect is noise
GPHN13812-0.41+1.47no dependency in either half
MLLT614022+0.17-0.05no dependency in either half
MYO1F16967+1.18-0.00no dependency; mild anti-dependency in the myeloid half
MEN1 (reference)1-2.29-2.28lineage-independent
KMT2A (reference)—-1.48-1.22lineage-independent

The reference genes behaving correctly is what makes the nulls interpretable. Only EPS15 shows anything, and it shows the exact signature the confound analysis flags as a myeloid-identity artefact — on a difference of only -0.10 Chronos, in a panel containing no EPS15-fusion line, and in the opposite lineage from the fusion's 40%-infant-ALL distribution.

Two Ensembl identifiers were caught resolving to the wrong gene by the script's approvedSymbol assertion: ENSG00000213445 resolves to SIPA1, not MLLT11 (correct id ENSG00000213190), and ENSG00000079462 resolves to PAFAH1B3, not EPS15L1 (correct id ENSG00000127527). Both were corrected against Ensembl REST xrefs/symbol before the numbers above were generated 5.

Facts and figures
scriptnot a script; E-utilities queries, Europe PMC full-text reads and a DepMap lineage split performed for this beat and recorded verbatim below. The lineage split is reproducible at analyses/depmap-deferred-partners-lineage-split/run.py
statusrun
run date2026-09-19
nomenclature noteThree distinct alias failure modes appeared in this beat. (1) RETIRED SYMBOLS deflate: AF1q returns 58 against MLLT11's 45 and dominates the KMT2A intersection; AF1p returns 8 against EPS15's 609; AF17 dominates the MLLT6 intersection. (2) TYPOGRAPHIC VARIANTS hide mentions: the recombinome prints MYO1F as 'MYOF1', which is not an HGNC symbol and collides with the unrelated gene MYOF. (3) A count can be neither deflated nor inflated but simply ABOUT A DIFFERENT MOLECULE: the 88 AF1q/MLLT11 records look like a healthy fusion literature and are almost all about wild-type AF1q in solid tumours and Mllt11 in mouse neurodevelopment.
deferred with real literature
ARHGEF12/LARG3.000ARHGEF12/LARG: 3NEBL3.000NEBL: 3PRPF193.000PRPF19: 3MAML24.000MAML2: 4FLNA4.000FLNA: 4KNL1/CASC5/AF15q146.000KNL1/CASC5/AF15q14: 6USP230.0USP2: 30
inputs
[
  "PubMed E-utilities esearch/efetch",
  "Europe PMC full-text XML of PMC10169636 (PMID 37019990)",
  "UniProt REST",
  "RCSB PDB entry API",
  "Ensembl REST xrefs/symbol",
  "Open Targets Platform GraphQL (DepMap Chronos)",
  "analyses/depmap-kmt2a-selective-scan-full/results.json"
]
written this beat
[
  "gene-gphn",
  "fusion-kmt2a-gphn",
  "gene-mllt11",
  "fusion-kmt2a-mllt11",
  "fusion-kmt2a-eps15",
  "gene-gas7",
  "fusion-kmt2a-gas7",
  "fusion-kmt2a-mllt6",
  "gene-myo1f",
  "fusion-kmt2a-myo1f",
  "hyp-microtubule-binding-as-a-recurrent-kmt2a-partner-module",
  "exp-af1q-oligomeric-state-and-fusion-transformation",
  "ana-kmt2a-deferred-partner-enumeration"
]
deliberately not written
[
  {
    "id": "gene-eps15",
    "reason": "protein-eps15-domains already exists in atlas2 and carries the protein-level content including the domain map and the missing-coiled-coil-coordinate problem. Writing a second EPS15 protein entry would duplicate an existing thing. fusion-kmt2a-eps15 links to it."
  },
  {
    "id": "gene-mllt6 (second copy)",
    "reason": "gene-mllt6 already exists in interactome3 with a verified accession (P55198, 1093 aa) and DotCom content. Only the fusion entry was written, as the dispatch instructed."
  },
  {
    "id": "fusion-kmt2a-gphn breakpoint exon numbers",
    "reason": "the two index-case reports describe retained protein regions, not exon junctions for both genes; no exon number was inferred"
  }
]
rejected outright
[
  {
    "partner": "TNRC18",
    "recombinome_n": 6,
    "pubmed_records": 0,
    "reason": "six patients in the compendium and zero indexed publications; the same shape as SEPTIN3 — a partner known only from a breakpoint table"
  },
  {
    "partner": "CLTC",
    "recombinome_n": 3,
    "pubmed_records": 0,
    "reason": "zero records on the intersection with either the symbol or the spelled-out protein name"
  }
]
deferred with real literature — the numbers behind the chart
[
  {
    "partner": "USP2",
    "recombinome_n": 30,
    "recombinome_rank": 9,
    "pubmed_records_raw": 11,
    "pubmed_records_checked": 8,
    "reason": "THE BIGGEST GAP LEFT BY THIS BEAT. A larger partner than MLLT11 or MLLT6 with a recent, substantive literature: a second KMT2A breakpoint cluster region for recurrent MLL-USP2 fusions (PMID 30899083), cryptic KMT2A-USP2 and KMT2A-USP8 rearrangements found by mate-pair sequencing, whole-transcriptome identification in infant AML, and clinical blinatumomab reports. The second-breakpoint-cluster claim is about KMT2A itself and this graph does not carry it."
  },
  {
    "partner": "MAML2",
    "recombinome_n": 4,
    "pubmed_records_raw": 20,
    "pubmed_records_checked": "12-13",
    "reason": "inv(11)(q21q23), therapy-related T-lineage ALL after treatment of another cancer, a recurrent KMT2A-MAML2 in aggressive thymoma histologies (a KMT2A fusion in a solid tumour), and one KMT2A-MAML2 clone that emerged and regressed during neuroblastoma therapy without leukaemia over 12.8 years. The only count in this beat materially inflated by off-topic hits."
  },
  {
    "partner": "ARHGEF12/LARG",
    "recombinome_n": 3,
    "pubmed_records_raw": 13,
    "pubmed_records_checked": null,
    "reason": "Rho GEF partner; the count was NOT title-checked and 13 is a raw upper bound"
  },
  {
    "partner": "LASP1",
    "recombinome_n": null,
    "pubmed_records": 7,
    "reason": "the impersonation partner at t(11;17)(q23;q21), cited inside fusion-kmt2a-mllt6 but not given its own entry"
  },
  {
    "partner": "NEBL",
    "recombinome_n": 3,
    "pubmed_records": 5,
    "reason": "shares the 10p11-12 ambiguity with ABI1 and MLLT10"
  },
  {
    "partner": "KNL1/CASC5/AF15q14",
    "recombinome_n": 6,
    "pubmed_records": 6,
    "reason": "three symbols searched; a real but small literature"
  },
  {
    "partner": "FLNA",
    "recombinome_n": 4,
    "pubmed_records": 6,
    "reason": "filamin A; another cytoskeletal partner, small literature"
  },
  {
    "partner": "PRPF19",
    "recombinome_n": 3,
    "pubmed_records": 2,
    "reason": "3 of 3 cases are spliced fusions, the highest spliced proportion of any partner in the recombinome, which makes it interesting to the spliced-fusion thread rather than to this beat"
  }
]
corrections to other entries
[
  {
    "target_entry": "ana-kmt2a-nonelongation-partner-systematic-enumeration",
    "field": "data.deferred_with_real_literature[GPHN].reason and the 'Deferred with real literature' section",
    "current_text": "the cleanest minimal-oligomerisation-domain result in the field, from a single therapy-related patient",
    "should_say": "two independently published index cases: a de novo acute monoblastic leukaemia (PMID 11289145) and a paediatric therapy-related acute undifferentiated leukaemia after VP16 (PMID 11579461, DOI 10.1002/gcc.1185). Only the second is therapy-related.",
    "citation": "PMID 11289145 and PMID 11579461, both read in abstract",
    "read_level": "abstract only, both"
  },
  {
    "target_entry": "ana-kmt2a-nonelongation-partner-systematic-enumeration",
    "field": "data.deferred_with_real_literature[MYO1F].pubmed_records and .reason",
    "current_text": "pubmed_records: 1 — 'real patient count, nothing to read'",
    "should_say": "pubmed_records: 2. The second is PMID 21620472, Leuk Res 2011, DOI 10.1016/j.leukres.2011.04.022, 'The t(11;19)(q23;p13) fusing MLL with MYO1F is recurrent in infant acute myeloid leukemias' — which establishes recurrence and is why MYO1F was promoted to an entry in this beat.",
    "citation": "PMID 21620472, record fetched; PubMed carries no abstract for it, so only the title is relied on",
    "read_level": "title and PubMed record; no abstract exists"
  },
  {
    "target_entry": "ana-kmt2a-nonelongation-partner-systematic-enumeration",
    "field": "data.queries.mll_or_kmt2a_intersections.MLLT11 and .AF1q, and data.deferred_with_real_literature[MLLT11].pubmed_records",
    "current_text": "MLLT11: 7, AF1q: 14; headline pubmed_records 14",
    "should_say": "the (MLL OR KMT2A) AND (MLLT11 OR AF1q) [Title/Abstract] intersection returns 20 on the 2026-09-19 index, and the gene-level query AF1q OR MLLT11 returns 88. The conclusion to defer was correct; the count understated the literature roughly fourfold, and the 88 records are almost entirely about wild-type AF1q in solid tumours and Mllt11 in mouse neurodevelopment rather than about the fusion.",
    "citation": "NCBI E-utilities esearch, queries and counts in data.queries below, all titles read",
    "read_level": "all 88 titles read"
  }
]
queries
{
  "mll_or_kmt2a_intersections_tiab": {
    "gephyrin OR GPHN": 3,
    "MLLT11 OR AF1q": 20,
    "EPS15 OR AF1p": 15,
    "MLLT6 OR AF17": 31,
    "GAS7": 5,
    "MYO1F": 2,
    "USP2": 11,
    "MAML2 OR MAMLD": 20,
    "ARHGEF12 OR LARG": 13,
    "LASP1": 7,
    "KNL1 OR CASC5 OR AF15q14": 6,
    "FLNA OR filamin": 6,
    "NEBL OR nebulette": 5,
    "PRPF19 OR SNEV OR PSO4": 2,
    "TNRC18": 0,
    "CLTC OR clathrin heavy chain": 0
  },
  "gene_level_controls_tiab": {
    "AF1q": 58,
    "MLLT11": 45,
    "AF1q OR MLLT11": 88,
    "EPS15": 609,
    "AF1p": 8,
    "gephyrin": 912,
    "GPHN": 76,
    "GAS7": 143,
    "MYO1F": 70
  },
  "fusion_name_queries_all_fields": {
    "\"MLL-AF1q\" OR \"MLL/AF1q\" OR \"MLL-MLLT11\" OR \"KMT2A-MLLT11\"": 10,
    "same AND (mice OR mouse OR retrovir* OR colony OR replating)": 0
  },
  "mechanism_queries": {
    "dimerization[Title] AND (MLL[Title] OR KMT2A[Title])": 4,
    "(MLL OR KMT2A) AND (MLLT11 OR AF1q) AND (transform* OR immortali* OR leukemogen*)": 4,
    "AF1q AND (crystal OR NMR OR 'intrinsically disordered' OR structure)": 5,
    "MLLT11 AND (knockout OR knock-out OR null OR conditional)": 3,
    "gephyrin[Title] AND (tubulin OR microtubule)": 23,
    "gephyrin[Title] AND (trimer* OR dimeri* OR oligomer* OR lattice OR crystal)": 46
  },
  "checked_vs_raw": {
    "MAML2": "20 raw, ~12-13 on topic after title-checking; the remainder are MAML2 in mucoepidermoid carcinoma and thymoma without a KMT2A fusion",
    "USP2": "11 raw, ~8 directly on KMT2A::USP2",
    "ARHGEF12": "13 raw, NOT title-checked, reported as an upper bound",
    "all_others": "title-checked; raw equals checked"
  }
}
lineage split results
{
  "method": "KMT2A-rearranged index arm split into 8 myeloid and 5 B-lineage lines, each half scored in SD of the other-leukaemia arm; reproduces analyses/lineage-confound-myeloid-tf-dependencies/",
  "EPS15": {
    "rank": 1155,
    "myeloid_sd": -1.24,
    "b_lineage_sd": -0.08,
    "verdict": "myeloid-half only, small absolute effect, opposite lineage to the fusion's epidemiology"
  },
  "MLLT11": {
    "rank": 8022,
    "myeloid_sd": 0.3,
    "b_lineage_sd": -0.58,
    "verdict": "no dependency"
  },
  "GAS7": {
    "rank": 12299,
    "myeloid_sd": 0.89,
    "b_lineage_sd": -0.89,
    "verdict": "no dependency"
  },
  "GPHN": {
    "rank": 13812,
    "myeloid_sd": -0.41,
    "b_lineage_sd": 1.47,
    "verdict": "no dependency"
  },
  "MLLT6": {
    "rank": 14022,
    "myeloid_sd": 0.17,
    "b_lineage_sd": -0.05,
    "verdict": "no dependency"
  },
  "MYO1F": {
    "rank": 16967,
    "myeloid_sd": 1.18,
    "b_lineage_sd": 0,
    "verdict": "no dependency; possible non-expression artefact"
  },
  "MEN1_reference": {
    "myeloid_sd": -2.29,
    "b_lineage_sd": -2.28,
    "verdict": "lineage-independent"
  },
  "KMT2A_reference": {
    "myeloid_sd": -1.48,
    "b_lineage_sd": -1.22,
    "verdict": "lineage-independent"
  }
}
identifier failures
[
  "ENSG00000213445 (used for MLLT11 in the first lineage-split run) resolves to SIPA1 in the Open Targets target record. Correct id: ENSG00000213190, confirmed by Ensembl REST xrefs/symbol. Caught by the script's approvedSymbol assertion, not by inspection.",
  "ENSG00000079462 (used for EPS15L1) resolves to PAFAH1B3. Correct id: ENSG00000127527.",
  "Three PubMed records in this beat's drafts initially carried author lists or DOIs typed from recall rather than copied from the fetched record; all three were wrong and were corrected by re-fetching (PMIDs 22682626, 10700861, 12750700 in one entry; 16897742, 12527918, 35091795 in another). The failure mode is exactly the one the project brief names: fetching a record, reading one field off it, and typing the next from memory."
]
How this could bite us back
  • Every absence claim in this beat is an absence in the INDEXED literature. The counted zero for KMT2A::MLLT11 functional work is the load-bearing one, and it would be defeated by a single construct in a supplementary table of a screen about something else. It is reported as a bounded claim for that reason and should not be paraphrased as 'the experiment has not been done'.
  • Counting partners by current HGNC symbol under-counts this class, and counting them by retired symbol can retrieve a literature about the wrong molecule entirely. Any automated partner enumeration in this project must search both AND title-check, and the MLLT11 case shows that even a correct, title-checked count of 88 records can correspond to zero records about the fusion.
  • The lineage split protects against one confound and introduces a false sense of completeness: no cell line carrying ANY of this beat's six fusions exists in DepMap, so a clean null in both halves is measuring the wrong thing carefully.
Still open
  1. USP2 is rank 9 with 30 patients, a recent substantive literature, and a published claim of a SECOND KMT2A breakpoint cluster region. It is the largest uncovered partner in the landscape and deserves the next partner beat.
  2. How many of the 63 partners seen only once in the recombinome exist solely in that table, as TNRC18 (6 patients, 0 papers) and SEPTIN3 (1 patient, 0 papers) do? A systematic pass would quantify how much of the partner landscape has never been described in prose.
  3. MAML2 puts a KMT2A fusion in a solid tumour (thymoma) and in a clone that arose and regressed during neuroblastoma therapy without causing leukaemia. Both are directly relevant to this project's preleukaemic and non-leukaemia sections and neither is in the graph.
  4. Is there a systematic way to detect typographic gene-symbol variants in full text? 'MYOF1' for MYO1F defeated symbol-keyed search in a paper this project reads repeatedly, and nobody had noticed.
Every source behind this entry (6)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Patient counts for this beat were read from Table 1 and the AML/ALL prose of the 2023 KMT2A recombinome (3401 patients; 107 in-frame fusions known in total, 37 recurrent, 63 seen only once): EPS15 55 (rank 8), USP2 30 (rank 9), MLLT11 28 (rank 10), MLLT6 27 (rank 11), MYO1F 5 (rank 22), MAML2 4, GAS7 3 (rank 28), NEBL 3, ARHGEF12 3, CLTC 3, PRPF19 3, KNL1 6, TNRC18 6, FLNA 4, GPHN 0 (present only in the historical partner table, at t(11;14)(q23.3;q23.3-q24.1), Kuwada et al. 2001, disease annotation AML and t-AML); spliced fusions account for EPS15 12 of 51, PRPF19 3 of 3 and MYO1F 1 of 5; the paper's AML prose prints MYO1F as 'MYOF1'.

    The KMT2A recombinome of acute leukemias in 2023. · Meyer C, Larghero P, Almeida Lopes B, Burmeister T, Gröger D, Sutton R, et al. · Leukemia 2023
    PMID 37019990DOI
    What this evidence cannot show

    Read directly from the Europe PMC full-text XML of PMC10169636 for this beat, with the Table 1 column header verified rather than assumed; the abstract reports none of these counts. A referral cohort accrued 2003-2022 from 20 countries, so counts are ascertainment-weighted floors and not population incidence. The paper is internally inconsistent for EPS15 (Table 1 gives 55; the spliced-fusion prose says 'out of 51 cases').

    Evidence that argues against this (1)
    • The compendium's counts are floors by construction for exactly the partners in this beat, three of which are documented impersonators: MLL-ENL, MLL-ELL, MLL-EEN and MLL-MYO1F are difficult to distinguish at t(11;19)(q23;p13); t(11;17)(q23;q21) produces both MLLT6 and LASP1; and the index MLL-GAS7 presented as a karyotypic del(11)(q23) requiring panhandle PCR to resolve. Taki T et al. 2005
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    PubMed record counts run for this beat on 2026-09-19 by raw E-utilities esearch, every result set title-checked. Intersections with (MLL OR KMT2A) [Title/Abstract]: gephyrin OR GPHN = 3; MLLT11 OR AF1q = 20; EPS15 OR AF1p = 15; MLLT6 OR AF17 = 31; GAS7 = 5; MYO1F = 2; USP2 = 11 (~8 on the fusion); MAML2 OR MAMLD = 20 raw, ~12-13 checked; ARHGEF12 OR LARG = 13; LASP1 = 7; KNL1 OR CASC5 OR AF15q14 = 6; FLNA OR filamin = 6; NEBL OR nebulette = 5; PRPF19 OR SNEV OR PSO4 = 2; TNRC18 = 0; CLTC OR 'clathrin heavy chain' = 0. Gene-level controls: AF1q = 58, MLLT11 = 45, AF1q OR MLLT11 = 88, EPS15 = 609, AF1p = 8, gephyrin = 912, GPHN = 76, GAS7 = 143, MYO1F = 70. Fusion-name queries [All Fields]: "MLL-AF1q" OR "MLL/AF1q" OR "MLL-MLLT11" OR "KMT2A-MLLT11" = 10, and the same AND (mice OR mouse OR retrovir* OR colony OR replating) = 0.

    What this evidence cannot show

    Counts are of records whose title or abstract (or, for the fusion-name queries, any field) contains the terms. They proxy literature volume, not quality or relevance. Every set that drove a decision was title-checked and the checked number is what is reported where the two differ, which is the case only for MAML2 (20 raw, ~12-13 checked) and USP2 (11 raw, ~8 on the fusion). The ARHGEF12 count of 13 was NOT title-checked and is reported as a raw upper bound. PubMed indexing changes, so these are dated.

    Evidence that argues against this (1)
    • Title/abstract counting and patient compendia have non-overlapping blind spots, demonstrated twice within this beat alone: GPHN has 3 papers and 0 patients in the compendium, while TNRC18 has 6 patients and 0 papers. Neither method alone enumerates the partner set, and a bounded, well-written absence claim from either one is the most trustworthy-looking way to be wrong. Meyer C et al. 2023
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    KMT2A::GPHN has two independently published index cases, not one: a de novo acute monoblastic leukaemia with t(11;14)(q23;q24) in which a human gephyrin homologue fused in frame to MLL exon 9, the chimera carrying the gephyrin C terminus including the tubulin binding site; and a paediatric therapy-related acute undifferentiated leukaemia arising three years after VP16-containing chemotherapy, whose genomic breakpoints in both MLL and GPHN were spanned by potential in vitro topoisomerase-II DNA-binding sites with no homologous-recombination motifs.

    t(11;14)(q23;q24) generates an MLL-human gephyrin fusion gene along with a de facto truncated MLL in acute monoblastic leukemia. · Kuwada N, Kimura F, Matsumura T, Yamashita T, Nakamura Y, Wakimoto N, et al. · Cancer Res 2001
    PMID 11289145
    What this evidence cannot show

    Both cases read in abstract only; full texts were not obtained. No DOI is registered in PubMed for this 2001 Cancer Research record, so the citation is anchored on PMID and URL. Two case reports remain two case reports: this correction changes the description of the evidence base, not its weight.

    Evidence that argues against this (1)
    • The second index case is the therapy-related one and describes the retained MLL portion differently — AT-hooks plus the DNA-methyltransferase-homology domain fused to the C-terminal half of gephyrin — so the two cases do not share an identical junction and 'two index cases' should not be read as 'two instances of one fusion protein'. Eguchi M et al. 2001
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    The KMT2A-MYO1F literature contains a second record beyond the founding case report, establishing that the t(11;19)(q23;p13) fusing MLL with MYO1F is recurrent in infant acute myeloid leukaemias.

    The t(11;19)(q23;p13) fusing MLL with MYO1F is recurrent in infant acute myeloid leukemias. · Duhoux FP, Ameye G, Libouton J, Bahloula K, Iossifidis S, Chantrain CF, et al. · Leuk Res 2011
    PMID 21620472DOI
    What this evidence cannot show

    PubMed carries no abstract for this record, so the claim asserted is exactly the title's content. Case numbers and breakpoints are not asserted. It is cited here to correct a record count, and the correction stands on the record existing and being on-topic, both of which were verified by fetching it.

    Evidence that argues against this (1)
    • Recurrence at this scale means very few cases: the largest systematic series records 5 MYO1F patients in 3401, so the correction raises the record count from 1 to 2 and the partner from 'nothing to read' to 'two descriptive papers', not to a characterised fusion. Meyer C et al. 2023
  5. E5moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    A lineage-split DepMap re-analysis was run for every partner in this beat, splitting the KMT2A-rearranged index arm into its 8 myeloid and 5 B-lineage lines and scoring each half in standard deviations of the other-leukaemia arm. Results (myeloid SD / B-lineage SD): EPS15 -1.24 / -0.08; MLLT11 +0.30 / -0.58; GAS7 +0.89 / -0.89; GPHN -0.41 / +1.47; MLLT6 +0.17 / -0.05; MYO1F +1.18 / -0.00; references MEN1 -2.29 / -2.28 and KMT2A -1.48 / -1.22, both lineage-independent. Only EPS15 shows any signal and it is myeloid-half only, on a genome-wide difference of -0.0997 Chronos. Two Ensembl identifiers were rejected by the run's approvedSymbol assertion: ENSG00000213445 resolves to SIPA1 rather than MLLT11 (correct id ENSG00000213190) and ENSG00000079462 resolves to PAFAH1B3 rather than EPS15L1 (correct id ENSG00000127527).

    What this evidence cannot show

    No cell line carrying ANY of these six fusions exists in the DepMap panel, so every number measures the wild-type partner gene in leukaemias driven by other partners. Each half of the split has 8 or 5 lines, so a single line moves a mean and a ±1 SD swing on a near-zero absolute effect is noise. For MYO1F there is an additional problem: its founding report found the gene expressed in only 3 of 22 leukaemic cell lines, so a CRISPR null may reflect non-expression rather than dispensability. DepMap release version is not stated by the API.

    Evidence that argues against this (1)
    • The coordinator's own analysis states the limits of this procedure: the split is by lineage LABEL rather than by expression signature, KMT2A-rearranged disease includes genuinely ambiguous-lineage cases that a label forces into one bin, and the method bounds a dependency's lineage specificity without testing therapeutic relevance — so a clean null in both halves is weaker reassurance than it appears. Oncofusion knowledge graph et al. 2026
  6. E6moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    The KMT2A::USP2 literature contains a structural claim about KMT2A itself that this graph does not carry: the human MLL/KMT2A gene exhibits a SECOND breakpoint cluster region for recurrent MLL-USP2 fusions.

    Human MLL/KMT2A gene exhibits a second breakpoint cluster region for recurrent MLL-USP2 fusions. · Meyer C, Lopes BA, Caye-Eude A, Cavé H, Arfeuille C, Cuccuini W, et al. · Leukemia 2019
    PMID 30899083DOI
    What this evidence cannot show

    Cited from the fetched PubMed record's title, authors, year, venue and DOI in a single read; the abstract and full text were NOT read for this entry, so the location and extent of the second cluster region, the number of patients and the detection method are not asserted. It is recorded here to name the gap precisely enough that the next agent can act on it, not to make a claim about KMT2A breakpoint architecture.

    Evidence that argues against this (1)
    • The same group's later compendium describes the KMT2A breakpoint cluster region without, in its abstract, foregrounding a second cluster, and reports USP2 among ordinary partners at 30 patients — so how general or how load-bearing the second-cluster finding is cannot be judged from the title alone, which is exactly why this is flagged as a gap rather than written up. Meyer C et al. 2023
Connected to