Oncofusion

start / graph / ana-the-arid1a-kmt2a-combination-exists-in-patients-so-the-empty-cell-line-intersection-is-a-panel-artefact

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

In St. Jude paediatric AML, 2 of the 6 ARID1A-mutant cases are KMT2A-rearranged and BOTH ARID1A mutations are TRUNCATING frameshifts in KMT2A::MLLT10 cases - so the exact combination cycle 86 showed the ARID1B dependency is confined to DOES occur in patients, and cycle 88's empty DepMap intersection is a cell-line sampling artefact rather than a fact about the disease. The paediatric ALL cohort returned zero ARID1A mutations, which is panel coverage and not biology

also called does ARID1A mutation co-occur with KMT2A rearrangement in patients · the empty intersection is a panel artefact · 2 of 6

Eight cycles of work showed that one chromatin gene becomes important when its partner gene is broken, and then found that no laboratory cell line carries both that broken partner and the gene rearrangement this project studies - which left the whole line of work stranded. This asks the version of the question that matters: do actual patients have both? In a study of nearly nine hundred children with acute myeloid leukaemia, six had the broken partner gene and two of those six also carried the rearrangement, so the combination is real in people even though no cell line models it. A second, larger children's study of a different leukaemia reported no cases at all - but it also reported none anywhere in the cohort, which means its sequencing panel did not cover the gene, so that zero is about the test and not about patients.

The same thing, for a scientist

**Statement.** Cycles 81–87 established ARID1B's dependency on ARID1A status pan-cancer (−0.2475, p = 5e-05, 124 truncating lines). Cycle 88 found the DepMap intersection of truncating-ARID1A and KMT2A-rearranged is **empty**, and wrote: *"the cell-line panel lacking the combination says little about the disease… cBioPortal serves paediatric and adult AML/ALL cohorts that could answer it in one query. That is the version of this question worth asking, and it replaces waiting for a fifth cell line."* Asked. **`aml_stjude_2024` — Pediatric AML, n = 887, structural-variant profile:** | | | |---|---:| | KMT2A structural-variant records | **237** | | KMT2A-rearranged samples | **180** | | ARID1A-mutant samples, cohort-wide | **6** | | **KMT2A-rearranged AND ARID1A-mutant** | **2** | The fusion partners recovered are the real spectrum — **AFF1**, AFDN, ELL, EPS15, MLLT1, MLLT3, MLLT10, MLLT11 — so the rearrangement variable is capturing what it should. ARID1A mutation types include **Frame_Shift_Del, Frame_Shift_Ins and Nonsense_Mutation**, i.e. genuinely truncating calls, not only missense. **So the combination exists.** Read from the rearranged side it is 2/180 = **1.1%**. Read from the ARID1A side it is **2 of 6 ARID1A-mutant cases**, against a cohort-wide rearrangement rate of 180/887 ≈ 20% — so ARID1A mutation is **not depleted** among rearranged cases, though at n = 6 that comparison carries no weight and is stated only to forestall the opposite reading. **`all_phase2_target_2018_pub` — Pediatric ALL, n = 1,978, `MLL_STATUS` clinical attribute:** 48 Positive, 1,464 Negative, 431 Unknown — **and ZERO ARID1A mutations anywhere in the cohort.** By the clause this analysis declared in advance, a cohort-wide count of exactly zero is a **likely panel-coverage artefact**, not biology. TARGET Phase II used targeted sequencing and ARID1A appears not to be on the panel. **That cohort contributes nothing in either direction** and its 48 rearranged cases must not be counted as evidence of absence. **And both ARID1A mutations are TRUNCATING** — `S795Gfs*39` and `S11Qfs*100`, two frameshift insertions. That was the load-bearing follow-up, because cycle 86 confined the ARID1B dependency **entirely** to truncating ARID1A with a tight missense null. Two missense cases would have meant the relevant combination is documented zero times in patients as well as zero times in DepMap. It is not: **the exact combination exists.** **The deflation is the partner.** Both double-positives are **KMT2A::MLLT10** — myeloid, not the B-lymphoid KMT2A::AFF1 this project exists for. So the honest headline is *documented twice, in a different fusion, in a different lineage.* **The conclusion cycle 88 could not reach:** the empty DepMap intersection is a property of **which cell lines exist and have been screened**, not of the disease. Patients with both lesions are documented, and the lesions are of the right kind. The ARID1B buffering biology therefore has a population it could be relevant to — a small one, in the wrong fusion, and one no cell line currently models.

Go deeper
My pre-registered scoring pooled a cohort that could not contribute, and the clause that should have caught it was applied to the wrong level

I declared in advance: "If a cohort's ARID1A mutation count is exactly zero across all samples, that is reported as a LIKELY PANEL-COVERAGE ARTEFACT and not as biology." That is the right clause and it fires for the ALL cohort.

But my scoring code applied it to the POOLED count. Pooled ARID1A-mutant cases across both cohorts is 6, which is non-zero, so the artefact test never triggered — and the script pooled 48 uninformative ALL rearranged cases into the denominator, producing "2 of 228 (0.88%), below the 2% threshold, verdict declined". The correct denominator is 180, the AML cohort alone, giving 1.1% — still below my threshold, but a materially different number reached for a defensible reason rather than by diluting with a cohort that could not contribute.

This is the sixth time in this project a pre-registered criterion has been sound in form and wrong in application, after a reagent that could not distinguish its subject, a population never stated, a measure saturated at its sample size, a control not independent of the treatment, and an or that let a dead clause decide. The new variant: a clause written per-cohort and evaluated per-pool. The clause caught the problem anyway — because I printed the per-cohort counts and read them — which is the argument for reporting a test's own control at the same granularity the clause is written at.

The 2% threshold itself was arbitrary and I should say so. I chose it before pulling, which protects against choosing it afterwards, but nothing justified 2% over 1% or 5%. The honest output is the count and the denominator; the threshold adds a verdict the data does not need.

Facts and figures
truncating verdictT1 - both truncating. The EXACT combination cycle 86 confined the ARID1B dependency to (truncating ARID1A, not missense) exists in patients.
but both partners are mllt10not AFF1. The combination is documented in a fusion this project does not study, in myeloid rather than B-lymphoid lineage.
mutationtype audit partially dischargedcycle 88 listed 'audit cBioPortal mutationType' as a standing debt across six analyses. On the records that carry THIS result the vocabulary is exactly four clean values with no Splice_Site and nothing unclassified. That discharges the debt for this study only; the other five analyses use other studies and remain unaudited.
questiondoes ARID1A mutation co-occur with KMT2A rearrangement IN PATIENTS?
why it replaced the previous questioncycle 88 found the DepMap intersection of truncating-ARID1A and KMT2A-rearranged is EMPTY, and that a cell-line panel lacking a combination says little about the disease
the answerTHE COMBINATION EXISTS IN PATIENTS. The empty DepMap intersection is a property of which cell lines exist and have been screened, not of the disease.
independence check that deflates itexpected double-positives under independence = 6 x 0.203 = 1.2; observed 2. This establishes OCCURRENCE and gives no evidence of enrichment.
my scoring errorI declared the panel-coverage clause PER COHORT and evaluated it PER POOL. Pooled ARID1A-mutant count is 6, non-zero, so the artefact test never fired, and 48 uninformative ALL rearranged cases were pooled into the denominator - printing '2 of 228' where the defensible figure is '2 of 180'. Sixth instance of a criterion sound in form and wrong in application; the new variant is a clause written per-cohort and evaluated per-pool.
what caught itprinting the per-cohort counts and reading them. That is the argument for reporting a test's own control at the same granularity the clause is written at.
the threshold was arbitrary2% was chosen before pulling, which protects against choosing it afterwards, but nothing justified 2% over 1% or 5%. The count and denominator are the output; the threshold adds a verdict the data does not need.
absence classmeasured-not-curated
absence class noteBoth cohorts have been public and queryable for the whole life of this project, and the co-occurrence question was askable in one query from cycle 81 onward. Eight cycles reasoned about whether the biology transfers to this disease while the patient data that bounds the question sat unqueried. Not never-measured: the sequencing was done, deposited and indexed.
the two cases
[
  {
    "sampleId": "SJAML016538_D1",
    "arid1a": "Frame_Shift_Ins S795Gfs*39",
    "class": "TRUNCATING",
    "kmt2a_partner": "MLLT10"
  },
  {
    "sampleId": "SJAML030324_D1",
    "arid1a": "Frame_Shift_Ins S11Qfs*100",
    "class": "TRUNCATING",
    "kmt2a_partner": "MLLT10"
  }
]
aml stjude 2024
{
  "disease": "Pediatric Acute Myeloid Leukemia (St. Jude 2024)",
  "n_samples": 887,
  "kmt2a_sv_records": 237,
  "kmt2a_rearranged_samples": 180,
  "arid1a_mutant_cohort_wide": 6,
  "both": 2,
  "rate_of_rearranged_carrying_arid1a_pct": 1.11,
  "fusion_partners_recovered": [
    "AFF1",
    "AFDN",
    "ELL",
    "EPS15",
    "MLLT1",
    "MLLT3",
    "MLLT10",
    "MLLT11"
  ],
  "arid1a_mutation_types": [
    "Frame_Shift_Del",
    "Frame_Shift_Ins",
    "Missense_Mutation",
    "Nonsense_Mutation"
  ]
}
all phase2 target 2018 pub
{
  "disease": "Pediatric Acute Lymphoid Leukemia (TARGET Phase II)",
  "n_samples": 1978,
  "mll_status": {
    "Negative": 1464,
    "Unknown": 431,
    "Positive": 48
  },
  "arid1a_mutant_cohort_wide": 0,
  "verdict": "UNINFORMATIVE - a cohort-wide ARID1A count of exactly zero is a LIKELY PANEL-COVERAGE ARTEFACT by this analysis's own declared clause, not biology. Contributes nothing in either direction."
}
Still open
  1. ANSWERED IN THIS ENTRY (E2): both are truncating frameshifts, and both are KMT2A::MLLT10. The follow-up that replaces it: is there ANY documented case of truncating ARID1A with KMT2A::AFF1 specifically? Both cases here are myeloid and MLLT10-partnered, so the B-lymphoid fusion this project exists for still has zero documented instances.
  2. Do the two frameshifts escape nonsense-mediated decay? S11Qfs*100 is at codon 11 of 2,285 and is a strong NMD candidate, so 'truncating by annotation' may understate or misdescribe the protein consequence. This study has expression data and the check is cheap.
  3. Is there an adult AML or B-ALL cohort with both ARID1A panel coverage and KMT2A fusion calls? The disease this project exists for is B-lymphoblastic, and the only B-lineage cohort available here could not test ARID1A at all.
  4. Does the St. Jude structural-variant call for KMT2A agree with that cohort's own clinical subtype annotation? Two instruments, not cross-validated here, and the 180 figure carries the result.
Every source behind this entry (2)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.analysis

    In cBioPortal study aml_stjude_2024 (Pediatric AML, n=887), the STRUCTURAL_VARIANT profile returns 237 KMT2A records across 180 samples, with partners including AFF1, AFDN, ELL, EPS15, MLLT1, MLLT3, MLLT10 and MLLT11; the mutations profile returns 6 ARID1A-mutant samples cohort-wide, with Frame_Shift_Del, Frame_Shift_Ins, Nonsense_Mutation and Missense_Mutation calls; and 2 samples carry both a KMT2A structural variant and an ARID1A mutation. In all_phase2_target_2018_pub (Pediatric ALL, n=1978), MLL_STATUS is Positive in 48, Negative in 1464 and Unknown in 431, and ARID1A mutations number ZERO cohort-wide.

    What this evidence cannot show

    n = 2 IS TWO PATIENTS AND CARRIES NO STATISTICAL WEIGHT. Nothing here supports a claim about enrichment, depletion, or a mechanistic relationship - only that the combination is documented. The ALL cohort's zero is a PANEL-COVERAGE ARTEFACT by this analysis's own declared clause and contributes nothing in either direction; its 48 rearranged cases must not be read as 48 cases lacking ARID1A mutation. My scoring code pooled that cohort into the denominator anyway - the clause was written per-cohort and evaluated per-pool - so the script's printed verdict quotes 2 of 228 where the defensible figure is 2 of 180. I did not check whether the two double-positive samples' ARID1A calls are truncating or missense in THIS evidence item; E2 resolves it and both are truncating frameshifts. cBioPortal's structural-variant calls and the trial's MLL_STATUS are two different instruments and are not cross-validated here.

    Evidence that argues against this (3)
    • Two cases out of 887 is also consistent with pure chance co-occurrence of two independent lesions. With ARID1A mutant in 6/887 (0.68%) and KMT2A-rearranged in 180/887 (20.3%), the expected number of double-positives under independence is 6 x 0.203 = 1.2, and 2 observed against 1.2 expected is exactly what independence predicts. So this establishes that the combination OCCURS and provides no evidence whatever that it occurs more than by chance - and the entry's framing should not be read as suggesting otherwise.
    • The finding rescues the eight-cycle ARID1B thread in a weak sense that deserves naming plainly: it restores a population for the biology to be relevant to, but that population is roughly 1% of KMT2A-rearranged paediatric AML and zero measurable cases in the B-lymphoblastic disease KMT2A::AFF1 actually defines. The ALL cohort could not test it, so the specific disease this project exists for remains untouched - and a reader could reasonably conclude the thread's relevance is now bounded at 'rare subset of a different leukaemia'.
    • I treated MLL_STATUS 'Positive' as KMT2A rearrangement without checking how the trial defined or assayed it, and the 431 'Unknown' cases are 22% of that cohort. If Unknown is enriched for anything, the 48 Positive is not a clean denominator - but since that cohort turned out uninformative on ARID1A coverage anyway, the error would not change the conclusion. It would matter if anyone reused the MLL_STATUS numbers for a different question.
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.analysis

    Both KMT2A-rearranged ARID1A-mutant samples in aml_stjude_2024 carry TRUNCATING ARID1A mutations: SJAML016538_D1 with Frame_Shift_Ins S795Gfs*39 and SJAML030324_D1 with Frame_Shift_Ins S11Qfs*100. Both are KMT2A::MLLT10. Cohort-wide the ARID1A mutationType vocabulary in this study is exactly four values - Frame_Shift_Del, Frame_Shift_Ins, Missense_Mutation, Nonsense_Mutation - with no Splice_Site and no value outside the pre-declared truncating and non-truncating lists.

    What this evidence cannot show

    STILL TWO PATIENTS. Establishing that both are truncating removes an alternative reading; it adds no statistical weight whatever. The truncating/non-truncating vocabulary is MINE, declared before the pull, and Splice_Site was deliberately placed in neither list because its consequence depends on position - that clause did not have to fire here, so it remains untested on this instrument. I did not confirm the frameshifts are in ARID1A's ARID or BAF250 domains or that they escape nonsense-mediated decay, so 'truncating by annotation' is not the same as 'protein-null'. The partner in BOTH cases is MLLT10, not AFF1, so the combination is documented in a fusion this project does not study.

    Evidence that argues against this (2)
    • That both partners are MLLT10 substantially narrows what this supports, and the coincidence deserves suspicion rather than enthusiasm at n = 2. KMT2A::MLLT10 AML and KMT2A::AFF1 infant B-ALL differ in lineage, in HOXA dependence and in menin-inhibitor response; nothing here licenses carrying the co-occurrence across to the fusion this project exists for. A reader could fairly say the honest headline is 'documented twice, in a different fusion, in a different lineage'.
    • S11Qfs*100 is at codon 11 of a 2,285-residue protein, which is so close to the start that the transcript is a strong nonsense-mediated-decay candidate and may produce no protein at all rather than a truncated one - a stronger loss than 'truncating' implies, but also a case where re-initiation at a downstream ATG is possible and the actual consequence is unverified. I classified both by annotation string alone and did not check either against the cohort's own expression data, which for this study exists.
Connected to