The pre-registered three-branch test of 'the HOXA-negative blast's oncoprotein is on the GMP programme' returns NEITHER/UNINFORMATIVE, on four independent gates
We wrote down in advance what would count as proof that the fusion protein moves to myeloid genes, what would count as proof it does not, and what would make the whole test say nothing. The third one happened, four times over.
The same thing, for a scientist
🟠 **CONFIRMED AT INTERVAL SCALE 2026-09-24 (cycle 139).** This entry named interval scoring as its cheapest remaining gap and as *the one measurement that could overturn the headline*. Q150 ran it: the `AML_specific` class is enriched in **0 of 24** untreated non-circular samples, while `Oncoprotein_Target` reaches **19 of 24** and `ALL_specific` **10 of 24** on the same intervals and null. **The controls make the zero a measurement.** And the width rationale was weaker than it looked - a promoter window already captures **0.62-0.66** of the GMP class's signal. **Your NEITHER verdict is confirmed by the test you nominated to overturn it.** See `ana-scoring-the-authors-intervals-instead-of-promoters-leaves-the-gmp-like-class-at-zero-of-twenty-four`. --- Branches were fixed before any programme score was computed. **GMP-DESTINATION did not fire. NO-DESTINATION did not fire. NEITHER/UNINFORMATIVE fired**, on four of the seven pre-declared uninformative conditions: - **U1** - the pre-registered reagent gate failed (and the gate itself was the wrong test; see the concept entry). - **U4** - A73051's AML/ALL ratio, **3.493**, sits **inside** the range of the five depth-adequate, HOXA-POSITIVE Fred Hutch paediatric AMLs (**1.263-7.189**), rank 3 of 6. The GMP-like elevation is a lineage property, not the HOXA-negative blast's destination. - **U5** - AML/ALL separates on institution and lineage (Fred Hutch median 3.031 vs Texas Children's 1.182, Mann-Whitney **two-sided p = 0.0087**; AML 3.031 vs B-lineage 1.268, two-sided p = 0.0080) and not at all on HOXA status (HOXA-negative 0.940 and 3.493, straddling the HOXA-positive median). - **U7** - A73051 fails its own readability: its canonical `Oncoprotein_Target` share is **1.208x**, below the 1.5x gate, the lowest of the 25 untreated samples. **An UNINFORMATIVE outcome is a real outcome and it is handed back as one.** It is not evidence for NO-DESTINATION, and it is not evidence for GMP-DESTINATION.
Every group that cannot carry a statistic, named rather than computed
Depth-ADEQUATE, untreated, non-circular samples by partner:
| partner | n | smallest attainable exact two-sided p | verdict |
|---|---|---|---|
| KMT2A-AFF1 | 3 (p179, p247, p454) | 2/6 = 0.33 | cannot carry a statistic |
| KMT2A-ENL | 4 (TB971992, p160, p186, p214) | 2/24 = 0.083 | cannot carry a statistic |
| KMT2A-AFDN | 3 (A70498, A82485, A89691) | 0.33 | cannot carry a statistic |
| KMT2A-AF10 | 2 (A73051, A93692) | - | cannot carry a statistic |
| KMT2A-AF9 | 1 (A95941) | - | cannot carry a statistic |
| KMT2A-EPS15 | 2 (R3595, R3652) | - | cannot carry a statistic |
No partner group in this deposit can support a partner-level statistic on this question. Every partner number in the output is a point value with its n stated.
The decisive comparison - A73051 against the five HOXA-positive Fred Hutch AMLs - is n = 1 versus n = 5. The smallest attainable exact two-sided p for A73051 being the extreme of six is 2/6 = 0.33, so no p is reported for it at all; only the point values, the rank (3 of 6) and the range.
The gates that did NOT fire, which matters as much
- U2 did not fire: every class retained at least 20 joined promoters (37, 40, 827).
- U3 did not fire: A73051's eight-locus non-expressed negative-control class sits at 0.173x, far below 1.0, so the within-sample share does discriminate in that sample.
- U6 did not fire: Spearman(AML/ALL, p_dz) over the 15 depth-adequate non-circular untreated samples is rho = +0.268, two-sided p = 0.334. The ordering is not depth.
What a POST-HOC, better-controlled reading says, labelled as post-hoc
Under the coverability-matched null that the pre-registration did not specify, none of the three classes is enriched in A73051: ALL_specific 0.541x its matched null (two-sided p 0.0014, 0.0002 - significantly DEPLETED), AML_specific 0.936x (ns, p 0.46 and 0.96), Oncoprotein_Target 0.667x (p 0.001, 0.001 - significantly DEPLETED). That is pre-registered branch NO-DESTINATION condition 1 - nothing is concentrated anywhere - reached by a null I did not pre-declare.
Both readings are reported and neither replaces the other. The pre-registered verdict is UNINFORMATIVE; the post-hoc verdict is NO-DESTINATION; and the two agree that GMP-DESTINATION does not fire.
| branch fired | NEITHER/UNINFORMATIVE |
|---|---|
| post hoc branch | NO-DESTINATION (condition 1), under a coverability-matched null |
| script | analyses/q148-what-the-hoxa-negative-blast-occupies/analyze148.py |
| status | run |
| result summary | NEITHER/UNINFORMATIVE on U1, U4, U5, U7 |
gates fired
[ "U1_G1_fails", "U4_A73051_inside_HOXApos_AML_range", "U5_institution_separation", "U7_A73051_globally_low_oncoprotein" ]
gates not fired
[ "U2_class_under_20_promoters", "U3_neg_control_not_below_1_in_A73051", "U6_depth" ]
groups that cannot carry a statistic
{
"KMT2A-AFF1": 3,
"KMT2A-ENL": 4,
"KMT2A-AFDN": 3,
"KMT2A-AF10": 2,
"KMT2A-AF9": 1,
"KMT2A-EPS15": 2,
"note": "depth-ADEQUATE untreated non-circular; every one of the six"
}measurements — the numbers behind the chart
[
{
"quantity": "AML_over_ALL",
"subject": "A73051 (KMT2A-AF10, HOXA-negative, Fred Hutch)",
"value": 3.493,
"units": "ratio",
"n": 2,
"universe": "40 AML_specific over 37 ALL_specific promoters, KMT2A-N1 and N2",
"null": null,
"note": "inside the 1.263-7.189 range of the 5 HOXA-positive Fred Hutch AMLs, rank 3 of 6; n=1 vs n=5 so no p"
},
{
"quantity": "spearman_rho",
"subject": "AML/ALL against promoter double-zero probability p_dz",
"value": 0.268,
"units": null,
"n": 15,
"universe": "depth-ADEQUATE, non-circular, untreated GSE252378 samples",
"null": 0,
"note": "two-sided p = 0.334; the pre-declared depth gate at |rho| >= 0.6 does not fire"
},
{
"quantity": "mannwhitney_p_two_sided",
"subject": "AML/ALL, Fred Hutch (n=6) vs Texas Children's (n=6)",
"value": 0.0087,
"units": null,
"n": 12,
"universe": "depth-ADEQUATE, non-circular, untreated GSE252378 samples",
"null": "smallest attainable two-sided p at 6 vs 6 is 0.0022",
"note": "institution and lineage are the SAME split in this subset and cannot be separated"
}
]- A pre-registered UNINFORMATIVE is the most quotable kind of nothing. It should not be read as 'the GMP hypothesis is dead' - it is 'this deposit cannot answer it', which is a statement about the deposit.
- Would a deposit with more than one depth-adequate HOXA-negative non-AFF1 sample change the verdict? GSE252378 has exactly one and every branch here is therefore an n = 1 statement dressed in cohort context.
- Should a pre-registration be allowed seven uninformative conditions? Four fired here, and two of them were about my own instrument rather than about the data.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
Of the seven pre-declared uninformative conditions, U1, U4, U5 and U7 fired and U2, U3 and U6 did not; GMP-DESTINATION and NO-DESTINATION both failed to fire under the pre-registered gates.
analyses/q148-what-the-hoxa-negative-blast-occupies - PREREGISTRATION.md sections 5-7 and results148.json 'uninformative_gates' and 'branch' · Q148 (Oncofusion Knowledge Graph) · this repository 2026What this evidence cannot show
The branch logic is mine and its thresholds (mean >= 1.5, p < 0.05, mean/median agreement) were chosen before the data were seen but not externally calibrated. Two of the four firing gates - U1 and U7 - are about the instrument, not about the biology.
Evidence that argues against this (1)
- A pre-registration with seven uninformative conditions is easier to render uninformative than to pass, and an author who writes many escape hatches can always find one open. The defence offered here is that the post-hoc, better-controlled analysis reaches NO-DESTINATION rather than GMP-DESTINATION, so the UNINFORMATIVE verdict is not hiding a positive. Q148 (Oncofusion Knowledge Graph) 2026
- E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
A73051's AML/ALL ratio of 3.493 is inside the 1.263-7.189 range of the five depth-adequate HOXA-positive Fred Hutch paediatric AMLs, rank 3 of 6.
analyses/q148-what-the-hoxa-negative-blast-occupies - results148.json 'G4_lineage_confound' · Q148 (Oncofusion Knowledge Graph) · this repository 2026What this evidence cannot show
n = 1 versus n = 5. This is a point comparison and no statistic is attached to it, by design. Institution and lineage are the same split in this subset, so 'Fred Hutch paediatric AML' is one stratum and not two controls.
Evidence that argues against this (1)
- Being inside a range of five is weak evidence of belonging to it. The stronger statement is the N-over-C one: A73051 shows no fusion-specific excess at any class, which the ratio comparison cannot show on its own. Q148 (Oncofusion Knowledge Graph) 2026
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The GSE252378 sample set the deposit's paper describes contains 36 profiled MLL-rearranged samples including three lineage-switching events; the 27 KMT2A-fusion patient samples reachable here are 24 independent patients.
MLL oncoprotein levels influence leukemia lineage identities. · Janssens DH, Duran M, Otto DJ, Wu W, Xu Y, Kirkey D, Mullighan CG, Yi JS, Meshinchi S, Sarthy JF, Ahmad K, Henikoff S · Nature Communications 15:9341 2024What this evidence cannot show
The 36 in the paper includes cell lines and samples outside the 27-sample KMT2A-fusion patient arm used here; the three pairs (p279/p318, 148752/152985, R3595/R3652) are each one patient sampled twice, per the GEO descriptions' own 'paired with' wording.
Evidence that argues against this (1)
- GEO's 'treated with the Menin inhibitor' on 148752 and 152985 is a CASE-level label. 148752 is the untreated diagnosis aliquot; only 152985 is drug-exposed. Counting both as treated, or both as untreated, changes n by one in either direction and has produced an n=6/n=5 inconsistency in this corpus before. Janssens DH et al. 2024
confirms NEITHER at interval scale, 0 of 24 with controls at 19 and 10
the branch outcome of the measurement in that entry
- depends on →A73051 is the one HOXA-negative non-AFF1 sample in GSE252378, and it is IRX1-negative too
A73051's HOXA/IRX1 calls and depth grade are carried from q147 unchanged
- supports →A group of four cannot support a correlation at any effect size: name it, do not compute it
applies the same rule: all six partner groups here are named as unable to carry a statistic
depth grades reused unchanged from the q147 gate
this is gate U5, one of the four that fired
gate U1 fired on a test the reagent was never built to pass