The occupancy rule this project has used since q145 calls a tenth of ALL promoters occupied, and no entry using it records that
also called q151 null for the q147 rule · the 10.75% pass rate
The test this project uses to decide whether a gene is 'switched on' by the fusion protein says yes for about one gene in ten, chosen at random. That does not make the test wrong, but it means a single gene passing it is much less surprising than it looks, and nobody had written that number down.
The same thing, for a scientist
🔴 **NULL SUPPLIED, AND THE OBSERVED VALUE IS BELOW IT (cycle 141).** This entry reported **332 of 19,404 promoters (1.71%)** showing the AFF1-only pattern and left `data.measurements[].null` as `null`. **The null is 426.4** - a label permutation over the fixed occupancy matrix redrawing only which 8 of 27 samples form the arm (10,000 draws: median 423, 5th-95th 287-579, p_lower 0.1517). Q152 computed 425 and the coordinator re-derived 426.4 independently. **So the 332 is not an enrichment and cannot support a partner-specificity reading.** Everything else in this entry re-derives to the digit. See `ana-the-aff1-only-promoter-pattern-is-below-its-own-null-and-the-figure-the-irx1-line-rests-on-is-not-an-enrichment`. --- The rule - within-sample share >= 2.0x an average promoter in BOTH KMT2A-N antibodies AND above the 95th percentile of eight width-matched cannot-be-occupied loci - was scored against **all 19,404 refGene promoters** in each of the 27 GSE252378 samples. It passes a **median of 10.75% of promoters per sample** (range 1.86% in A80956 to 14.01% in p247), so an arbitrary promoter is expected to be called occupied in **2.73 of 27** samples [E1]. Against that reference, HOXA9 (21 of 27) and MEIS1 (24 of 27) are far above; **IRX1 at promoter scale is 2 of 27, at or below the null**; IRX2 is 0 of 27, below it [E1]. Restricting to the pattern the whole IRX story turns on - occupied in at least one of the 8 AFF1 samples and in none of the 19 non-AFF1 - **332 of 19,404 promoters (1.71%, and 5.49% of the 6,049 occupied anywhere) show it** [E1]. QUEEN_BRIEF rule 84 is explicit that a proportion is not a result until its null is written down, and **no entry in this corpus that applies this rule records one**. This is the reference, not a defect report: a fusion oncoprotein that genuinely occupies thousands of promoters would produce exactly this rate.
What was measured
analyses/q151-is-irx2-fusion-occupied-in-gse252378/null151.py builds a boolean 19,404 x 27 occupancy matrix by applying the identical rule - the same 2.0x gate, the same two KMT2A-N antibodies, the same eight-locus floor computed per library - to every promoter rather than to a chosen one.
| quantity | value |
|---|---|
| promoters scored | 19,404 |
| samples | 27 |
| median pass rate per sample | 10.75% |
| range | 1.86% (A80956) - 14.01% (p247) |
| expected samples occupied, arbitrary promoter | 2.73 of 27 |
| promoters occupied in 0 samples | 13,355 |
| promoters occupied in exactly 1 sample | 1,254 |
| promoters occupied somewhere | 6,049 |
The low tail is not a stringent gate. A80956's 1.86% comes from an unusually high floor - its KMT2A-N1 eight-locus maximum is 4.005x, above the 2.0x gate itself - so in that library clause 2 rather than clause 1 is binding. Reporting the range without that explanation would invite the wrong reading. 1
What it does to the loci this project has been arguing about
| locus | samples occupied of 27 | vs null 2.73 |
|---|---|---|
| MEIS1 promoter | 24 | far above |
| HOXA9 promoter | 21 | far above |
| PROM1 promoter | 11 | above |
| IRX1 promoter | 2 | at or below |
| IRX2 promoter | 0 | below |
| IRX2-DT (C5orf38) promoter | 0 | below |
This deflates a claim of mine and constrains one of q147's, and the two are different sizes.
Mine: this beat's IRX2 zero is strengthened by the null, because 0 is below 2.73 - a zero under a permissive rule is more informative than a zero under a strict one.
q147's: IRX1 at promoter scale is 2 of 27, which is not distinguishable from an arbitrary promoter. That is NOT a refutation of q147, whose 3-of-8 was measured at gene-body scale (5,572 bp). No 19,404-region gene-body matrix exists on disk, so q147's own null cannot be computed from what is available and is not claimed here to be 10.75%. The narrow, defensible statement is: a promoter-scale replication of q147's IRX1 occupancy result does not exist in this deposit, and q147's gene-body result is currently reported without a null. 12
The base rate for 'AFF1-restricted', which is the shape of the whole IRX argument
Of 19,404 promoters, 332 (1.71%) are occupied in at least one of the 8 KMT2A::AFF1 samples and in none of the 19 non-AFF1 samples. That is 5.49% of the 6,049 promoters occupied anywhere.
So the pattern is uncommon but not rare, and it is shared with 331 other loci. "AFF1-restricted at promoter scale" is a description, not by itself evidence of a fusion-specific mechanism; what would make a locus interesting is the effect size and the reproducibility across the AFF1 arm, not the restriction per se.
For scale at the single-sample end: 56 promoters (0.29%) are occupied in p454 and in no other sample. IRX2 does not reach that set - it fails the rule in p454 by 0.102 in N2 - but 56 is the number a p454-only IRX2 call would have had to be read against. 1
(An earlier draft of this analysis's own output field called the AFF1-only pattern "COMMON by construction". That was written before the number was read and it is wrong: 5.49% is not common. The correction is recorded inside results151_null.json under AFF1_restriction_base_rate.CORRECTION.)
What this is not
- Not a claim that the rule is broken. A fusion oncoprotein with thousands of genuine promoter targets produces this rate legitimately. The number is the reference a single-locus call must be read against, and that is all.
- Not transferable to gene-body scale. Stated twice because it is the easiest thing to quote wrongly out of this entry.
- Not a test of the eight-locus floor. The floor and the 2.0x gate are entangled here; A80956 shows that which clause binds varies by library, and this analysis does not decompose them per locus.
- Not corrected for depth. This deposit has no spike-in, and pass rate correlates with library depth in the obvious direction, which is part of why the range spans 1.86-14.01%.
| script | analyses/q151-is-irx2-fusion-occupied-in-gse252378/null151.py |
|---|---|
| status | run |
| result summary | the q147 rule passes a median 10.75% of all promoters per sample at promoter scale; expected 2.73 of 27 samples for an arbitrary promoter |
| rule scored | >=2.0x in both KMT2A-N antibodies AND above the per-library maximum of eight width-matched cannot-be-occupied 2 kb promoters |
| n promoters | 19404 |
| n samples | 27 |
| pass rate PCT median | 10.75 |
| pass rate PCT min | 1.86 |
| pass rate PCT min sample | A80956 (its KMT2A-N1 floor is 4.005x, above the 2.0x gate, so clause 2 binds) |
| pass rate PCT max | 14.01 |
| pass rate PCT max sample | p247 |
| expected samples for arbitrary promoter | 2.73 |
| n promoters occupied in zero samples | 13355 |
| n promoters occupied in exactly one sample | 1254 |
| n promoters occupied anywhere | 6049 |
| n promoters AFF1 only | 332 |
| PCT of occupied that are AFF1 only | 5.49 |
| n promoters p454 only | 56 |
| scope limit | PROMOTER scale only. No 19,404-region gene-body matrix exists on disk, so q147's gene-body null cannot be computed and is NOT claimed to be the same. |
inputs
[ "C:/Temp/q147part/promoter_matrix.tsv" ]
observed counts — the numbers behind the chart
{
"MEIS1": 24,
"HOXA9": 21,
"PROM1": 11,
"IRX1": 2,
"IRX2": 0,
"C5orf38_IRX2DT": 0
}measurements
[
{
"quantity": "rule_pass_rate",
"subject": "an arbitrary refGene promoter under the q147 two-antibody 2.0x plus floor rule",
"value": 10.75,
"units": "percent of promoters per sample (median)",
"n": 27,
"universe": "19,404 refGene promoters x 27 GSE252378 KMT2A-fusion samples, 2 kb windows, within-sample shares",
"null": null,
"note": "this IS the null; range 1.86-14.01%, expectation 2.73 of 27 samples for an arbitrary promoter"
},
{
"quantity": "proportion_AFF1_restricted",
"subject": "promoters occupied in >=1 of 8 AFF1 and 0 of 19 non-AFF1 samples",
"value": 0.0549,
"units": null,
"n": 6049,
"universe": "the 6,049 of 19,404 promoters occupied in at least one sample under the same rule",
"null": 426.4,
"note": "332 of 6,049; as a share of all promoters it is 1.71%. The AFF1-restricted pattern is shared with 331 other loci.",
"null_source": "label-permutation null over the fixed occupancy matrix, 10,000 draws; Q152 computed 425 and the coordinator re-derived 426.4 (5th-95th 287-579, p_lower 0.1517). The observed 332 is BELOW this null."
}
]- Any therapeutic rationale built on 'the fusion occupies gene X' in this deposit inherits a one-in-ten false-positive reference rate at promoter scale. A target nominated on a single-sample occupancy call is the most likely kind of target to fail replication.
- What is the pass rate of the same rule at GENE-BODY scale? It cannot be computed from what is on disk; it needs a genome-wide gene-body matrix, which would mean re-reading 54 bigWigs.
- How much of the 1.86-14.01% spread is library depth? A depth-matched version of this null would say whether a low-pass-rate sample is a strict library or a shallow one - and this deposit has no spike-in, so that may not be separable.
- Should the 2.0x gate be re-set at promoter scale, e.g. to a per-library percentile rather than a fixed multiple? Re-setting it would change every call in q145, q147 and this beat, so it is a coordinator decision and not one this entry makes.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Applied to all 19,404 refGene promoters in each of the 27 GSE252378 samples, the two-antibody 2.0x plus width-matched-floor occupancy rule passes a median of 10.75% of promoters per sample, giving an expectation of 2.73 of 27 samples for an arbitrary promoter; 332 promoters (1.71%) show the AFF1-only pattern and 56 (0.29%) are occupied in p454 alone.
q151 null for the q147 occupancy rule over 19,404 refGene promoters x 27 GSE252378 samples · Iqbal Lab oncofusion knowledge graph, beat Q151 · this project 2026What this evidence cannot show
Computed by this project, not published. Specific to 2 kb promoter windows and to this deposit's depth distribution. The 19,404 promoters are one refGene transcript per symbol, deduplicated by first occurrence, so alternative promoters are not represented and the denominator is a gene set rather than a promoter set. No depth matching was applied, and the pass rate correlates with library depth, which is most of the 1.86-14.01% spread.
Evidence that argues against this (1)
- A high pass rate is what a promiscuous chromatin-bound oncoprotein should produce. KMT2A::AFF1 is reported to occupy thousands of promoters genome-wide, so this rate may be biology rather than a loose gate - in which case the right reading is that single-locus occupancy calls in this system are intrinsically low-information, not that the rule is mis-set. NCBI Gene Expression Omnibus 2024
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
q147 reports IRX1 fusion occupancy in 3 of 8 KMT2A::AFF1 samples at gene-body scale (5,572 bp) under this rule and states no null for it.
ana-q147-irx1-fusion-occupancy-is-confined-to-kmt2a-aff1 (this graph) · Iqbal Lab oncofusion knowledge graph, beat Q147 · this project 2026What this evidence cannot show
This is a statement about a sibling entry's contents, read off disk on 2026-09-24, not a literature claim. expected_absent_terms names 'null' because the absence of a recorded null IS the claim. q147's result is not contested here; only the missing reference is noted.
Evidence that argues against this (1)
- q147 does supply controls of a different kind - the eight width-matched cannot-be-occupied loci and the HOXA positive control in 17 of 18 samples - so it is not uncontrolled. What it lacks specifically is a genome-wide pass rate for its own rule, which is a different and weaker omission than having no control at all. Iqbal Lab oncofusion knowledge graph et al. 2026
supplies the null that entry left empty: 332 against a permutation mean of 426.4, so it is not an enrichment
the same rule: 10.75% of promoters pass, and 44% of the calls are at windows belonging to two genes
- supports →IRX2 is not fusion-occupied at promoter scale in GSE252378: 0 of 27 samples, against a null of 2.73
supplies the null the 0-of-27 is read against: 2.73 of 27 expected for an arbitrary promoter
- related to →IRX1 fusion occupancy is confined to KMT2A::AFF1: 0 of 18 samples across five other partners
q147's rule has no recorded null; at promoter scale it passes a median 10.75% of all promoters
in A80956 the 8-locus floor is 4.005x, above the 2.0x gate, so the floor rather than the multiple binds
scores the null q151 calibrated; every figure re-derived to the digit
- ← refutesThe 332-promoter AFF1-only set is BELOW the count a random eight-sample arm produces (null 425)
supplies the missing null for q151's AFF1-restricted share; the observed 332 is BELOW it
audits the per-library floor that q151's rule uses
same rule, a second limitation: the rule that passes a median 10.75% of promoters also cannot separate two overlapping ones
the same rule's permissiveness is what makes the promoter-activity explanation of the 2.03-fold plausible
supplies the missing gene-body half of the same calibration: 0.608% against that entry's 10.75%