IRX2 is not fusion-occupied at promoter scale in GSE252378: 0 of 27 samples, against a null of 2.73

My dispatch brief predicted the trap of this beat would be strand: IRX1 is plus-strand with its TSS at the lower coordinate, IRX2 is minus-strand with its TSS at the higher coordinate (chr5:2,751,677), so copying q147's IRX1 recipe would place the IRX2 window ~16 kb away at the wrong end of the gene.

The strand fact is right and I verified it independently (UCSC hg38 refGene, fetched 2026-09-24: NM_001134222 chr5:2,745,844-2,751,677 strand -; MANE NM_033267 chr5:2,746,164-2,751,677 strand -; NCBI esummary GeneID 153572 returns chrstart 2751676 > chrstop 2735634). But the trap does not exist, because C:\Temp\q145irx1\promoters.tsv - the table q145 built and q147 copied its IRX1_promoter_2kb out of - is strand-aware and already contains IRX2:

IRX2  chr5  2751677  2750677  2752677  -  NM_001134222
IRX1  chr5  3595831  3594831  3596831  +  NM_024337

So IRX2's correct 2 kb promoter has been measured in all 133 libraries since q145, at row 14,352 of promoter_matrix.tsv, and nobody had looked at it. There was no interval to add and no bigWig to re-read. 1

q147's data.floor_rule, quoted: ">=2.0x in both KMT2A-N antibodies AND above the 95th percentile of eight width-matched cannot-be-occupied loci". Reused unchanged in both clauses.

The one change: q147 width-matched its floor to the 5,572 bp IRX1 gene body. This beat measures a 2,000 bp promoter, so the floor is the 2 kb promoter of the same eight loci - ALB, RHO, MYH7, MYOD1, NEUROD1, KRT14, GFAP, INS - read from the same matrix, where every row is 2 kb by construction. Same eight loci, width matched to the measurement.

Input assertion. promoter_matrix.tsv holds RAW mean bigWig coverage, not shares. Asserted at load in the direction that would break the gate: column means span 0.9177-18.7951 and 0 of 133 is within 1% of 1.0. Each column is then divided by its own mean. Every quantity reported is a ratio. 1

Instrument consistency check, and what it is NOT. q147's regions147.py computed IRX1_promoter_2kb for 79 libraries by downloading each bigWig and averaging chr5:3,594,831-3,596,831 over promoter_sum/19404; this beat reads matrix row IRX1 and divides by the column mean. They agree to a median relative difference of 2.2e-6 across all 79. That is agreement between two expressions of the same arithmetic over the same interval and the same denominator, so it confirms the interval and the normaliser and nothing more. It is a derivation check, not an independent measurement, and it is reported here as one. 1

Within-sample share of the IRX2 2 kb promoter in the two fusion-side antibodies. T = menin-inhibitor treated.

Denominators. 27 samples; 24 independent patients (pairs p279/p318, 148752/152985, R3595/R3652); AFF1 8, non-AFF1 19 - or 18 in q147's style, which drops one of the ENL lineage-switch pair. 25 untreated: the committed manifest records treated: true for both 148752 and 152985, and both desc strings say "treated with the Menin inhibitor". They differ by lineage phase (infant B-ALL vs infant AML in a lineage switch), not by treatment - so neither is an untreated control. 1

Groups that cannot carry a statistic, named rather than computed: KMT2A-EPS15 n=2 and both samples are the same patient (n_independent = 1); KMT2A-AF10 n=3; KMT2A-AF9 n=3.

Positive control at the same scale. HOXA9's 2 kb promoter clears the identical rule in 17 of the 18 non-AFF1 samples q147 scored, exactly reproducing q147's 17 of 18 at 17.9 kb scale, and in 21 of 27 overall. MEIS1 clears it in 24 of 27. So a 2 kb window in this assay can report occupancy; my pre-registered expectation that this gate would fail was wrong. 1

Mappability control. 37 of 133 libraries carry non-zero raw coverage at the IRX2 window and the share reaches 3.81x, so the interval is not a dark region. A zero in an unmappable window is not a measurement; this one is. 1

The null, which is the part that matters most. Scoring all 19,404 refGene promoters against the identical rule in each of the 27 samples: the rule calls a median of 10.75% of promoters occupied per sample (range 1.86%-14.01%; the low tail is depth-driven). An arbitrary promoter is therefore expected to be occupied in 2.73 of 27 samples. IRX2 at 0 is below that; HOXA9 at 21 and MEIS1 at 24 are far above. A gate that admits one promoter in ten is not stringent, and no entry in this graph previously recorded that. 2

p454, KMT2A::AFF1, untreated. IRX2 promoter share 3.8119 in KMT2A-N1 (library floor 0.015) and 1.8979 in KMT2A-N2 (floor 0.2022). Both are above the floor; N2 is 0.102 below the 2.0x gate, i.e. 5.1% of the threshold. IgG is exactly 0.0. The C-side antibodies read 1.595 and 1.106. In the same sample IRX1 reads 4.592 / 2.359 and is called occupied.

A threshold sweep from 0.5x to 4.0x shows the count go 0 -> 1 of 27 between 1.90x and 1.89x, and the one sample gained is p454. At every threshold in that range the count of IRX2-occupied non-AFF1 samples is 0. So the magnitude of this result is threshold-dependent and the partner-restriction of it is not. 3

This is reported as branch IRX2-IS-NOT-OCCUPIED because that is what the pre-registered rule returns, and the boundary is declared rather than resolved by moving the rule. 3

• Scale. This is a 2 kb promoter result. q147's IRX1 call was at gene-body scale (5,572 bp) and q145's headline rho was too. A gene-body IRX2 measurement does not exist and would need the bigWigs re-read; the interval would be chr5:2,745,844-2,751,677 (NM_001134222, 5,833 bp) or 2,746,164-2,751,677 (MANE, 5,513 bp). See exp-q151-rescore-irx2-at-gene-body-scale-in-gse252378.
• Gene identity. IRX2's TSS is 215 bp from IRX2-DT's on the opposite strand, so 1,785 bp (89.2%) of the 2 kb window is shared with the divergent transcript. Signal in this window cannot be assigned to IRX2 rather than IRX2-DT. See concept-q151-a-divergent-promoter-pair-shares-its-window-and-a-2-kb-call-cannot-name-its-gene.
• Locus independence. The test for "one 844 kb domain rather than two genes" is underpowered: only 3 refGene promoters in the matrix lie in the 1.2 Mb spanning IRX2 and IRX1, and two of the three (IRX2, IRX2-DT/C5orf38) are the same divergent promoter. The flat profile between them is not evidence of independence, it is evidence of a gene desert. 2
• Cohorts. GSE252378 (occupancy) and GSE68720 (expression) are different patients. concept-an-expression-subgroup-is-not-an-occupancy-subgroup applies; no expression conclusion is transferred here.
• Depth. This deposit has no spike-in (171/171 GSMs library_strategy = OTHER), so no claim of the form "sample X has little oncoprotein anywhere" is made.