A width-matched panel of loci that cannot be occupied is a usable per-locus floor; IgG in this deposit is not
To know whether a protein is really sitting on a gene, you need to know what 'nothing' looks like. The usual control antibody is useless here, so we measured eight genes that a leukaemia cell cannot possibly be using - liver, retina, muscle, skin - and used those as the zero line.
The same thing, for a scientist
q145 established that IgG is nearly worthless as a per-locus floor in GSE252378: the IgG libraries are **57-94% zero at promoters**, and EGR1 fails its own IgG control (N/IgG = 0.50 in p247). Measured here across all 25 IgG libraries in the 27-sample panel, the situation is worse than 'noisy': **the 95th-percentile IgG share at eight loci that cannot be occupied in a leukaemia ranges from exactly 0.000 (9 of 25 libraries) to 13.12x an average promoter (TB051939), with 28% of IgG libraries above 2.0x** [E1]. An IgG floor drawn from those libraries would be simultaneously **absent** in a third of samples and **higher than the signal** in another quarter. The replacement, declared in the pre-registration before any value was read: **ALB (liver), RHO (retina), MYH7 (cardiac), MYOD1 (skeletal muscle), NEUROD1 (neuroendocrine), KRT14 (skin), GFAP (glia), INS (beta cell)** - eight loci whose promoters a leukaemic blast cannot be using - scored in the *same* KMT2A-N library as the signal, with the **95th percentile** as that library's floor. Measured: median floor **0.206x**, 90th percentile **0.745x**, **only 1 of 54 KMT2A-N libraries above 2.0x** (A80956, itself depth-inadequate) [E1]. At gene-body scale, using **width-matched 5,572 bp windows** so the comparison is like-for-like with the IRX1 gene, the floor is median 0.223x and **never exceeds 1.586x** [E1]. So the floor is separable from the signal by an order of magnitude, and the pre-registered U3 uninformative clause (floor above 2.0x in more than half of libraries) does not fire at 0.019.
Why it has to be width-matched, and why the panel has to be in the same library
Width-matched. Mean coverage over a window is not scale-free: a 2 kb promoter window and a 5.6 kb gene-body window have different variances and different chances of containing a stray fragment. Comparing IRX1's 5,572 bp gene body against a 2 kb control window would make the control look quieter than it is. The eight negative loci are therefore scored as 5,572 bp windows - exactly the IRX1 gene width - centred on their own TSS.
Same library. There is no spike-in in this deposit, so between-library comparison of absolute coverage is not licensed at all. The floor is a within-library quantity and the occupancy call is a comparison of two numbers from the same library - which is the only form in which the unknown per-library scale cancels.
Eight, not one. A single non-expressed locus can be a background hotspot, which is exactly what q145 found for EGR1. Eight gives a distribution, and the 95th percentile is chosen so that one hotspot among the eight cannot lower the bar.
The general rule, which is not about IgG
A rank statistic is scale-free and cannot tell you whether anything was there. q145 was nearly caught out by its own flat control CSPG4 reaching Spearman -0.619 while every CSPG4 value was at or below 0.09x an average promoter. Reproduced here at 2 kb promoter scale: CSPG4 gives rho -0.619 in AFF1 and +0.718 in ENL, with max shares of 0.557x and 0.345x respectively. At n = 5-8 the eight cannot-be-occupied loci reach |rho| up to 0.600 1.
So: any locus placed on a correlation must independently clear an absolute floor, and the noise band of |rho| that unoccupied loci reach at your n should be printed beside your result. It is the difference between a correlation that means something and one that is ordering background.
| floor statistic | 95th percentile of the eight within the same library |
|---|---|
| igg libraries | 25 |
| igg floor median | 0.808 |
| igg floor max | 13.121 |
| igg floor frac above 2 | 0.28 |
| igg floor exactly zero | 9 |
| kmt2an libraries | 54 |
| kmt2an floor median | 0.206 |
| kmt2an floor p90 | 0.745 |
| kmt2an floor max | 2.994 |
| kmt2an floor frac above 2 | 0.019 |
| gene scale window bp | 5572 |
| gene scale floor median | 0.223 |
| gene scale floor max | 1.586 |
| cspg4 rho aff1 | -0.619 |
| cspg4 rho enl | 0.718 |
| cspg4 max share | 0.557 |
negative control loci
[ "ALB", "RHO", "MYH7", "MYOD1", "NEUROD1", "KRT14", "GFAP", "INS" ]
control locus rho band
[ -0.571, 0.6 ]
analyses
[ "ana-q147-irx1-fusion-occupancy-is-confined-to-kmt2a-aff1" ]
- Is the eight-locus panel mappable end to end under repeat-masked hg38 with bowtie2? A locus that reads zero because reads cannot be placed there deflates the floor, and that was not separately checked.
- Would a matched-GC, matched-mappability random promoter panel be a better floor than a hand-picked tissue-specific one? It would be defensible against the objection that these eight are unusually quiet for a reason other than not being occupied.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
Across the 27-sample GSE252378 panel: the 95th-percentile share at eight cannot-be-occupied loci is 0.000-13.12x in the 25 IgG libraries (median 0.808x, 28% above 2.0x, 9 of 25 exactly 0.000), against 0.010-2.994x in the 54 KMT2A-N libraries (median 0.206x, 90th percentile 0.745x, 1 of 54 above 2.0x). At gene-body scale with width-matched 5,572 bp windows the KMT2A-N floor is median 0.223x and never exceeds 1.586x. The same eight loci reach |Spearman| up to 0.600 against HOXA at n = 5-6 and up to 0.571 at n = 8.
q147 re-analysis of GSE252378: HOXA and IRX1 oncoprotein occupancy across all six KMT2A fusion partners, 27 samples, 133 libraries (analyses/q147-hoxa-irx1-across-partners) · Iqbal Lab oncofusion knowledge graph, beat q147 · internal analysis over deposited GSE252378 bigWigs 2026What this evidence cannot show
The panel's validity rests on the biological premise that ALB, RHO, MYH7, MYOD1, NEUROD1, KRT14, GFAP and INS promoters are not used in a leukaemic blast. That premise is strong but not measured here; it is not certain for GFAP or NEUROD1, which have reported expression in some non-neural contexts. Mappability was not separately assessed for the eight, so a locus reading zero because it is unmappable would deflate the floor - the 95th percentile rather than the median is used partly for that reason. Every number is a WITHIN-SAMPLE share: an interval's mean coverage divided by that same library's mean coverage over all 19,404 hg38 RefSeq promoters, so 1.0 = an average promoter in that library. GSE252378 has no spike-in, so no between-library absolute statement is licensed; the only cross-sample quantity used is a ratio of two loci measured inside one library, where the unknown per-library scale cancels.
Evidence that argues against this (1)
- The deposit paper uses IgG as its own control and reports oncoprotein-target calls against it, so this entry is a methodological disagreement with the source study's choice of floor, not a report of its result. The paper's N-over-C ratio criterion is a different and defensible floor that does not depend on IgG depth at all. Janssens DH et al. 2024
the floor-referenced ratio is NOT coverage-invariant: onco/neg tracks raw coverage at +0.631
the enrichment axis, not depth, is what the floor-referenced ratio fails to cancel
uses IgG anyway, as an antibody-background arm rather than as a floor, and states why that is not an accessibility control
the eight-locus non-expressed floor is reused as the second denominator
a non-expressed-locus panel worked for occupancy; the transcript analogue of it did NOT - three of its ten genes sit above IRX1's cohort median
- ← depends onIRX2 is not fusion-occupied at promoter scale in GSE252378: 0 of 27 samples, against a null of 2.73
the eight cannot-be-occupied loci, width-matched here to 2 kb rather than 5,572 bp
in A80956 the 8-locus floor is 4.005x, above the 2.0x gate, so the floor rather than the multiple binds
usable at promoter scale, where floors reach 4.005; inert at gene-body scale, where the floor never exceeds 1.648 against a 2.0x gate
re-derives that decision from the denominator side: IgG promoter density is 8.45-fold below KMT2A-N1's