The 2.03-fold over-representation of shared windows among fusion-occupied promoters is generic promoter activity, not a fusion preference - and if it is not, the rule is finding something different from what it claims
also called bidirectional promoter occupancy confound · shared window enrichment explanation
Windows that cover two genes at once turn up in nearly half the positive results, though they are only a fifth of all windows. The boring explanation is that those spots are busy in every cell for reasons that have nothing to do with leukaemia. If the boring explanation is wrong, then the test is measuring something other than what it is named after.
The same thing, for a scientist
Measured: **43.87% of 53,180 occupancy calls** made by the q147 rule across 27 GSE252378 samples fall on promoter windows that contain another gene's TSS within 1 kb, against a base rate of **21.60%** — a 2.03-fold enrichment, in **27 of 27 samples**, range 41.3–50.8% [E1]. **Two explanations, and they have opposite consequences for everything built on this rule.** (H-A, the confound) Divergent pairs share one CpG-island bidirectional promoter; such promoters are nucleosome-depleted and transcribed in most cells, so they carry high coverage in any chromatin assay, and a rule scoring *within-sample share relative to the average promoter* will preferentially clear them. Under H-A the enrichment is a fact about chromatin and the rule is partly a promoter-activity detector — which is consistent with its already-recorded permissiveness of a median **10.75% of all 19,404 promoters called occupied per sample** [E2]. (H-B, the biology) KMT2A fusions preferentially load at bidirectional/CpG-island promoters over and above their activity, e.g. through the CXXC domain's unmethylated-CpG address. Under H-B the enrichment is a real property of fusion targeting and would be a finding. **Prior: H-A at roughly 0.8, H-B at roughly 0.2**, because H-A requires nothing beyond known chromatin biology and H-B requires an effect beyond activity that nothing in this corpus has measured. **Both are testable from data already on disk** [E3]. **And the hypothesis is stated because it cuts against this beat's own convenience:** if H-A holds, then the correct reading of the q147 rule's positive calls is weaker than any beat using it has assumed, and that weakening applies to the clean 56% as much as to the shared 44%.
The discriminating test, which needs no new data
Match, do not adjust. For each shared-window promoter called occupied, find an unshared promoter matched on its activity proxy, and ask whether the occupancy rate is still higher at shared windows. Three activity proxies are available on disk without a single new download:
- The KMT2A-C arms. Every sample has
KMT2Ac_1andKMT2Ac_2libraries. If shared windows are simply busier chromatin, they will be enriched in the C-terminal arms too — and the C-terminal fragment is not the oncoprotein. This is the cleanest available control and it is already in the matrix. - The IgG arms. Same logic, weaker signal, but a shared-window enrichment in IgG is pure accessibility.
- The per-promoter mean across all 133 libraries, which is a within-deposit activity index needing nothing but the matrix.
Pre-declared reading: if the shared-window enrichment in the KMT2A-C arms is statistically indistinguishable from the 43.87% in the KMT2A-N arms, H-A is supported and H-B is not needed. If the N arms are enriched above the C arms, the residual is the part H-B would have to explain, and its size is the effect size to report.
A third possibility that must be pre-declared: both arms may be enriched and differ, in which case the answer is quantitative and neither hypothesis wins outright. The output is then the residual with an interval, not a verdict.
Why this matters beyond bookkeeping
This project's promoter-scale occupancy rule is load-bearing. It produced IRX1 occupied in 2 of 27, IRX2 in 0 of 27, HOXA9 in 21 of 27, MEIS1 in 24 of 27, and the 2.73-of-27 null that makes the IRX2 zero interpretable. Every one of those numbers means something different under H-A than under H-B.
Under H-A, HOXA9 21/27 is partly a statement that HOXA9's promoter is active in KMT2A-rearranged blasts — which is true and well known and not what the measurement was supposed to add. Under H-B it is a statement about fusion targeting.
The KMT2A-C control is the existing discriminator for exactly this and, as far as this beat could establish, it has been used as a within-locus comparison and never as a genome-wide activity control. That is the cheap missing analysis.
| statement | The 2.03-fold over-representation of shared (divergent) promoter windows among fusion-occupied calls in GSE252378 is caused by generic promoter activity at bidirectional CpG-island promoters (H-A), not by a preference of the KMT2A fusion for such promoters over and above their activity (H-B). |
|---|---|
| origin | data-pattern |
| prior probability | 0.8 |
predictions
[ "H-A predicts the same shared-window enrichment in the KMT2A-C arms as in the KMT2A-N arms, within sampling error.", "H-A predicts shared windows have a higher per-promoter mean across all 133 libraries, including IgG, than unshared windows.", "H-B predicts a residual enrichment in the KMT2A-N arms after matching on any activity proxy, and predicts that residual should be larger at CpG-island promoters given the CXXC domain's unmethylated-CpG address.", "Both predict the enrichment is insensitive to fusion partner, since neither mechanism is partner-specific; a partner-dependent enrichment would refute both as stated." ]
analyses
[ "ana-q153-43-9-percent-of-every-occupancy-call-this-rule-makes-is-at-a-window-it-cannot-attribute", "ana-q153-a-promoter-occupancy-call-agrees-with-its-overlapping-neighbour-91-percent-of-the-time" ]
experiments
[ "exp-q153-base-resolution-coverage-across-the-irx2-irx2-dt-shared-promoter" ]
falsifiers
[ "H-A IS REFUTED IF the shared-window fraction among occupancy calls in the KMT2A-N arms exceeds that in the KMT2A-C arms of the SAME 27 samples by more than 5 percentage points, with the direction consistent in at least 20 of 27 samples. Both arms are already in C:\\Temp\\q147part\\promoter_matrix.tsv, so this is scoreable today.", "H-B IS REFUTED IF the KMT2A-C arms show a shared-window fraction within 2 percentage points of the KMT2A-N arms' 43.87%, in which case no residual remains for a fusion-specific mechanism to explain.", "BOTH ARE REFUTED AS STATED IF the enrichment differs by fusion partner - e.g. present in KMT2A::AFF1 and absent in KMT2A::MLLT1 - since neither mechanism is partner-specific. Scoreable across the 6 partners represented in the 27 samples, though several partner groups have n <= 3 and CANNOT carry a statistic; those groups must be named as such rather than pooled silently.", "UNINFORMATIVE IF both arms are enriched and differ by between 2 and 5 percentage points. Pre-declared now: the output is then the residual with an interval and no verdict, and it must not be written up as support for either hypothesis." ]
measurements
[
{
"quantity": "fold_enrichment_of_shared_windows_among_occupancy_calls",
"subject": "occupancy calls by the q147 rule over all 19,404 promoters in 27 GSE252378 samples, KMT2A-N arms",
"value": 2.03,
"units": "fold over the base rate",
"n": 53180,
"universe": "23,331 of 53,180 calls at shared windows; base rate 4,191 of 19,404 = 21.60%",
"null": 1,
"note": "the quantity this hypothesis is about; the KMT2A-C arm version of the same number is the discriminating test and has not been computed"
}
]- If H-A holds, the rule that produced this project's occupancy landscape is partly an accessibility detector, and target lists derived from it will be enriched for constitutively active housekeeping-adjacent promoters - precisely the genes whose inhibition is most likely to be toxic in normal tissue. The confound and the toxicity risk point the same way.
- If H-B holds instead, the finding is real and immediately awkward: a fusion that prefers bidirectional promoters is a fusion whose targets mostly cannot be assigned to one gene, so the better the targeting story gets, the worse the attribution problem becomes.
- What is the shared-window fraction among occupancy calls in the KMT2A-C arms? One run over the existing matrix answers it and decides between H-A and H-B. It is the cheapest unanswered question this beat leaves behind.
- Are shared windows CpG-island promoters? The census classified neighbours by symbol, not by CpG content, and a CpG-island track was not fetched, so the mechanistic half of H-A is asserted from general chromatin biology rather than measured here.
- If H-A holds, how much of the q147 rule's positive-call set is promoter activity rather than fusion occupancy - across ALL windows, not just shared ones? That question is larger than this beat and would revise several existing entries.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
43.87% of 53,180 occupancy calls made by the q147 rule over all 19,404 promoters across 27 GSE252378 samples fall on windows containing another gene's TSS within 1 kb, against a 21.60% base rate, in 27 of 27 samples with a range of 41.3% to 50.8%.
q153 enrichment of shared-window promoters among occupancy calls, GSE252378 · Iqbal Lab oncofusion knowledge graph, beat Q153 · this project 2026What this evidence cannot show
The rule is a faithful reimplementation of q147's without its depth-grade relabelling of zeros, which cannot create a positive. Matrix asserted at load to be RAW coverage and converted to within-sample shares by column mean. No p-value: the 19,404 calls within a sample are not independent, because neighbouring promoters share coverage - which is the very thing being measured - so any nominal p would be fabricated precision. 27 samples from 24 patients.
Evidence that argues against this (1)
- The enrichment is exactly what promoter activity predicts, so on its own it is evidence for nothing beyond chromatin accessibility. Reporting it as a fusion-targeting result without the KMT2A-C control would be the same error this project made when it read an in-promoter enrichment fraction as library depth. Trinklein ND et al. 2004
- E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
The same occupancy rule already calls a median 10.75% of all 19,404 refGene promoters occupied per sample, with a minimum of 1.86% and a maximum of 14.01%, so it is permissive at promoter scale and an arbitrary promoter is expected to be called occupied in 2.73 of 27 samples.
The two-antibody 2x occupancy rule calls a tenth of all promoters occupied · Iqbal Lab oncofusion knowledge graph, beat Q151 · this project 2026What this evidence cannot show
Read from another beat's entry and results file, not independently re-derived here beyond confirming the same rule reproduces on the same matrix. A per-sample median of 10.75% is a property of the rule and this deposit, not a general fact about CUT&RUN.
Evidence that argues against this (1)
- A permissive rule is not a wrong rule - it was chosen to make a NEGATIVE strong, and for that purpose permissiveness is conservative. The criticism here applies only to its positive calls, which is the direction this hypothesis is about. Iqbal Lab oncofusion knowledge graph et al. 2026
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
Every library needed to discriminate the two explanations is already in the 133-library promoter matrix on disk: each of the 27 samples carries two KMT2A-C libraries and, in most cases, an IgG library, alongside the two KMT2A-N libraries the occupancy rule uses.
q153 audit of the GSE252378 promoter matrix column set · Iqbal Lab oncofusion knowledge graph, beat Q153 · this project 2026What this evidence cannot show
Verified by reading the matrix header: 133 columns with the naming pattern <sample>_IgG, <sample>_KMT2Ac_1/2, <sample>_KMT2An_1/2. Not every sample has all five - the discriminability analysis found 27 of 27 with both KMT2A-N libraries, and IgG presence was not counted sample by sample here. The KMT2A-C arms are a control for accessibility only to the extent that the C-terminal fragment is genuinely not part of the oncoprotein at these loci, which is the deposit's own design assumption and not something this beat verified.
Evidence that argues against this (1)
- The KMT2A-C control is imperfect: wild-type KMT2A is present in these cells and its C-terminal fragment does occupy promoters, so a shared-window enrichment in the C arms could reflect wild-type KMT2A targeting rather than generic accessibility. IgG is the cleaner accessibility control and the weaker signal, so the two must be read together and neither alone. Iqbal Lab oncofusion knowledge graph et al. 2026
the 2.03-fold enrichment this hypothesis exists to explain
- related to →KMT2A / MLL1 — the gene that breaks
H-B would require the fusion's CXXC unmethylated-CpG address to prefer bidirectional CpG-island promoters beyond their activity; prior 0.2