The fusion-biased peak at the AFF1 locus fires in 23 of 27 samples across all six partners, so it is not the fusion visiting its own partner
also called Q155 partner breakdown at the AFF1 locus · AFF1 autoregulation control
The hottest spot next to AFF1 lights up in almost every patient, whichever partner gene the fusion happens to carry. If the fusion were drawn there because AFF1 is its own partner, it should light up in the AFF1 patients and not the others. It does not.
The same thing, for a scientist
The obvious story after finding a strong, N-biased peak beside AFF1 in KMT2A::AFF1 leukaemia is **autoregulation**: the fusion protein occupying the locus of the gene that supplies its own C-terminal half. That story is **excluded by the partner breakdown**, which is why this control was run before anything was written. A 400 bp window on the current-RefSeq long C4orf36 TSS (`chr4:86,936,001-86,936,401`, peak B) scored per sample with the project's rule [E1]: | fusion partner | samples occupied | |---|---| | KMT2A::AFF1 | **7 / 8** | | KMT2A::MLLT1 (ENL) | 6 / 7 | | KMT2A::MLLT3 (AF9) | **3 / 3** | | KMT2A::AFDN | 3 / 4 | | KMT2A::MLLT10 (AF10) | 2 / 3 | | KMT2A::EPS15 | **2 / 2** | | **total** | **23 / 27** | **KMT2A::AFF1 samples are not enriched; they are at the cohort rate.** The same holds at peak A (AFF1 6/8 against 20/27 overall) and at AFF1's own TSS +/- 200 bp (7/8 against 21/27). **There is no partner-specific occupancy at this locus in this deposit.** That is the correct control for the autoregulation hypothesis and it is a **negative**, which by Q153's asymmetry is the kind of result that survives attribution: a null over the shared window entails a null over each constituent, so *"the fusion does not preferentially occupy this locus when AFF1 is its partner"* holds regardless of which of the three genes the signal belongs to. **This is the one attributable statement the beat produces.** It also fits the coordinator's control that the shared-window enrichment is **not fusion-specific at all** (KMT2A-C 44.86% against KMT2A-N 43.86%, IgG 33.86%): this locus behaves like a busy, accessible, bidirectional promoter region that most KMT2A-rearranged blasts have open, not like a fusion target.
All four intervals, per partner
Rule: within-sample share >= 2.0x in both KMT2A-N antibodies and above that library's floor, the per-library maximum over eight tissue-restricted loci taken at the same 400 bp width.
| interval | occupied | N/C median | AFF1-partner samples | non-AFF1 samples |
|---|---|---|---|---|
| peak A, 86,934,650-86,935,050 | 20/27 | 1.41 | 6/8 | 14/19 |
| AFF1 TSS +/- 200, 86,934,801-86,935,201 | 21/27 | 1.23 | 7/8 | 14/19 |
| peak B, 86,936,001-86,936,401 | 23/27 | 2.32 | 7/8 | 16/19 |
| IgG hotspot, 86,936,301-86,936,701 | 20/27 | 2.53 | 7/8 | 13/19 |
A Fisher test on peak B (7/8 against 16/19) is not worth computing: with 8 AFF1-partner samples the design cannot detect anything short of a total on/off difference, and it sees none. The correct statement is that the experiment is underpowered for a modest partner effect and excludes a large one.
Note the fourth row. The interval centred on the IgG arm's own maximum scores 20/27 with an N/C median of 2.53 - higher than peak B's. That is the strongest caution in this beat: an N/C ratio at this locus is not by itself evidence of fusion occupancy, because a region where IgG is also high can produce one.
The outlier I am not excluding
Sample TB986292 (KMT2A::AFF1) reads peak B at N mean 143.69 with N/C 33.00, against a cohort median around 9. Its KMT2A-C libraries are near zero at this locus (0.243 and 0.078 in the 2 kb promoter window), so the ratio is a small denominator as much as a large numerator, and its AFF1 promoter call fails the rule because N2 is 0.000 - i.e. one of its two N libraries has no coverage here at all.
It is reported rather than dropped. Two readings are open and the beat does not choose: a genuinely exceptional occupancy in one patient, or a depth artefact in a library whose C arm and one N arm are both empty at this locus. The q147 depth gate (p_dz = frac_zero(N1) x frac_zero(N2)) would separate them and was not run.
| script | analyses/q155-can-an-aff1-promoter-measurement-be-attributed/q155_peaks.py |
|---|---|
| status | run |
| result summary | Peak B occupied 23/27 with no partner enrichment (AFF1 7/8 vs others 16/19); autoregulation excluded at the effect size this design can see; an IgG-adjacent interval scores a HIGHER N/C, which caps how much peak B can be read into. |
| attributable statement | The fusion does not preferentially occupy this locus when AFF1 is its own partner. This is a NEGATIVE over the shared window, so by the superset rule it holds for each of AFF1, AFF1-AS1 and C4orf36 individually and is the one fully attributable statement this beat produces. |
| outlier not excluded | TB986292 (KMT2A::AFF1): peak B N mean 143.69, N/C 33.00, with near-zero C libraries and one N library at 0.000 in the 2 kb window. Reported, not dropped; the depth gate that would adjudicate it was not run. |
inputs
[ "GSE252378, all 133 bigWigs read remotely", "q144 manifest for fusion partner labels" ]
measurements — the numbers behind the chart
[
{
"quantity": "n_samples_occupied",
"subject": "peak B, chr4:86,936,001-86,936,401, KMT2A-N1+N2, 2.0x rule with a 400 bp width-matched floor",
"value": 23,
"units": "samples",
"n": 27,
"universe": "27 GSE252378 KMT2A-fusion patient samples across six 3' partners",
"null": null,
"note": "peak A 20/27, AFF1 TSS +/-200 21/27, IgG-hotspot interval 20/27"
},
{
"quantity": "fraction_occupied",
"subject": "peak B in KMT2A::AFF1 samples",
"value": 0.875,
"units": "fraction of samples",
"n": 8,
"universe": "the 8 KMT2A::AFF1 samples of GSE252378",
"null": 0.842,
"note": "null is the rate in the 19 non-AFF1-partner samples, 16/19 = 0.842. The difference is 0.03 and the design cannot resolve it"
},
{
"quantity": "n_over_c",
"subject": "peak B versus the IgG-hotspot interval 300 bp away, median over 27 samples",
"value": 2.32,
"units": "ratio, median",
"n": 27,
"universe": "peak B chr4:86,936,001-86,936,401 against the IgG-maximum interval chr4:86,936,301-86,936,701",
"null": 2.53,
"note": "the null here is the IgG-hotspot interval's own N/C median, which is HIGHER - so N/C at this locus does not distinguish a fusion peak from a sticky region"
}
]- If a programme were built on 'KMT2A::AFF1 autoregulates its own partner', this control kills it at the level of occupancy in primary patient chromatin, and it kills it cheaply. Anyone proposing that mechanism has to explain why every other partner does the same thing.
- Conversely, a locus that every KMT2A fusion occupies is a worse therapeutic handle than a partner-specific one, not a better one: perturbing it cannot be selective for the index fusion.
- Does peak B fall under menin inhibition? GSE252378 carries treated samples and the per-sample table carries the flag; splitting on it is a one-line change to q155_peaks.py and would separate a fusion-bound peak from an accessibility feature far better than an N/C ratio can.
- Is peak B present in a non-KMT2A-rearranged CUT&RUN library? Nothing in GSE252378 is fusion-negative, so this deposit cannot supply the control; the q144 CD34 comparator used elsewhere in this project could.
- How many of this project's occupancy calls sit within 400 bp of an IgG-arm local maximum? If the answer is large, the N/C ratio is a worse discriminator than the project has been assuming everywhere, not just here.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
A 400 bp window on the current-RefSeq long C4orf36 TSS (chr4:86,936,001-86,936,401) is occupied in 23 of 27 GSE252378 samples under the project's two-antibody rule with a width-matched floor, with 7 of 8 KMT2A::AFF1 samples, 6 of 7 KMT2A::MLLT1, 3 of 3 KMT2A::MLLT3, 3 of 4 KMT2A::AFDN, 2 of 3 KMT2A::MLLT10 and 2 of 2 KMT2A::EPS15, so the KMT2A::AFF1 samples are not enriched.
Q155 per-sample peak scoring at the AFF1 locus with fusion partner attached, GSE252378, all 133 libraries read remotely · Iqbal Lab oncofusion knowledge graph, beat Q155 · this project 2026What this evidence cannot show
Eight KMT2A::AFF1 samples against nineteen others is a small and unbalanced design; it excludes a large partner effect and cannot exclude a modest one, and no test statistic is reported because none would be informative. The partner labels come from the q144 manifest. No spike-in exists in this deposit. The region gate asserted all twelve intervals at exactly 400 bp before any value was read, and the run is complete at 133 of 133 libraries with zero failures.
Evidence that argues against this (1)
- An interval centred on the IgG arm's own maximum 300 bp further along (chr4:86,936,301-86,936,701) scores 20 of 27 with an N/C median of 2.53, HIGHER than peak B's 2.32. So a high N/C at this locus is not by itself a fusion signature, and the partner-negative result is the robust half of this entry while the peak-B-is-special half is not. Iqbal Lab oncofusion knowledge graph et al. 2026
- E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
The partner-negative result at this locus is consistent with the project's own control finding that the bidirectional/shared-window occupancy enrichment is not fusion-specific, where the same rule calls 44.86% of KMT2A-C calls, 43.86% of KMT2A-N calls and 33.86% of IgG calls at shared windows.
The bidirectional-promoter enrichment is mostly accessibility, partly KMT2A, and not fusion-specific at all · Iqbal Lab oncofusion knowledge graph, coordinator control run · this project 2026What this evidence cannot show
Those three percentages are a corpus-wide property of the occupancy rule, not a measurement at this locus, so this is consistency rather than corroboration. Citing another entry's number is weaker than re-deriving it and it was not re-derived here; the figures are quoted from the coordinator's control as reported in this beat's dispatch and should be re-run before being used as a load-bearing quantity.
Evidence that argues against this (1)
- Consistency with a base rate is what a base rate produces. That this locus behaves like the average shared window is evidence that it is unremarkable, which is the claim - but it would look exactly the same if the rule were simply insensitive, and no positive control locus was scored alongside it in this beat. Iqbal Lab oncofusion knowledge graph et al. 2026
per-sample scoring of the two modes the profile found, with the fusion partner attached
no partner-specific occupancy: 7/8 KMT2A::AFF1 against 16/19 other partners, so the fusion does not preferentially occupy its own partner's locus
a worked single-locus instance of the coordinator's corpus-wide control