The KMT2A-N signal at the AFF1 locus is bimodal, neither mode is on AFF1's TSS, and the flank test I ran first said the opposite
also called Q155 base-resolution profile · AFF1 locus two peaks
Looking at the DNA fifty letters at a time instead of two thousand at a time shows two separate hotspots around AFF1, and AFF1's own starting point sits in the dip between them. The bigger hotspot is just outside the window this project measures, sitting on a different gene's start.
The same thing, for a scientist
50 bp bins across `chr4:86,932,500-86,938,500`, averaged over all **54 KMT2A-N**, **54 KMT2A-C** and **25 IgG** libraries of GSE252378, read remotely with no bigWig downloaded [E1]. The KMT2A-N profile is **bimodal**: - **Peak A**, maximum in the bin `86,934,750-86,934,800` at **12.83x an average promoter**, sitting between C4orf36's near TSS (86,934,732) and AFF1-AS1's TSS (86,934,850), i.e. **150-250 bp UPSTREAM of AFF1's TSS**. Over `86,934,700-86,935,000`: N **11.12**, C **8.49**, **N/C 1.31**. - **Peak B**, maximum in the bin `86,936,250-86,936,300` at **21.23x** - the **global maximum of the entire 6 kb interval** - sitting on the long C4orf36 TSS (86,936,204) and Ensembl's unnamed antisense lncRNA TSS (86,936,202), **203-299 bp OUTSIDE the project's AFF1 window**. Over `86,936,000-86,936,350`: N **15.24**, C **4.42**, **N/C 3.45**, N/IgG **3.85**. **AFF1's own TSS sits in the trough.** The bin containing it reads 6.58 and the signal falls to 1.31 within 200 bp downstream. Over the 1 kb either side of AFF1's TSS: upstream **5.247**, downstream **2.704**, **downstream/upstream = 0.515** for KMT2A-N. There is **twice as much KMT2A-N signal on the wrong side of AFF1's TSS as on AFF1's side** [E1]. **This overturns the reading I had two steps earlier and the correction is the point of this entry.** The interval deconvolution found the 151 bp unique to an AFF1 window occupied in **15 of 27** samples against **1 of 27** for the mirror flank unique to an AFF1-AS1 window, and AFF1's gene body carrying 2.247x the width-matched AFF1-AS1 body. Read alone, that says the promoter signal is AFF1's. **It is an artefact of where the flanks are.** The 'unique to AFF1' flank is `86,935,850-86,936,001` - the **rising shoulder of peak B**, whose summit is 249 bp further on, on a minus-strand gene's TSS. The flank test was measuring the approach to somebody else's promoter and scoring it for AFF1 [E2].
The profile, in the bins that matter
Mean over libraries of (bin mean coverage / that library's own mean promoter coverage):
| position | KMT2A-N | KMT2A-C | IgG | N/C | landmark |
|---|---|---|---|---|---|
| 86,934,700 | 8.86 | 5.78 | 6.13 | 1.53 | C4orf36 TSS (near), minus |
| 86,934,750 | 12.83 | 7.89 | 7.69 | 1.63 | peak A summit |
| 86,934,850 | 9.58 | 7.10 | 1.22 | 1.35 | AFF1-AS1 TSS, minus |
| 86,934,950 | 11.91 | 11.42 | 2.03 | 1.04 | |
| 86,935,000 | 6.58 | 4.54 | 0.60 | 1.45 | AFF1 TSS, plus |
| 86,935,200 | 1.31 | 4.67 | 1.29 | 0.28 | trough; the C arm is HIGHER here |
| 86,935,850 | 4.19 | 2.35 | 7.14 | 1.78 | the '151 bp unique to AFF1' flank |
| 86,936,000 | 10.16 | 4.48 | 3.99 | 2.27 | window upper edge |
| 86,936,200 | 20.95 | 6.02 | 4.92 | 3.48 | C4orf36 TSS (far) / Ensembl lncRNA TSS |
| 86,936,250 | 21.23 | 3.23 | 3.38 | 6.58 | peak B summit, global max |
| 86,936,400 | 6.73 | 2.31 | 11.86 | 2.91 | IgG's own maximum |
Interval means:
| interval | N | C | IgG | N/C | N/IgG |
|---|---|---|---|---|---|
| peak A, 86,934,700-86,935,000 | 11.12 | 8.49 | 4.00 | 1.31 | 2.78 |
| AFF1 TSS +/- 100 | 8.51 | 7.43 | 1.35 | 1.15 | 6.30 |
| trough, 86,935,100-86,935,500 | 1.52 | 3.29 | 0.72 | 0.46 | 2.12 |
| peak B, 86,936,000-86,936,350 | 15.24 | 4.42 | 3.96 | 3.45 | 3.85 |
| the project's whole 2 kb window | 3.98 | 2.87 | 1.77 | 1.38 | 2.24 |
Peak A is shared between the arms (N/C 1.31) and peak B is not (N/C 3.45). The C-terminal epitope is lost from the fusion, so an N-and-C signal is wild-type KMT2A and an N-only signal is the fusion. The fusion-biased peak at this locus is peak B, and peak B is outside the window this project measures.
Why the flank test misled me, stated as a method error
The deconvolution compared two 151 bp intervals chosen to be symmetric: each lies 849-1,000 bp into its own gene's body from its own TSS, and each is 1,000-1,151 bp from the other gene's TSS. That symmetry is real and it is why the test looked sound. It gave 15/27 vs 1/27, and I read it as the AFF1 direction winning.
What the design did not control for is a third TSS just outside the window on the AFF1 side. The 'unique to AFF1' flank is 203-354 bp from the long C4orf36 / lncRNA TSS at 86,936,204 - closer to that TSS than to AFF1's own, which is 849 bp away. A peak centred there produces exactly the observed asymmetry, and so does AFF1-directed elongation. The two models are indistinguishable by the flank test and distinguishable by where the signal decays, which is what the 50 bp profile shows: the signal rises monotonically from 86,935,200 to 86,936,250 and then falls, so it is decaying away from 86,936,250, not away from 86,935,001.
The general form, which is the transferable part: a deconvolution designed around two features cannot be rescued by making it symmetric in those two features if a third feature is sitting just past the edge. The pre-registration named this branch - 'if the window contains a third gene's TSS, neither attributable-to-A nor attributable-to-B is the right question' - and I then designed the deconvolution before I had looked for the third gene, which is the ordering mistake.
| script | analyses/q155-can-an-aff1-promoter-measurement-be-attributed/q155_profile.py |
|---|---|
| status | run |
| result summary | Bimodal KMT2A-N signal; peak A (N/C 1.31) 150-250 bp upstream of AFF1's TSS between two minus-strand TSSs, peak B (N/C 3.45, global max) 203-299 bp OUTSIDE the project's window on the long C4orf36 / Ensembl-lncRNA TSS. AFF1's TSS is in the trough; downstream/upstream = 0.515. |
| window | chr4:86,932,500-86,938,500, 50 bp bins, 120 bins |
| self correction | This entry overturns the directional reading of q155_deconv.py's 151 bp flank test, which was run first by the same beat. |
inputs
[ "GSE252378, all 133 bigWigs read remotely by HTTP range request; no file downloaded", "C:/Temp/q147part/extract_meta.json for per-library promoter denominators" ]
peaks
{
"KMT2An": "chr4:86,936,250-86,936,300 at 21.228",
"KMT2Ac": "chr4:86,934,900-86,934,950 at 11.597",
"IgG": "chr4:86,936,400-86,936,450 at 11.860"
}centre of mass in project window — the numbers behind the chart
{
"KMT2An": 86934984.7,
"KMT2Ac": 86935028.8,
"IgG": 86935274
}measurements — the numbers behind the chart
[
{
"quantity": "downstream_over_upstream",
"subject": "mean KMT2A-N share in the 1 kb downstream of AFF1's TSS over the 1 kb upstream",
"value": 0.515,
"units": "ratio",
"n": 54,
"universe": "54 KMT2A-N libraries, 50 bp bins, chr4:86,934,001-86,936,001 split at AFF1's TSS 86,935,001",
"null": 1,
"note": "1.0 is what a peak centred on AFF1's TSS would give. KMT2A-C gives 0.700 and IgG 1.145"
},
{
"quantity": "n_over_c",
"subject": "peak A (chr4:86,934,700-86,935,000), between C4orf36's near TSS and AFF1-AS1's TSS",
"value": 1.31,
"units": "ratio of arm means",
"n": 108,
"universe": "54 KMT2A-N and 54 KMT2A-C libraries, mean of within-sample shares over the interval",
"null": 1,
"note": "1.0 is what wild-type KMT2A gives, because the C epitope is lost from the fusion. Peak A is barely above it"
},
{
"quantity": "n_over_c",
"subject": "peak B (chr4:86,936,000-86,936,350), on the long C4orf36 / Ensembl lncRNA TSS, OUTSIDE the project's AFF1 window",
"value": 3.45,
"units": "ratio of arm means",
"n": 108,
"universe": "same 108 libraries, same construction",
"null": 1,
"note": "the highest N/C anywhere in the 6 kb; N/IgG 3.85 over the interval and 6.28 at the best single 50 bp bin"
},
{
"quantity": "distance_from_window_edge",
"subject": "peak B's summit bin relative to the upper edge of the project's 2 kb AFF1 window",
"value": 249,
"units": "bp outside",
"n": 1,
"universe": "window ends chr4:86,936,001; summit bin 86,936,250-86,936,300",
"null": null,
"note": "a 2.5 kb window would have contained it; a 2 kb window does not"
}
]- Any experiment designed to test 'does the fusion sit on AFF1's promoter' using a 2 kb window will return yes, and the yes will be about a peak whose summit is outside the window on a different gene's TSS. Designing the experiment from the project's existing promoter table reproduces the error rather than testing it.
- The trough immediately downstream of AFF1's TSS has the C arm HIGHER than the N arm (N/C 0.46). Anyone reading a promoter mean as 'fusion occupancy at AFF1' is averaging a fusion-biased shoulder, a wild-type-biased summit and a wild-type-biased trough into one number.
- Is peak B a promoter or an artefact? An IgG maximum sits 150 bp away, and nothing here distinguishes a fusion-bound TSS from a sticky region with an N-arm-specific antibody bias. The cheapest discriminator is whether peak B appears at the equivalent position in a non-KMT2A-rearranged CUT&RUN library from any deposit.
- Does peak B move under menin inhibition? GSE252378 contains treated samples, the per-sample table carries the treated flag, and the split was not run. If peak B is fusion-bound it should fall; if it is accessibility it should not.
- Does the same bimodality appear at IRX2/IRX2-DT? Q153 proposed exactly this tiling experiment there (exp-q153-base-resolution-coverage-across-the-irx2-irx2-dt-shared-promoter) and it has not been run. The code in q155_profile.py runs it in about three minutes with the interval changed.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
Averaged over 54 KMT2A-N, 54 KMT2A-C and 25 IgG GSE252378 libraries in 50 bp bins, the KMT2A-N profile across chr4:86,932,500-86,938,500 has its global maximum at 86,936,250-86,936,300 (21.23) and a second mode at 86,934,750-86,934,800 (12.83), with AFF1's TSS bin at 6.58 between them; peak A has N/C 1.31 and peak B has N/C 3.45, and the mean KMT2A-N share in the 1 kb downstream of AFF1's TSS is 0.515 times that in the 1 kb upstream.
Q155 50 bp base-resolution profile across chr4:86,932,500-86,938,500 over all 133 GSE252378 libraries, read remotely · Iqbal Lab oncofusion knowledge graph, beat Q155 · this project 2026What this evidence cannot show
This is a MEAN over libraries of within-sample shares, so a single extreme library can move it - and one sample (TB986292) has near-zero C libraries and reaches 143.69 at peak B, which inflates the N/C ratios in the per-sample table though not the per-bin means shown here, which are N and C averaged separately. No spike-in exists in this deposit. All 133 libraries were read and the script exits non-zero on any failure; the run was complete. Bin means are not peak calls: no peak-calling algorithm, background model or significance test was applied, and 'bimodal' is a description of the averaged trace, not a statistical claim.
Evidence that argues against this (1)
- The IgG arm has its OWN maximum at 86,936,400-86,936,450 (11.86), 150-200 bp from peak B's summit, so this stretch of chromosome is generically sticky and a fraction of peak B is accessibility rather than occupancy. Peak B's separation from that is N/IgG 6.28 at its best bin and 3.85 over the 350 bp interval - real, but not large, and the whole finding would change if the IgG hotspot were wider than measured. Iqbal Lab oncofusion knowledge graph et al. 2026
- E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
The interval deconvolution run before the profile found the 151 bp unique to an AFF1 promoter window occupied in 15 of 27 samples against 1 of 27 for the symmetric mirror flank unique to an AFF1-AS1 window, and AFF1's width-matched gene body carrying a median 2.247x the AFF1-AS1 body; the profile shows that the 'unique to AFF1' flank is the rising shoulder of a peak whose summit is 249 bp beyond it on a minus-strand TSS, so the asymmetry does not establish AFF1 directionality.
Q155 interval deconvolution over all 133 GSE252378 libraries · Iqbal Lab oncofusion knowledge graph, beat Q155 · this project 2026What this evidence cannot show
The two 151 bp flanks ARE symmetric with respect to the AFF1/AFF1-AS1 pair - each is 849-1,000 bp into its own gene from its own TSS and 1,000-1,151 bp from the other's - so the design is not wrong in its own terms; it is under-specified, because a third TSS 203 bp past the window's upper edge breaks the symmetry it assumes. The gene-body ratio (2.247, above 1.0 in 25 of 27) is NOT overturned by the profile and remains the one directional statistic that survives: it is measured 1.2 to 11 kb away from peak B.
Evidence that argues against this (1)
- The gene-body asymmetry survives and points at AFF1, so the correction is partial rather than total. AFF1's width-matched first 9,957 bp is occupied in 4 of 27 and AFF1-AS1's body in 0 of 27, and that interval starts at 86,936,204 - i.e. it begins exactly where peak B is and extends 10 kb past it, so even this statistic is not cleanly separated from peak B's shoulder. Iqbal Lab oncofusion knowledge graph et al. 2026
the base-resolution profile shows the 20/27 call is not centred on AFF1's TSS; twice as much KMT2A-N signal sits upstream of it as downstream
the global maximum of the 6 kb interval sits on C4orf36's current-RefSeq long TSS, 249 bp outside the project's AFF1 window
Q153 proposed this tiling at IRX2/IRX2-DT; the same code was run at AFF1/AFF1-AS1 instead and the IRX2 version is still unrun
per-sample scoring of the two modes the profile found, with the fusion partner attached