Oncofusion

start / graph / ana-the-bidirectional-promoter-enrichment-is-mostly-accessibility-partly-kmt2a-and-not-fusion-specific-at-all

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.

Q153's named threat, run: the reciprocal KMT2A-C arm shows the same bidirectional-promoter enrichment as the fusion-side N arm (44.86% against 43.86%), and IgG shows 33.86% - so the 2.03x is mostly accessibility, carries a real 7-point KMT2A-family component, and is not fusion-specific in any part

also called N 43.86 vs C 44.86 · IgG 33.86% at shared windows · the C-arm control on the shared-window enrichment · the enrichment is not a fusion preference

Nearly half of this project's promoter measurements sit at spots where two genes start back to back, far more than you would expect by chance. Running the same test on the half of the protein that is NOT part of the fusion, and on a negative control antibody, shows the effect is mostly just open chromatin and is not something the fusion does.

The same thing, for a scientist

Q153 found that **43.87% of every promoter-scale occupancy call this project's rule makes is at a window containing another gene's TSS within 1 kb**, against a **21.60%** base rate - a **2.03x** enrichment in 27 of 27 samples. **I re-derived it at 43.86% from my own occupancy calls against its persisted census of 4,191 windows.** It named the control as its cheapest remaining gap and as *"the one threat this beat found to an instrument several recent beats depend on"*: run the same enrichment on the **KMT2A-C** arm. **Run, plus an IgG arm Q153 did not propose** [E1]: | antibody | calls | at a shared window | share | x base | |---|---:|---:|---:|---:| | **KMT2A-N** (fusion side) | 53,027 | 23,258 | **43.86%** | **2.03x** | | **KMT2A-C** (reciprocal side) | 47,111 | 21,133 | **44.86%** | **2.08x** | **N minus C = -1.00 percentage points.** Q153 pre-declared an UNINFORMATIVE band of 2-5 points and a verdict below 2; this is inside it, and in the direction *against* the fusion side. **The enrichment is not a property of the fusion.** 🟠 **But Q153 predicted that a matching C arm would mean the effect is "chromatin accessibility", and that is not quite right.** The C-terminal antibody detects wild-type KMT2A from the unrearranged allele and the reciprocal product, both of which bind chromatin - so the C arm is a control for *fusion specificity*, not for accessibility. Separating the two needs IgG, which the same matrix carries. On a single-library form of the rule so IgG is comparable: | antibody | libraries | share at a shared window | x base | |---|---:|---:|---:| | base rate of shared windows | - | **21.60%** | 1.00x | | **IgG** | 24 | **33.86%** | **1.57x** | | **KMT2A-N** | 54 | **40.88%** | 1.89x | | **KMT2A-C** | 54 | **41.59%** | 1.93x | 🔵 **So the effect is three-tiered, and all three tiers matter.** Accessibility carries the bulk of it - IgG background alone moves 21.60% to 33.86%. **The specific antibodies exceed IgG by about 7 points**, which is a genuine KMT2A-family signal and not background. And **the fusion side has no advantage over the non-fusion side.** 🔴 **What this licenses, stated narrowly.** The shared-window enrichment is **not evidence of fusion targeting** and must not be read as the oncoprotein preferring bidirectional promoters. It does **not** by itself weaken `HOXA9 21/27` or `MEIS1 24/27` as fusion targets - those rest on other evidence including drug response - but it does remove this enrichment as support for them, and it means **any promoter-scale claim in this corpus is about a window that is 44% likely to belong to two genes and whose enrichment is largely accessibility.**

Go deeper
Why the C arm is the wrong control for accessibility, and the right one for specificity

In a KMT2A-rearranged cell the fusion retains KMT2A's N-terminus; the C-terminal fragment is a separate species - wild-type KMT2A from the unrearranged allele, plus the reciprocal product where one exists. Both are chromatin-binding proteins with their own promoter preferences.

So N matches C does not mean the signal is background. It means the preference is shared by fusion-side and non-fusion-side KMT2A signal, i.e. it is a KMT2A-family property rather than a fusion-specific one. That is a weaker claim than Q153's predicted "chromatin accessibility" and a stronger one than "no signal".

🔵 The IgG arm is what separates them, and it does: 33.86% against ~41%. Had IgG matched the specific antibodies, the whole enrichment would have been background. Had IgG sat at the 21.60% base rate, the whole enrichment would have been KMT2A binding. It sits between, so both explanations are partly true and neither alone is.

The caveat on the IgG comparison, stated because it is not small

The IgG arm uses a single-library form of the rule, because IgG libraries are not replicated in pairs the way KMT2An_1/_2 and KMT2Ac_1/_2 are. That changes the statistic: the two-antibody version requires agreement between replicates and is therefore stricter. Under the single-library form the N arm reads 40.88% rather than 43.86%, so the 7-point gap over IgG is measured on the looser rule and the comparison to the 43.86% headline is not like-for-like.

Also: only libraries clearing 50 calls are included, and 24 IgG libraries qualify against 54 for each specific antibody, so the IgG estimate rests on less than half the material. Neither issue changes the ordering, and both are reasons to treat the 7 points as approximate.

Facts and figures
census sourceanalyses/q153-can-an-irx2-promoter-measurement-be-attributed/q153_census_1kb_members.tsv (4,191 windows)
base rate shared PCT21.6
q153 preregistered bandverdict below 2 points, UNINFORMATIVE 2-5, real preference above 5
band outcome-1.00 points: below 2, so not fusion-specific
correction to q153it predicted a matching C arm would show the effect is chromatin accessibility; the C arm is a fusion-specificity control, not an accessibility control, and IgG at 33.86% shows accessibility explains most but not all of the enrichment
two antibody rule
{
  "KMT2A_N": {
    "calls": 53027,
    "at_shared": 23258,
    "pct": 43.86,
    "x_base": 2.03
  },
  "KMT2A_C": {
    "calls": 47111,
    "at_shared": 21133,
    "pct": 44.86,
    "x_base": 2.08
  },
  "N_minus_C_points": -1
}
single library rule
{
  "IgG": {
    "libraries": 24,
    "pct": 33.86,
    "x_base": 1.57
  },
  "KMT2A_N": {
    "libraries": 54,
    "pct": 40.88,
    "x_base": 1.89
  },
  "KMT2A_C": {
    "libraries": 54,
    "pct": 41.59,
    "x_base": 1.93
  },
  "N_minus_IgG_points": 7.02
}
measurements
[
  {
    "quantity": "share of occupancy calls at a window containing another gene's TSS within 1 kb",
    "value": {
      "KMT2A_N": 43.86,
      "KMT2A_C": 44.86,
      "IgG_single_library": 33.86,
      "base_rate": 21.6
    },
    "unit": "percent of calls",
    "subject": "the two-antibody rule on the N and C arms; IgG on a single-library form of the same rule",
    "universe": "19,404 refGene promoters x 27 samples; 4,191 windows in the shared census"
  },
  {
    "quantity": "fusion-side minus reciprocal-side enrichment",
    "value": -1,
    "unit": "percentage points",
    "subject": "KMT2A-N 43.86% against KMT2A-C 44.86%",
    "universe": "Q153's pre-declared bands: verdict below 2, uninformative 2-5, real preference above 5"
  }
]
Still open
  1. Does the KMT2A-C arm's slight excess over N (44.86 against 43.86) mean anything, or is it within run-to-run variation? One point on 47,000 calls is nominally large but the arms differ in library count and depth, and no confidence interval was computed.
  2. Rebuild the census independently. This verifies the enrichment GIVEN Q153's 4,191-window census; an error in the census would propagate silently into every number here.
  3. Is there a depth-matched input or unrelated-factor control anywhere in the deposit? IgG is an antibody-background control, not an accessibility control, and the 7-point gap rests on treating it as the latter.
Every source behind this entry (1)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.

    Applying the project's two-antibody occupancy rule to the KMT2A-C arm of the same matrix gives 21,133 of 47,111 calls at a window containing another gene's TSS within 1 kb (44.86%, 2.08x the 21.60% base rate) against 23,258 of 53,027 for the KMT2A-N arm (43.86%, 2.03x), a difference of -1.00 percentage points; and on a single-library form of the rule IgG gives 33.86% (1.57x) against 40.88% for KMT2A-N and 41.59% for KMT2A-C.

    What this evidence cannot show

    Run 2026-09-24 (cycle 141). The matrix was asserted RAW at load (column means 0.9177-18.7951, 0 of 133 within 1% of 1.0) and converted by column mean. The census of 4,191 shared windows is Q153's, built from fetched refGene annotation, and was not rebuilt here - so this verifies the enrichment given that census rather than the census itself. The IgG comparison uses a single-library rule because IgG is not replicated in pairs, and only 24 IgG libraries clear 50 calls against 54 per specific antibody. My N-arm total of 53,027 differs from Q153's 53,180 by 0.3%, a marginal floor or tie difference.

    Evidence that argues against this (2)
    • The KMT2A-C antibody is not a clean negative control for the fusion. In a KMT2A-rearranged cell it detects wild-type KMT2A from the unrearranged allele, which is a chromatin-binding protein with genuine promoter targets including HOXA. So N matching C is consistent with BOTH proteins preferring bidirectional promoters for the same reason, and does not establish that the fusion lacks such a preference - only that it has no preference the wild-type protein does not share.
    • IgG in CUT&RUN is not a pure accessibility readout either: it reports non-specific tethering, which itself tracks open chromatin but with its own bias profile. A sequencing-depth-matched input or an unrelated-factor control would separate accessibility from antibody background properly, and neither exists in this deposit.
Connected to