Oncofusion

start / graph / ana-q152-aff1-only-is-a-property-of-mid-activity-promoters-and-is-empty-in-the-top-decile

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q152-aff1-only-recurrent-core

'AFF1-only' is a property of mid-activity promoters: zero in the bottom three deciles, zero in the top decile, and 81% of the 332 in deciles 6-8

also called q152 activity stratification

Sorting all nineteen thousand gene switches by how much oncoprotein signal they carry on average, the 'only in KMT2A::AFF1' ones are all in the middle of the range. The quietest thirty per cent never pass the test at all, and the loudest ten per cent pass it in every patient, so neither can ever look partner-specific. Partner-specificity is only visible in the middle, which is where a threshold decision is fragile.

The same thing, for a scientist

Binning the 19,404 promoters into deciles of mean within-sample share across all 54 KMT2A-N libraries: deciles 1-3 (mean share 0.079-0.159) contain zero promoters occupied in any sample; decile 10 (mean share 4.099) contains 1,941 of 1,941 occupied somewhere and **zero** AFF1-only; the 332 AFF1-only promoters sit 6 / 27 / 91 / 177 / 31 / 0 across deciles 5-10, and 15 of the 17 recurrent core are in deciles 8-9. The AFF1-only label is therefore not available to either end of the activity distribution by construction. Any comparison of an 'AFF1-only' set against an unmatched background is comparing activity bands, which is why the identity test in this beat uses an activity-matched null.

Go deeper
The decile table
decilenmean shareoccupied anywhereAFF1-onlyrecurrent core
119410.079000
219400.120000
319400.159000
419410.224700
519400.3665360
619400.606149271
719410.934525911
819401.36514651779
919402.0481909316
1019414.099194100

The AFF1-only count peaks at decile 8 and collapses to zero at decile 10 - not because the top decile is uninteresting but because everything in it is occupied in every sample, so no exclusion pattern can form there 1.

What this forces on any downstream test

A gene set drawn at random from the 6,049 occupied-anywhere promoters is drawn mostly from deciles 8-10. The 17 core promoters come from deciles 6-9 with a mode at 8. Any functional-enrichment comparison against the unmatched 6,049 is therefore partly a comparison of activity bands, and this project already holds the general version of that caveat from the q146 Jaccard work. The identity test in ana-q152-the-lymphoid-look-of-the-core-does-not-survive-an-activity-matched-null-inside-the-same-332 is decile-matched for this reason, and the matching is what changes the answer.

Facts and figures
scriptanalyses/q152-the-aff1-only-recurrent-promoter-core/step3.py
statusrun
result summaryAFF1-only is structurally unavailable in deciles 1-3 (nothing occupied) and decile 10 (everything occupied everywhere); 295 of 332 sit in deciles 6-8.
measurements
[
  {
    "quantity": "share_of_AFF1_only_in_deciles_6_to_8",
    "subject": "the 332 AFF1-only promoters",
    "value": 0.889,
    "units": "fraction",
    "n": 332,
    "universe": "19,404 refGene promoters binned into deciles of mean within-sample share across 54 GSE252378 KMT2A-N libraries",
    "null": 0.3,
    "note": "295 of 332; the null 0.3 is the share of all promoters in three deciles"
  },
  {
    "quantity": "n_AFF1_only_in_top_activity_decile",
    "subject": "AFF1-only promoters in decile 10",
    "value": 0,
    "units": "promoters",
    "n": 1941,
    "universe": "the top activity decile, in which all 1,941 promoters are occupied in at least one sample",
    "null": null,
    "note": "structurally zero: an exclusion pattern cannot form where every sample is occupied"
  }
]
Still open
  1. Is the decile-8 peak in AFF1-only promoters a general property of the 2.0x rule, i.e. does the same peak appear for a pseudo-index arm? The permutation counts were not decomposed by decile.
  2. What fraction of the decile-10 promoters are CpG islands, and does the rule simply recover CpG-island promoters in every sample?
Every source behind this entry (1)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    Across activity deciles of mean within-sample share over the 54 GSE252378 KMT2A-N libraries, the 332 AFF1-only promoters distribute 0/0/0/0/6/27/91/177/31/0 from decile 1 to 10 and the 17 recurrent core distribute 0/0/0/0/0/1/1/9/6/0, while occupancy anywhere rises monotonically from 0 to 1,941 of 1,941.

    Q152 activity-decile stratification of the AFF1-only and recurrent-core promoter sets · Iqbal Lab oncofusion knowledge graph, beat Q152 · this project 2026
    What this evidence cannot show

    Deciles are computed on the mean across all 54 KMT2A-N libraries including the AFF1 ones, so the stratification variable is not independent of the sets being stratified. A leave-arm-out activity estimate was not computed; it would shift the deciles slightly and cannot change the two structural zeros (deciles 1-3 and decile 10).

    Evidence that argues against this (1)
    • The deciles are computed from the same 54 KMT2A-N libraries that define occupancy, so the stratification variable is not independent of the outcome. A leave-arm-out activity estimate, or an orthogonal activity measure such as an input or IgG track, could move promoters between deciles and change the shape of the AFF1-only distribution; the two structural zeros (nothing occupied in deciles 1-3, everything occupied in decile 10) cannot move, but the decile-8 peak could. NCBI Gene Expression Omnibus 2024
Connected to