Oncofusion

start / graph / finding-q152-the-seventeen-promoter-aff1-only-recurrent-core-is-not-more-recurrence-than-a-label-permutation-gives

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

The 17-promoter AFF1-only recurrent core is not more recurrence than a label permutation of the same matrix gives

also called q152 branch A · AFF1-only recurrent core null

Seventeen gene switches look like they are turned on only in the KMT2A::AFF1 leukaemias and in three or more of the eight samples, which feels like a shared programme. When we shuffle which eight of the twenty-seven patients we call 'the AFF1 group' and re-count, a random group of eight gives nine such genes on average and twenty-one at the 95th percentile - so seventeen is unremarkable.

The same thing, for a scientist

Pre-registered test (branch A). Under the q147/q151 two-antibody 2.0x-plus-floor occupancy rule, 17 of 19,404 promoters are occupied in >=3 of the 8 KMT2A::AFF1 samples of GSE252378 and 0 of the 19 non-AFF1 samples. A label permutation that holds the 19,404 x 27 binary occupancy matrix fixed and re-draws the 8-sample index arm 10,000 times gives mean 9.1 (SD 6.1, median 8, 5th-95th 2-21, full range 0-56), one-sided p_upper = 0.1119 and p_lower = 0.9054. Restricting to the 4,335 draws whose 8-sample mean pass rate is within 10% of the real AFF1 arm's raises the null to mean 12.0 (95th 24), p_upper = 0.2004. Neither test clears 0.05. Branch A of the pre-registration fires: the recurrent core is not distinguishable from 17 recurrent singletons.

Go deeper
What was tested and why the permutation is the right null

The pre-registration named three branches and the null that separates them 1. The independence calculation this project would reach for first is not conservative here (see ana-q152-an-independence-null-overstates-recurrence-sixteenfold-in-this-matrix-and-is-not-a-lower-bound), so the primary null holds the real occupancy matrix fixed and permutes only the partner labels. That preserves every promoter's real cross-sample correlation, every sample's real pass rate, and the real CpG/activity clustering of occupancy - it is not an independence null 2.

The numbers, with both tails labelled
nullmeanmedian5th95thrangep (obs >= null)p (obs <= null)
N2 label permutation, 10,000 draws of 8 of 279.182210-560.11190.9054
N2b same, restricted to the 4,335 draws pass-rate-matched to the AFF1 arm12.0--24-0.20040.8286
C2 variant, 25 samples after dropping the two menin-inhibitor-treated non-AFF1 libraries11.9--25-0.21630.8183

Observed = 17 in all three. The +1-corrected permutation p-values are quoted; both tails are given because the lower tail is the more interesting one - the observed value is closer to the null's centre than to either extreme 2.

Leave-one-out: no single sample carries it

Dropping each AFF1 sample entirely and re-counting at the same >=3 threshold gives TB920488 17, TB051939 16, p318 16, p179 15, TB986292 14, p247 13, p279 10, p454 8. p454 is the largest single contributor. Because the observed value is already inside the null, escape condition C5 is moot - but the spread is recorded because it is a property of the set 2.

What this does NOT say

It does not say that no member of the 17 is a real KMT2A::AFF1 target. One of them, EPHA7, has a published ChIP showing ALL1/AF4 fusion-protein occupancy on its promoter and loses expression when ALL1/AF4 is knocked down in a t(4;11) line - see finding-q152-epha7-is-the-one-core-member-with-a-published-kmt2a-aff1-promoter-chip. A negative about a set is not a negative about each member. What fails is the inference from the recurrence count to a programme.

Facts and figures
scriptanalyses/q152-the-aff1-only-recurrent-promoter-core/step2.py
statusrun
result summaryObserved 17; label-permutation null mean 9.1 (95th 21), p_upper 0.1119; pass-rate-matched null mean 12.0 (95th 24), p_upper 0.2004. Branch A fires.
branch firedA - NOT A PROGRAMME (on the recurrence question); C3 fires concurrently on the identity question
leave one out counts
p4548.000p454: 8p27910.0p279: 10p24713.0p247: 13TB98629214.0TB986292: 14p17915.0p179: 15TB05193916.0TB051939: 16p31816.0p318: 16TB92048817.0TB920488: 17
measurements
permutation_p_upper0.112permutation_p_upper: 0.1119permutation_p_upper_passr…0.200permutation_p_upper_passrate_matched: 0.2004count_recurrent_partner_e…17.0count_recurrent_partner_exclusive_promoters: 17
inputs
[
  "C:/Temp/q147part/promoter_matrix.tsv (RAW mean coverage, 19,404 x 133)",
  "analyses/q147-hoxa-irx1-across-partners/results_gene_scale147.json (partner labels)"
]
core genes
[
  "AMZ2",
  "BLNK",
  "CCP110",
  "EPHA7",
  "GBA",
  "MAGI3",
  "MATR3",
  "MED28",
  "MTFR1",
  "NOP58",
  "PBLD",
  "POU2AF1",
  "SPRED1",
  "SYNE3",
  "TBX21",
  "UHRF1BP1L",
  "WT1"
]
leave one out counts — the numbers behind the chart
{
  "TB920488": 17,
  "TB051939": 16,
  "p318": 16,
  "p179": 15,
  "TB986292": 14,
  "p247": 13,
  "p279": 10,
  "p454": 8
}
measurements — the numbers behind the chart
[
  {
    "quantity": "count_recurrent_partner_exclusive_promoters",
    "subject": "promoters occupied in >=3 of the 8 KMT2A::AFF1 samples and 0 of the 19 non-AFF1 samples",
    "value": 17,
    "units": "promoters",
    "n": 19404,
    "universe": "19,404 refGene 2 kb promoters x 27 GSE252378 samples, two-antibody 2.0x within-sample share plus per-library width-matched floor",
    "null": 9.1,
    "note": "null is the mean of 10,000 label permutations of the same fixed occupancy matrix; 95th percentile 21, p_upper 0.1119"
  },
  {
    "quantity": "permutation_p_upper",
    "subject": "observed 17 against the label-permutation null",
    "value": 0.1119,
    "units": "one-sided permutation p, +1 corrected",
    "n": 10000,
    "universe": "10,000 random draws of 8 of the 27 samples as a pseudo-index arm",
    "null": 0.5,
    "note": "p_lower = 0.9054; neither tail is significant"
  },
  {
    "quantity": "permutation_p_upper_passrate_matched",
    "subject": "observed 17 against the pass-rate-matched label-permutation null",
    "value": 0.2004,
    "units": "one-sided permutation p, +1 corrected",
    "n": 4335,
    "universe": "the 4,335 of 10,000 draws whose 8-sample mean pass rate is within 10% of the AFF1 arm's",
    "null": 0.5,
    "note": "matched null mean 12.0, 95th percentile 24"
  }
]
How this could bite us back
  • If a menin-inhibitor programme were designed around the 'AFF1-specific' promoter core, 16 of the 17 genes would be selected on a statistic this analysis shows is at the chance level, and the trial would be enriching for a set with no partner specificity.
  • The one member with a real published ALL1/AF4 promoter ChIP, EPHA7, drives ERK phosphorylation, so a therapy aimed at it inherits every MAPK-pathway resistance route.
Still open
  1. Does a margin-based statistic - weakest occupied call divided by the strongest non-index call - separate partner-specific promoters where a recurrence count does not? WT1 (2.17x), EPHA7 (1.84x) and SPRED1 (1.59x) are the only three core members whose margin exceeds 1.5.
  2. Would the same permutation null applied to a within-family index arm (e.g. the 4 KMT2A-AFDN samples) also return ~9? If the null is partner-invariant then partner-exclusivity as a concept has no purchase in this deposit at promoter scale.
  3. The 8 AFF1 samples are 7 patients. Would an 8th independent KMT2A::AFF1 patient move the observed count, and by how much - i.e. how steep is the recurrence-versus-n curve?
Every source behind this entry (3)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    The three branches, the decision thresholds, the nulls and the five NEITHER triggers were written down before any of the statistics below were computed; only the coordinator's gene list, the q147 partner table, the matrix column names and the TSS table had been read.

    Q152 pre-registration, written before computation · Iqbal Lab oncofusion knowledge graph, beat Q152 · this project 2026
    What this evidence cannot show

    A pre-registration written by the same agent that runs the analysis is weaker than one filed with a third party. The file is committed before the results file and its git timestamp is the only external check.

    Evidence that argues against this (1)
    • A pre-registration authored and executed by the same agent in the same session cannot be audited independently; the git commit order is the only external evidence of sequence, and if the coordinator commits the results and the pre-registration together that evidence is lost. Iqbal Lab oncofusion knowledge graph et al. 2026
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    Re-deriving the whole q151 calibration from C:/Temp/q147part/promoter_matrix.tsv reproduces every published figure to the digit (median pass rate 10.75%, min 1.86% A80956, max 14.01% p247, 6,049 occupied anywhere, 13,355 nowhere, 332 AFF1-only, MEIS1 24 / HOXA9 21 / PROM1 11 / IRX1 2 / IRX2 0); the >=3-of-8 and 0-of-19 count is 17, and 10,000 label permutations of the same fixed matrix give mean 9.1, 95th percentile 21, p_upper 0.1119.

    Q152 label-permutation null over the GSE252378 promoter occupancy matrix · Iqbal Lab oncofusion knowledge graph, beat Q152 · this project 2026
    What this evidence cannot show

    The permutation draws an 8-sample arm across all six partner families, so it does not specifically test a single-family arm (e.g. the 4 AFDN samples plus 4 others). Named in the pre-registration as the cheapest refutation of a positive result; since the result is negative it makes the negative, if anything, conservative.

    Evidence that argues against this (1)
    • If the occupancy rule had a bias that suppresses recurrence in the non-AFF1 arm specifically, the permutation null would be too high and this negative would be wrong. The pass rates of the non-AFF1 samples (1.86-14.01%) span the AFF1 samples' range, which argues against such a bias but does not exclude it. NCBI Gene Expression Omnibus 2024
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    The 27 GSE252378 samples carry six KMT2A fusion partners and the 8 KMT2A::AFF1 samples are TB051939, TB920488, TB986292, p179, p247, p279, p318 and p454.

    MLL oncoprotein levels influence leukemia lineage identities · Janssens DH, Duran M, Otto DJ, Wu W, Xu Y, Kirkey D, Mullighan CG, Yi JS, Meshinchi S, Sarthy JF, Ahmad K, Henikoff S · Nat Commun 15:9341 2024
    PMID 39472576DOI
    What this evidence cannot show

    Partner assignments were taken from this project's own q147 per-sample table, which was itself built from the deposit metadata. p279 and p318 are two lineage states of ONE patient, so the arm is 7 patients.

    Evidence that argues against this (1)
    • Partner assignments were taken from this project's own q147 table rather than re-read from the GEO metadata in this beat. If any sample is mis-assigned - and this project has previously found a misidentified variable in this same deposit in cycle 138 - both the observed count and the null change. NCBI Gene Expression Omnibus 2024
Connected to