Oncofusion

start / graph / ana-q148-the-gmp-like-promoter-class-carries-no-fusion-specific-n-over-c-excess-outside-the-sample-that-defined-it

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q148-what-the-hoxa-negative-blast-occupies

The GMP-like promoter class carries no N-over-C excess in 25 of GSE252378's 27 samples: whatever occupies it is seen equally by the fusion-side and the wild-type-side antibody

Two antibodies are used: one sees both the fusion protein and the normal protein, the other sees only the normal one. At the myeloid gene set they give the same answer in almost every patient - which means the thing sitting there is not the fusion.

The same thing, for a scientist

The deposit's own test for fusion-specific occupancy is an N-terminal-over-C-terminal ratio, with the authors' `Count_N_C` using a log-fold-change cutoff of **1.75** at p < 1e-5. N reads the KMT2A N-terminal fragment that every KMT2A fusion retains, plus wild-type KMT2A; C reads the reciprocal/wild-type C fragment alone. Applied to the three promoter classes across all 27 samples: - `AML_specific` (GMP-like, 40 promoters): N/C spans **0.787 to 2.300** and **only 2 of 27 samples exceed 1.75** - **p318 at 2.300** (the sample that defined the class) and **TB986292 at 1.814** (depth-INADEQUATE, so a positive call is admissible but weak, and its matched-null ratio is 1.263, ns). Every other sample is between 0.787 and 1.266. - `ALL_specific` (37 promoters): N/C reaches **4.839** (p179) and **4.069** (A89691), and 18 of 27 exceed 1.75. - `Oncoprotein_Target` (827 promoters): **16 of 27 exceed 1.75**, range 0.965-2.846. **So the instrument detects fusion-specific occupancy freely at the canonical class and at the B-lineage class, and essentially never at the GMP-like class.** In A73051 - the HOXA-negative, IRX1-negative blast this beat was sent to explain - the GMP-like class N/C is **0.964**, the lowest of its three classes.

Go deeper
The full table, because a count of 2 of 27 needs its members named

AML_specific N/C, all 27 samples, highest first: p318 2.300; TB986292 1.814; p247 1.266; p179 1.249; A95941 1.214; A89691 1.203; p454 1.200; TB995291 1.188; p186 1.185; A93692 1.185; A82485 1.184; p279 1.168; TB051939 1.115; 148752 1.113; TB920488 1.112; p160 1.069; A70498 1.068; R3595 1.059; TB971992 1.014; p214 1.011; 152985 0.992; A73051 0.964; R3652 0.920; A93949 0.909; TB000694 0.904; A80956 0.847; TB040576 0.787.

Twenty-five of 27 sit between 0.787 and 1.266 - i.e. at or barely above parity. A ratio clustered that tightly around 1.0 across six partners, four institutions and three lineages is the signature of a class occupied by something both antibodies see: wild-type KMT2A, or non-specific coverage.

Why this is the strongest statement in the beat

Every other quantity here is a within-sample share, and a share cannot see a global change because there is no spike-in in this deposit. N/C is a ratio of two enrichments measured in the same sample over the same promoters, so it survives the objection that A73051 or any other sample has different coverage properties. It is also the deposit's own published criterion rather than one invented here, which means the threshold is not mine to choose.

Two honest weaknesses. First, N/C near 1.0 at a class could also mean the fusion and wild-type KMT2A both occupy it, which is a positive finding about wild-type KMT2A rather than a negative about the fusion. Second, the two antibodies are different clones with different efficiencies, so the absolute value of N/C is not calibrated; what carries information is that the same pair of antibodies reaches 2.8-4.8 at other classes in the same samples.

Facts and figures
criterionthe deposit's own Count_N_C fusion-specificity cutoff, log-fold-change 1.75, p < 1e-5
a73051 aml specific n over c0.964
scriptanalyses/q148-what-the-hoxa-negative-blast-occupies/controls148.py
statusrun
result summaryGMP-like class: 2 of 27 above N/C 1.75, one circular, one depth-inadequate
n over 1 75
AML_specific2.000AML_specific: 2Oncoprotein_Target16.0Oncoprotein_Target: 16ALL_specific18.0ALL_specific: 18denominator27.0denominator: 27
n over 1 75 — the numbers behind the chart
{
  "AML_specific": 2,
  "ALL_specific": 18,
  "Oncoprotein_Target": 16,
  "denominator": 27
}
aml specific n over c range
[
  0.787,
  2.3
]
aml specific above criterion
[
  "p318 (2.300, CIRCULAR - defined the class)",
  "TB986292 (1.814, depth-INADEQUATE)"
]
measurements
[
  {
    "quantity": "n_samples_above_the_deposits_own_n_over_c_criterion",
    "subject": "AML_specific (GMP-like) 40-promoter class",
    "value": 2,
    "units": "samples",
    "n": 27,
    "universe": "all GSE252378 KMT2A-fusion patient samples, KMT2A-N mean over KMT2A-C mean",
    "null": "N/C > 1.75 is the authors' own cutoff",
    "note": "p318 (circular) and TB986292 (depth-INADEQUATE); the other 25 lie between 0.787 and 1.266"
  },
  {
    "quantity": "n_samples_above_the_same_criterion",
    "subject": "Oncoprotein_Target 827-promoter class",
    "value": 16,
    "units": "samples",
    "n": 27,
    "universe": "same samples and libraries",
    "null": "N/C > 1.75",
    "note": "the instrument can reach the criterion; the GMP-like class specifically does not"
  }
]
How this could bite us back
  • If the GMP-like programme is held by wild-type KMT2A rather than by the fusion, then a menin inhibitor - which targets the interface both species use - would strip it from both, and the therapeutic reasoning that treats the GMP-like state as a fusion-driven vulnerability is resting on the wrong protein.
Still open
  1. Is wild-type KMT2A what sits on the GMP-like promoters? An N/C near 1.0 is consistent with it and the deposit cannot distinguish wild-type occupancy from background, because there is no KMT2A-null control library in this accession.
  2. Does the authors' own per-interval N/C test, run on the GMP-like intervals in all 27 samples rather than on the p279/p318 pair, agree with this class-mean version? Their code is public and this has not been run.
Every source behind this entry (2)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    Only 2 of GSE252378's 27 KMT2A-fusion samples show an N-over-C ratio above the deposit's own 1.75 criterion at the GMP-like promoter class, and one of them is the sample the class was defined in.

    What this evidence cannot show

    Promoter windows rather than the authors' peak intervals. The 1.75 threshold is the authors' log-fold-change cutoff from a peak-calling context and is applied here to a class mean, which is not the same operation. Different antibody clones, uncalibrated.

    Evidence that argues against this (1)
    • The authors' own N/C machinery operates per interval with a p-value, not on a class mean, so a class whose members individually pass could average below 1.75 and vice versa. This is a coarser instrument than theirs and disagreement with their peak-level result would not settle anything. Janssens DH 2024
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    The same antibody pair reaches N/C of 4.839 and 4.069 at the B-lineage class and 2.846 at the canonical class in the same deposit, so the near-parity at the GMP-like class is not a limitation of the readout.

    What this evidence cannot show

    The B-lineage class is 59.5% one tandem protocadherin cluster, so its high N/C values in B-lineage samples rest partly on 22 promoters of one 23.8 kb interval. The canonical class at 827 promoters is the cleaner control.

    Evidence that argues against this (1)
    • A class of 40 promoters has more sampling noise than one of 827, so the tight clustering of GMP-like N/C near 1.0 is partly a small-set effect; the counter-argument is that noise would widen the spread rather than pin it at parity. Q148 (Oncofusion Knowledge Graph) 2026
Connected to