The AML-over-ALL retention ratio is 93.5% denominator: the GMP-like numerator barely moves across 23 samples while the B-lineage denominator varies fourteenfold, so the ratio measures loss of the B-lineage class and not gain of the GMP-like one
A popular way to say 'the fusion moved to myeloid genes' is to divide myeloid signal by lymphoid signal. Across 23 patients the myeloid number hardly changes at all and the lymphoid number changes enormously, so the division is really reporting the lymphoid number upside down.
The same thing, for a scientist
Over the **23** untreated, non-circular GSE252378 samples, at 2 kb promoters: | | range | CV | above 1.5x | below 1.0x | |---|---|---:|---:|---:| | `AML_specific` (GMP-like, 40 promoters) | **1.305 - 2.607** | **0.142** | 22 of 23 | 0 | | `ALL_specific` (B-lineage, 37 promoters) | **0.330 - 3.148** | **0.545** | 5 of 23 | 7 | The numerator's coefficient of variation is **3.8x smaller** than the denominator's. Decomposed on the log scale, **var(log ALL) = 0.3003** against **var(log AML) = 0.0209**, so the denominator supplies **93.5%** of the variance the ratio can express. And: - **Spearman(ratio, denominator) = -0.936, two-sided p = 5.7e-11** - **Spearman(ratio, numerator) = +0.104, two-sided p = 0.638** **The ratio is a rank-inverted copy of the B-lineage class.** A sample with a high AML/ALL has not gained GMP-like occupancy; it has lost B-lineage occupancy. Every statement of the form 'the oncoprotein redistributed to the GMP-like programme' that rests on this ratio has to be rewritten as 'the oncoprotein left the B-lineage programme', which is a different mechanism with different consequences.
Why this is not the same claim as q143's
ana-q143-enl-redistributes-to-the-gmp-like-peaks-rather-than-simply-persisting-and-both-instruments-agree-on-the-direction reports ENL AML/ALL = 2.06 as a retention ratio - a ratio of before-and-after densities within one drug pair - not as a cross-sample occupancy ratio. These are different quantities and this entry does not contradict that one.
But the structural warning transfers. q143's own numerator and denominator were ENL retention 0.548 at AML-specific peaks against 0.266 at ALL-specific, and 0.266 is the lowest of its ten cells. So there too the striking ratio is carried by the ALL-specific side falling further, and q143 says as much in its own prose - it calls the result a redistribution rather than a retention, which is the correct word. What this entry adds is that the ALL-specific class is the volatile one in 23 samples with no drug involved at all, so its volatility is a property of the class and not of the treatment.
The general form, which is why this has its own entry
A ratio is only a measurement of both its terms if both terms move. When one term's CV is a fraction of the other's, the ratio is a noisy transform of the volatile term, and naming the ratio after the stable term is a naming error that survives every significance test. The diagnostic is two lines: the CV of each term, and the Spearman of the ratio against each term. Here they give -0.936 and +0.104 and settle it.
| n | 23 |
|---|---|
| population | GSE252378 untreated, non-circular (p279/p318 excluded) samples |
| aml specific cv | 0.1424 |
| all specific cv | 0.5447 |
| var log aml | 0.02088 |
| var log all | 0.3003 |
| denominator share of ratio variance | 0.935 |
| script | analyses/q148-what-the-hoxa-negative-blast-occupies/controls148.py |
| status | run |
| result summary | the ratio is 93.5% denominator; Spearman -0.936 against the denominator |
aml specific range
[ 1.3048, 2.6067 ]
all specific range
[ 0.33, 3.1483 ]
measurements — the numbers behind the chart
[
{
"quantity": "share_of_log_ratio_variance_from_the_denominator",
"subject": "AML_specific / ALL_specific promoter-share ratio",
"value": 0.935,
"units": "fraction",
"n": 23,
"universe": "GSE252378 untreated non-circular samples; 40 AML_specific and 37 ALL_specific promoters, KMT2A-N",
"null": 0.5,
"note": "var(log ALL) 0.3003 vs var(log AML) 0.0209"
},
{
"quantity": "spearman_rho",
"subject": "AML/ALL ratio against its ALL_specific denominator",
"value": -0.936,
"units": null,
"n": 23,
"universe": "same 23 samples",
"null": 0,
"note": "two-sided p = 5.7e-11"
},
{
"quantity": "spearman_rho",
"subject": "AML/ALL ratio against its AML_specific numerator",
"value": 0.104,
"units": null,
"n": 23,
"universe": "same 23 samples",
"null": 0,
"note": "two-sided p = 0.638"
},
{
"quantity": "coefficient_of_variation",
"subject": "AML_specific class share across samples",
"value": 0.142,
"units": null,
"n": 23,
"universe": "same 23 samples",
"null": null,
"note": "against 0.545 for the ALL_specific class in the same samples"
}
]- If a biomarker for lineage switch is built on an AML-over-ALL occupancy ratio, it will fire on any sample that has lost B-lineage occupancy for any reason, including poor sample quality, and it will not distinguish that from a myeloid programme being acquired.
- Does the same decomposition hold at the authors' peak intervals? If the GMP-like numerator moves at 20 kb scale but not at 2 kb, that localises the signal to gene bodies and enhancers and would be a positive finding rather than a correction.
- Is there any published KMT2A-r occupancy ratio in this literature whose numerator has been shown to move? The corpus has several ratios and none of them reports the CV of its terms.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Across 23 untreated non-circular GSE252378 samples the GMP-like promoter class has CV 0.142 and range 1.305-2.607 while the B-lineage class has CV 0.545 and range 0.330-3.148; the ratio correlates with the denominator at Spearman -0.936 and with the numerator at +0.104.
analyses/q148-what-the-hoxa-negative-blast-occupies - results148_controls.json C6_what_drives_the_ratio · Q148 (Oncofusion Knowledge Graph) · this repository 2026What this evidence cannot show
40 and 37 promoters. The log-variance decomposition assumes the two terms are the only inputs, which is true by construction for the ratio but says nothing about which term is biologically informative. Both classes are derived from one patient pair.
Evidence that argues against this (1)
- A stable numerator is not evidence that the numerator is uninformative - a class occupied at 1.3-2.6x in every sample could be a genuine constitutive target set. The reason to doubt that here is separate: against a coverability-matched null the class is not enriched in any of the 23, so the stable 1.3-2.6x is what an equally coverable promoter set gives. Q148 (Oncofusion Knowledge Graph) 2026
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The deposit's paper frames the lineage-switch result as the oncoproteins preferentially ACTIVATING GMP genes, which is a gain-of-occupancy framing.
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 · Nature Communications 15:9341 2024What this evidence cannot show
The word 'preferentially' is itself relative and the paper's own figures show both classes; this entry is about how a ratio should be described, not about whether the authors described theirs wrongly.
Evidence that argues against this (1)
- The authors also state that in lineage-switching samples the oncoprotein LEVELS are reduced, which is a loss framing and is consistent with the denominator reading. Their sentence contains both halves; a reader quoting only 'preferentially activate GMP genes' drops the half that matters. Janssens DH et al. 2024
different quantity (retention within a drug pair) but the same denominator-driven structure
explains why a stable 1.3-2.6x numerator is not occupancy
the volatile denominator is itself 59.5% one 23.8 kb interval