The two VTP-50469 H3K4me3 libraries hold 3.3% and 3.8% of their reads in the DMSO-defined peak set against 8.9% for DMSO and 9.4% for EPZ-5676 - one global fact that explains every raw locus difference in the SEM menin-inhibitor arm, and that cannot be resolved into drug effect or library quality
One number explains almost everything that looked like a drug effect here: the two menin-inhibitor samples have about two and a half times less of their data sitting on the sharp peaks the mark normally makes. That could be the drug genuinely spreading the mark out, or it could be that those two samples simply worked less well. The deposited data cannot tell the difference, and saying which it is would be guessing.
The same thing, for a scientist
Share of each SEM H3K4me3 library's fragments falling in the top-0.1% DMSO-defined H3K4me3 bin set: untreated 8.9%/9.4%, DMSO 8.9%/8.9%, EPZ-5676 9.4%/9.5%, **VTP-50469 3.3%/3.8%**. The same picture in the scale-free top-1%-of-bins statistic: 0.594/0.496 untreated, 0.561/0.523 DMSO, 0.585/0.546 EPZ, **0.258/0.303 VTP**. Correlation to IgG rises in the same two libraries (0.260/0.243 vs 0.184-0.201 elsewhere). Genome-wide Pearson reproduces it: every DMSO-vs-VTP and EPZ-vs-VTP pair (r 0.638-0.697) sits below every DMSO-vs-EPZ pair (0.777-0.836), and the two VTP replicates correlate with each other (0.621) **less** than either correlates with DMSO - the signature of lost structure rather than of a coherent new state.
The two readings, and why the data cannot choose
Reading A - biology. Menin displacement removes the fusion complex, focal H3K4me3 dissolves, and the library becomes background-dominated. This is a real prediction of the menin-interface model.
Reading B - technical. These two libraries have lower signal-to-noise for reasons unrelated to the drug: 72 h at 30 uM of a cytotoxic agent kills and stresses cells, dying cells tagmentate differently, and CUT&Tag efficiency drops.
The discriminator would be a spike-in or a matched second mark in the same cells. This deposit has neither in the SEM drug arms: there is no H3K27ac, no H3K79me2, no RNAP2 and no menin library under VTP-50469 (see absence-q140-the-sem-menin-inhibitor-arm-of-gse159608-carries-only-h3k4me3). So the confound is unresolvable from the deposit, and any raw cross-arm locus comparison inherits it. 12
What survives the confound
A uniform change in peak share is multiplicative and global: it scales every site's share by the same factor, so it cancels in a ratio between two site classes read inside the same library. That is why the surviving measurement in this beat is a within-library class contrast and not a locus-by-locus comparison. Reading B, if true, damages the raw comparison completely and the class contrast not at all. 1
A caution about reading A as support for the hypothesis
A 2.5-fold global collapse of focal H3K4me3 in a KMT2A::AFF1 line is larger than the menin-interface model predicts. Most H3K4me3 in any cell is deposited by SETD1A/B and by wild-type KMT2A at housekeeping promoters that have no fusion on them; a menin inhibitor should not touch those. A drug effect of this size at GAPDH is therefore evidence against reading A, not for it - which is a second, independent reason to suspect the libraries and a reason this entry does not claim a global drug effect.
| script | analyses/sem-arm-identity-from-signal/run2.py |
|---|---|
| status | run |
| verdict | unresolvable from this deposit; requires spike-in or a second mark in the same cells |
peak share — the numbers behind the chart
{
"GSM4835757": 0.0891,
"GSM4842299": 0.0938,
"GSM5139709": 0.0886,
"GSM5139710": 0.0887,
"GSM5139711": 0.0941,
"GSM5139712": 0.0945,
"GSM5139713": 0.0331,
"GSM5139714": 0.0378
}top1pct bin share — the numbers behind the chart
{
"GSM5139709": 0.561,
"GSM5139710": 0.523,
"GSM5139711": 0.585,
"GSM5139712": 0.546,
"GSM5139713": 0.258,
"GSM5139714": 0.303
}corr to IgG — the numbers behind the chart
{
"GSM5139709": 0.2009,
"GSM5139710": 0.1963,
"GSM5139711": 0.1835,
"GSM5139712": 0.1946,
"GSM5139713": 0.2595,
"GSM5139714": 0.2426
}within arm r — the numbers behind the chart
{
"DMSO": 0.7892,
"EPZ5676": 0.8272,
"VTP50469": 0.6211
}- If reading A is right, the menin inhibitor's chromatin effect extends far beyond fusion targets and the drug's therapeutic window is narrower than the fusion-selectivity story implies.
- If reading B is right, every published figure normalising these two libraries against DMSO overstates the drug's effect, and any biomarker derived from them will not reproduce in a better-controlled experiment.
- Do the VTP-50469 arms in the other cell lines of GSE159608 show the same peak-share drop? If it is specific to SEM it is more likely biology; if it is universal across lines and marks it is more likely the exposure.
- Is there any deposited CUT&Tag dataset anywhere with a spike-in and a menin-inhibitor arm? That single dataset would settle a question this project has now hit three times.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
The two SEM VTP-50469 H3K4me3 libraries hold 3.3% and 3.8% of their fragments in the DMSO-defined top-0.1% H3K4me3 bin set, against 8.9-9.5% for the untreated, DMSO and EPZ-5676 libraries, and correlate more strongly with IgG.
sem-arm-identity-from-signal: run2.py section C and results2.json (this repository) · Q140-sem-menin-arm · Oncofusion knowledge graph 2026What this evidence cannot show
Peak set defined from the DMSO arm only, so the drug arms are scored on a set they did not help define. Both statistics are within-library shares and therefore independent of depth; library depths are 2.83M and 2.16M fragments for the VTP arm, within the range of the others (1.65M-3.33M), so this is not a depth artefact.
Evidence that argues against this (1)
- The depositors published from these libraries and did not report them as failed, which is weak evidence for reading A. It is weak because a paper reporting a drug-induced loss of focal signal and a paper with two lower-quality libraries look the same from outside. Janssens DH et al. 2021
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
VTP-50469 kills MLL-rearranged cells and induces differentiation and apoptosis, so a 72-hour 30 uM exposure is expected to change the cell population that the assay samples.
A Menin-MLL Inhibitor Induces Specific Chromatin Changes and Eradicates Disease in Models of MLL-Rearranged Leukemia · Krivtsov AV, Evans K, Gadrey JY, Eschle BK, Hatton C, Uckelmann HJ, et al., Armstrong SA · Cancer Cell 36:660-673.e11 2019What this evidence cannot show
Establishes only that reading B has a plausible mechanism; it does not measure CUT&Tag efficiency under the drug, which nobody has reported for this system.
Evidence that argues against this (2)
- This cuts against the entry's own headline as hard as it supports its caution. If a 72-hour 30 uM exposure changes which cells the assay samples, then ANY difference observed in the VTP arm - including the 23% depletion this beat reports - may be a change in population composition rather than a change in chromatin, and the within-library class ratio cancels only the uniform part of that.
- The cited pharmacology is from MLL-rearranged models generally, not from SEM at 30 uM for 72 h. Whether this specific line at this specific exposure is differentiating, dying or merely stressed is not measured anywhere in the deposit - there is no viability, apoptosis or differentiation readout accompanying these libraries.
The 2.5-fold peak-share drop is the confound the within-library contrast is built to cancel
A uniform peak-share drop under depth normalisation is what produced the spurious three-fold locus effect
The confound is unresolvable because no second mark and no spike-in exist in the VTP arm
Reproducing or failing to reproduce the peak-share collapse decides drug effect versus library quality