Cycle 104's M1 - 'no scoreable mark moves at D1, so the menin inhibitor's chromatin effect is proportionally a peak phenomenon' - DOES NOT SURVIVE. The same-condition floor at D1 is 2.4%, not the 10% borrowed from the top stratum, and all three scoreable marks exceed it (H3K4me3 9.8%, H3K79me2 7.8%, H3K27me3 4.8%). Two of the three exceed even GSE260742's OWN stratum-matched input floor, which was on disk the whole time. Cycle 104's alternative M2 fires instead
also called M1 does not survive · the marks do move at D1 · M2 fires, not M1 · the drug is not only a peak phenomenon
Last cycle concluded that a menin inhibitor changes the chromatin marks only at a tiny set of the very strongest sites, and not in the much larger region just below them. That conclusion used a 'how much do identical samples differ' yardstick measured at the strongest sites and applied to the larger region, where the true yardstick is about five times tighter. With the right yardstick, all three measurable marks do change in the larger region - and that region holds twenty-three times more signal.
The same thing, for a scientist
**Cycle 104 pre-registered `M1` and `M2` as alternatives.** `M1`: *the chromatin response is as confined as the eviction if every mark's VTP/DMSO ratio at D1 sits within the input-derived floor of 4-10%*. `M2`: *the thread has been measuring a fraction if any mark departs from 1 at D1 by more than that floor*. It scored `M1` as firing. **`M1` used a floor measured at the TOP stratum and applied it at D1.** Two independent instruments now say that is four to five times too generous. | mark, VTP/DMSO at D1 (NT) | ratio | deviation | vs measured same-condition D1 floor **2.42%** | vs GSE260742's OWN stratum-matched D1 input floor **5.68%** | |---|---:|---:|---|---| | **H3K4me3** | 1.0982 | **9.82%** | **EXCEEDS** | **EXCEEDS** | | **H3K79me2** | 1.0777 | **7.77%** | **EXCEEDS** | **EXCEEDS** | | **H3K27me3** | 1.0478 | **4.78%** | **EXCEEDS** | inside (but outside q124's 1.32%) | | H2AK119ub `PAIRED` | 1.0524 | 5.24% | unscoreable | unscoreable | | H2AK119ub `REP2` | 0.9784 | 2.16% | unscoreable | unscoreable | **So `M2` fires and `M1` does not. The drug's chromatin effect is not proportionally confined to the peaks.** **The cheapest refutation needed no new data and was inside cycle 103's own `results.json`.** That file stores both input pairs' share ratios at every decile. At D1 they read **1.0568 (q125) and 1.0132 (q124)** - deviations of **5.68% and 1.32%**. Cycle 104 used **9.67%**, the q125 figure *at the TOP stratum*, as a global floor. **Applying the stratum-matched number that was already on disk breaks `M1` for H3K4me3 and H3K79me2 without touching GSE299100 at all.** **And the obvious counter-explanation was tested and fails.** The worry is that a depth or duplicate-removal difference between the two libraries manufactures a decile gradient. Measured directly: an input pair differing **1.74x** in total coverage, scored on an external cut exactly as cycle 104 scored its marks, stays within **0.73%** across all ten deciles and is **not monotone**. At these depths, depth differences do not make decile gradients on an external cut. **What this costs and what it does not.** Cycle 103's spatial separation stands - VTP's effect on the FUSION is 24.5% at the peaks and 0.63% at D1, and 0.63% is inside the measured 2.42% D1 floor. **The fusion is evicted at the peaks only. The marks are not.** Those are now different statements, where cycle 104 had made them the same one.
Why this matters more than the ratios look
D1 holds 19.4% of the FLAG signal against the top stratum's 0.84% - 23 times more. Cycle 104 already reported that absolute share change at D1 exceeds TOP by 5-33x for every mark, and then discounted that observation with its own counter-evidence: "D1 has 100x the bins, so identical proportional behaviour puts more absolute change there - the test that separates them IS the ratio, and it says no."
The ratio now says yes. Both halves point the same way: the bulk response is larger in absolute terms and is not proportionally smaller. The discount no longer applies.
The shape across the bulk is a redistribution, not a uniform shift
Reading cycle 104's own table across strata for the control genotype: H3K4me3 runs 1.1235 (TOP), 1.0982 (D1), then 0.9324, 0.9189, 0.9088, 0.9018, 0.8957, 0.8944, 0.8862, 0.8815, 0.8744 (D10). H3K79me2 runs 0.8065, 1.0777, 1.0220, 0.9959, 0.9633, 0.9386, 0.9167, 0.8963, 0.8787, 0.8517, 0.8383.
Against a 2.4% floor, almost every one of those deciles is outside it. The pattern for H3K4me3 is a gain at the top two strata and a loss everywhere else; for H3K79me2 a loss at the peaks, a gain at D1-D2 and a loss from D4 down. Shares are compositional, so some of this is arithmetic - a gain somewhere forces a loss elsewhere - but the shape is not forced, and the shapes differ between the two marks, which arithmetic alone cannot produce.
What would have to be true for M1 to be right after all
The measured D1 floor comes from a MEN1/MLL1 transcription-factor-style ChIP pair in human MV4;11. Cycle 104's D1 values are histone-mark ChIPs in mouse MLL-AF9 cells. If broad histone-mark ChIP has a bulk same-condition floor several times wider than a sharp-peak factor ChIP does, M1 could survive.
That is a real gap and it is testable. It is not currently supported: GSE260742's own input pairs give 1.32-5.68% at D1, in mouse, in the same libraries the marks come from, and 5.68% is still below H3K4me3's 9.82% and H3K79me2's 7.77%. The verdict for those two marks does not depend on GSE299100 at all. H3K27me3's 4.78% does, and is labelled accordingly.
What must now be re-read
Any entry stating that the menin inhibitor's chromatin consequence is confined to the top 0.1%, or that S_FLAG was the right stratification for the MARKS, is now wrong in that clause. The corresponding claim about the FUSION - cycle 103's 0.7554 at TOP against 0.9937 at D1 - is unaffected and both its halves survive re-testing.
ana-the-marks-respond-proportionally-at-the-peaks-and-quantitatively-in-the-bulk-and-h3k79me2-reverses-sign is the entry carrying the superseded clause. Its title's 'respond proportionally at the peaks' is the part that fails; its 'quantitatively in the bulk' half is strengthened. It is not owned by this beat and is reported rather than edited.
| script | analyses/the-same-condition-noise-floor/run.py |
|---|---|
| status | run |
| result summary | M1 does not survive. Measured same-condition D1 floor 2.42%; H3K4me3 9.82%, H3K79me2 7.77% and H3K27me3 4.78% all exceed it. Two of the three also exceed GSE260742's own stratum-matched D1 input floor of 5.68%, so the verdict for those two needs no new data. |
| d1 floor measured | 0.0242 |
| d1 floor borrowed used by cycle 104 | 0.1 |
| d1 floor stratum matched from gse260742 | 0.0568 |
marks at d1 control genotype — the numbers behind the chart
{
"H3K4me3": 1.0982,
"H3K79me2": 1.0777,
"H3K27me3": 1.0478,
"H2AK119ub_PAIRED": 1.0524,
"H2AK119ub_REP2": 0.9784
}marks breaking the measured floor
[ "H3K27me3", "H3K4me3", "H3K79me2" ]
marks breaking the stratum matched borrowed floor
[ "H3K4me3", "H3K79me2" ]
inputs
[ "analyses/do-the-marks-move-in-the-bulk/results.json", "analyses/where-does-the-displaced-fusion-share-go/results.json", "GSE299100 six control samples" ]
- This widens rather than narrows the off-target surface for a menin inhibitor. If H3K4me3 and H3K79me2 shift by 8-10% of total share across the bulk of the genome, the drug's chromatin consequence is not confined to the oncogenic programme, and a normal cell with the same complexes would carry the same bulk shift without any therapeutic benefit.
- It also weakens a reassurance the project had been offering itself. 'The effect is confined to the peaks' is the shape a targeted interface inhibitor predicts and was being read as support for the menin-interface hypothesis. That support is withdrawn for the marks - though not for the fusion, whose eviction remains peak-confined on both halves of the re-test.
- A pharmacodynamic assay reading a bulk-averaged histone mark would now be expected to move under a menin inhibitor - which sounds useful, but a bulk mark that moves in every cell is a poor selectivity marker and could make a non-responding patient look engaged.
- No same-condition histone-mark replicate pair exists in either deposit, so the mark-specific bulk floor is still inferred from a factor ChIP. Which public menin-inhibitor deposit has two vehicle histone-mark libraries under uniform processing? That is the next blocking search, and it is the same shape as the one cycle 99 ran.
- If the marks move across the whole bulk, what is the mechanism? An eviction confined to the peaks cannot directly explain a several-percent share shift across 1.4 million bins; a global change in total mark abundance would, and within-sample share cannot see one. A spike-in-normalised menin-inhibitor histone ChIP would settle it and none is known to exist in this thread.
- The two marks that break the floor hardest at D1 are the two ACTIVATING marks, both rising. H3K27me3 rises least and H2AK119ub is unscoreable. Is the bulk response an activation signature rather than a repression one? That ordering was not predicted and is not explained.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
In GSE299100, over 2 kb hg38 bins, an external peak-set cut and two vehicle-treated non-targeting hairpins in MV4;11, the same-condition share-ratio floor at the D1 decile is 2.42% (MEN1 0.9758) and 0.82% (MLL1 0.9918), against 9.37% and 8.83% for the same pairs at the top 0.1% stratum.
What this evidence cannot show
Human MV4;11 (KMT2A::AFF1, acute MONOCYTIC with FLT3-ITD, not infant B-ALL), hg38, a menin-complex factor ChIP; cycle 104's marks are mouse mm10 histone ChIPs in an MLL-AF9 line. The transfer between the two is an assumption, stated, and the independent GSE260742-internal check in E2 is what keeps two of the three verdicts from depending on it.
Evidence that argues against this (1)
- A broad histone mark spread over kilobase domains may have a narrower bulk floor than a sharp-peak factor, not a wider one, because its signal is less concentrated and less sensitive to IP efficiency at summits. On that reading the 2.42% used here would be conservative and M1 would fail harder. Neither direction is measured; no same-condition histone-mark pair exists in either deposit. source
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
Cycle 103's results.json records the two GSE260742 input pairs at every decile; at D1 they are 1.0568 (q125) and 1.0132 (q124), deviations of 5.68% and 1.32%, while cycle 104's run.py sets FLOOR = 0.10 from 'the widest of the two input-pair deviations, cycle 102', which is the TOP-stratum figure.
What this evidence cannot show
These input pairs are NT versus PCGF1-knockout, i.e. a genotype contrast, so they are themselves an imperfect null and, like cycle 102's, a lower bound with no antibody. They are cited here as the stratum-matched version of the very number cycle 104 used, not as an ideal floor.
Evidence that argues against this (1)
- The two GSE260742 input pairs disagree four-fold at D1 (5.68% versus 1.32%), so taking the wider is a choice. Taking the narrower would break M1 for all three scoreable marks; taking the wider spares H3K27me3. The wider is used throughout here, which is the conservative direction for the conclusion being drawn. source 2026
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
Cycle 104 reports VTP/DMSO ratios in the control genotype at D1 of 1.0478 for H3K27me3, 1.0982 for H3K4me3 and 1.0777 for H3K79me2, with H2AK119ub giving +5.24% under the PAIRED convention and -2.16% under REP2 and therefore declared unscoreable at D1 by its own pre-registered U2.
What this evidence cannot show
n = 1 per mark per genotype per treatment except H2AK119ub, so none of these ratios has a confidence interval of its own; the floor is the only scale available, which is exactly why the floor's stratum matters.
Evidence that argues against this (1)
- Per this project's standing policy, no H2AK119ub number from GSE260742 may be quoted without both conventions. Both are given above and they straddle the new 2.42% floor - PAIRED outside, REP2 inside - so the mark remains unscoreable at D1 under the new floor exactly as under the old one, and no part of this entry's verdict rests on it. source 2026
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
An input pair whose two arms differ 1.74-fold in total coverage bases (9.516e9 against 5.467e9), scored on an external peak-set cut, stays within 0.73% of 1 across all ten deciles and is not monotone - so at these depths a library-depth difference does not manufacture a decile gradient under the design cycle 104 used.
What this evidence cannot show
One pair, one depth ratio, one cell line. It bounds the artefact at 1.74x depth difference; a larger imbalance is untested. The VTP and DMSO mark libraries in GSE260742 have depth ratios that were not extracted here.
Evidence that argues against this (1)
- The same two input libraries DO produce a monotone decile gradient spanning 0.9040 to 1.2349 when the strata are cut on the pair itself rather than externally. So the artefact is real and large; it is specifically the external design that is immune. If any beat in this thread self-stratified, E4 does not protect it - and that is why the self-stratification result is written as its own entry rather than a footnote here. source 2026