PCGF1 loss flattens the fusion's genomic distribution - the knockout-over-control share ratio rises monotonically from 0.810 at the highest-signal decile to 1.136 at the lowest while both input pairs stay flat - and the drug does something spatially different: VTP-50469 evicts almost only at the top 0.1% (0.7554 against 0.9937 at the decile below). The drug acts at the peaks, the genotype acts in the bulk
also called PCGF1 loss flattens the fusion distribution · monotone from peaks into the bulk · where the displaced share went
Cells lacking a particular repressive protein hold less of the cancer-driving fusion at the places it normally sits most densely. Because these measurements are proportions, that missing amount has to turn up somewhere else, and this looked across the whole genome to find out where. It is not piling up at any particular new location - it spreads smoothly into the vast low-signal majority of the genome, and the effect grows steadily the weaker a region's normal signal is. The same analysis run on control samples with no antibody shows nothing, so this is not a technical artefact of sequencing depth.
The same thing, for a scientist
**Statement.** Cycle 102 found the fusion less concentrated at its strongest sites in a PCGF1 knockout — 16.3% lower at the parental top-FLAG bins, deepening to 21.4% at the knockout's own — and refuted relocation. Its replacement open question: *"A within-sample share must sum to 1, so a 21% deficit at the top sites is a 21% surplus somewhere, and nobody has looked at the rest of the genome."* Looked. **Cutting the 1,174,321-bin parental universe into the top 0.1% plus deciles of the remainder:** | stratum | NT share | C1 share | **FLAG C1/NT** | input q125 | input q124 | |---|---:|---:|---:|---:|---:| | TOP 0.1% | 0.008412 | 0.007045 | **0.8374** | 1.0967 | 1.0360 | | **D1** | **0.193581** | **0.156697** | **0.8095** | 1.0568 | 1.0132 | | D2 | 0.116826 | 0.111023 | 0.9503 | 0.9891 | 0.9967 | | D3 | 0.103370 | 0.102356 | 0.9902 | 0.9822 | 0.9958 | | D4 | 0.094975 | 0.096958 | 1.0209 | 0.9764 | 1.0012 | | D5 | 0.089326 | 0.092671 | 1.0374 | 0.9702 | 0.9934 | | D6 | 0.083973 | 0.089113 | 1.0612 | 0.9821 | 1.0007 | | D7 | 0.079402 | 0.085843 | 1.0811 | 0.9821 | 0.9988 | | D8 | 0.075072 | 0.082526 | 1.0993 | 0.9911 | 1.0002 | | D9 | 0.070345 | 0.078805 | 1.1203 | 1.0054 | 0.9965 | | D10 | 0.061679 | 0.070037 | 1.1355 | 1.0256 | 0.9931 | **`S1` — FLATTENING.** The ratio rises **monotonically across all ten deciles**, from 0.810 to 1.136. No single stratum absorbs more than **19%** of the total loss. **The fusion is redistributing from peaks into the bulk.** **`U1`, the control that decides it, passes.** Both input pairs are **flat** over the same strata — q124 spans 0.993–1.036, q125 spans 0.970–1.097 — while FLAG spans **0.810–1.136**. FLAG's largest deviation from 1 is **0.1905** against the inputs' **0.0967**, and, more tellingly, **the FLAG profile is monotone and neither input profile is.** This is not coverage. **And the loss is not where cycle 102 was looking.** The top 0.1% contributes a deficit of **0.0014**; **D1 contributes 0.0369, twenty-seven times more.** Every previous beat in this thread has stratified on `S_FLAG`, the top 0.1%. **The dominant signal is in the decile beneath it, which no beat has examined.** **AND THE OBVIOUS WORRY — that the whole thread has been looking in the wrong stratum — WAS TESTED BEFORE THIS SHIPPED AND IS REFUTED.** Computing VTP-50469's eviction per stratum in the control: | stratum | **NT VTP/DMSO** | share held | |---|---:|---:| | **TOP 0.1%** | **0.7554** | 0.84% | | D1 | 0.9937 | 19.4% | | D2–D10 | 0.991–1.016 | 79.8% | **The drug's effect is essentially confined to the top 0.1%.** That stratum holds **0.84%** of the signal and contributes **41%** of the total share lost under VTP; D1 holds 19.4% and contributes 24%. Per unit of signal the top stratum is affected some fifty-fold more. **So the two effects are spatially distinct, and both stratifications are right for their own question.** **The drug acts at the peaks** — which is why every beat in this thread was correct to stratify on `S_FLAG`. **PCGF1 loss acts in the bulk** — 0.0369 of the genotype deficit sits in D1 against 0.0014 at the top. A menin inhibitor and a PRC1.1 knockout move the fusion in different places, and nothing before this cycle could have shown that.
The verdict logic was wrong the first time, and the arithmetic that caught it
The first run returned S2 — RELOCATION, reporting that "D9 absorbs 619% of the top-stratum deficit". 619% is not a possible share of anything, and the nonsense was the diagnosis: I had divided every gain by the TOP_0.1pct deficit alone while the loss is dominated by D1, twenty-seven times larger. Against the correct denominator — the total across all four losing strata, 0.0451 — the largest gainer takes 19% and the verdict is S1.
Ninth criterion in this project sound in form and wrong in substance, after a reagent that could not distinguish its subject, a population never stated, a measure saturated at its sample size, a control not independent of the treatment, an or letting a dead clause decide, a clause written per-cohort and evaluated per-pool, a threshold test with no baseline count, and a majority test that passed on a sample of one. The new variant: a denominator that names only part of the quantity it divides.
What caught it was printing a percentage that could not exist. A ratio above 100% of a share is self-refuting in a way that a plausible-looking 45% would not have been, and I would probably have shipped the wrong verdict if the top stratum had been larger.
What the shares do not sum to, and what that means
The absolute share changes sum to −3.89 × 10⁻³ rather than zero. That is not rounding: each track's share is computed against its own genome-wide total, and roughly 0.4% of each track's signal falls outside the parental universe — bins the parental FLAG DMSO track did not cover. The knockout has slightly more of its signal out there than the control does.
So the flattening is measured within the parental universe and is very slightly understated by this accounting. The direction is unaffected; anyone recomputing an exact mass balance needs the out-of-universe fraction as an eleventh category.
And the compositional caveat declared before running still governs the reading. A deficit somewhere forces a surplus elsewhere. The existence of the surplus is arithmetic and carries no information. Only its shape does — and the shape here is monotone across ten strata, which is not something arithmetic alone produces.
| question | a 21% deficit at the top sites is a surplus somewhere - where? |
|---|---|
| verdict | S1 - FLATTENING. Monotone redistribution from peaks into the bulk; no stratum takes more than 19% of the loss. |
| downloads required | 0 |
| strata | top 0.1% of the 1,174,321-bin parental FLAG DMSO universe, then deciles of the remainder |
| U1 control | PASSES - FLAG max |ratio-1| is 0.1905 against the inputs' 0.0967, and the FLAG profile is MONOTONE while neither input profile is |
| largest gainer share of total loss PCT | 19 |
| THE LOSS IS NOT WHERE THE THREAD HAS BEEN LOOKING | the top 0.1% contributes 0.0014 of the deficit and D1 contributes 0.0369 - twenty-seven times more. Q117, Q122, Q124, Q125, Q130 and Q131 all stratify on S_FLAG, the top 0.1%. |
| my verdict logic was wrong first time | the first run returned S2 RELOCATION with 'D9 absorbs 619% of the top-stratum deficit'. 619% is not a possible share; I had divided gains by the TOP_0.1pct deficit alone while the loss is dominated by D1. Against the correct total of 0.0451 the largest gainer takes 19%. NINTH criterion in this project sound in form and wrong in substance; the new variant is a denominator that names only part of the quantity it divides. |
| shares do not sum to zero | absolute changes sum to -3.89e-3 because ~0.4% of each track's signal falls outside the parental universe, differentially between genotypes. The flattening is measured within the universe and is slightly understated. |
| absence class | measured-not-analysed |
| absence class note | Both FLAG tracks and both input pairs were binned and on disk; what had never been analysed is the genome-wide share profile, because every beat in this thread stratified only on the top 0.1%. |
FLAG C1 over NT — the numbers behind the chart
{
"TOP_0.1pct": 0.8374,
"D1": 0.8095,
"D2": 0.9503,
"D3": 0.9902,
"D4": 1.0209,
"D5": 1.0374,
"D6": 1.0612,
"D7": 1.0811,
"D8": 1.0993,
"D9": 1.1203,
"D10": 1.1355
}input q125 C1 over NT — the numbers behind the chart
{
"TOP_0.1pct": 1.0967,
"D1": 1.0568,
"D2": 0.9891,
"D3": 0.9822,
"D4": 0.9764,
"D5": 0.9702,
"D6": 0.9821,
"D7": 0.9821,
"D8": 0.9911,
"D9": 1.0054,
"D10": 1.0256
}input q124 C1 over NT — the numbers behind the chart
{
"TOP_0.1pct": 1.036,
"D1": 1.0132,
"D2": 0.9967,
"D3": 0.9958,
"D4": 1.0012,
"D5": 0.9934,
"D6": 1.0007,
"D7": 0.9988,
"D8": 1.0002,
"D9": 0.9965,
"D10": 0.9931
}losses — the numbers behind the chart
{
"TOP_0.1pct": -0.001368,
"D1": -0.036885,
"D2": -0.005803,
"D3": -0.001015,
"total": -0.04507
}drug versus genotype are spatially distinct
{
"ran": "before this entry shipped, because its own counter-evidence claimed the thread had been measuring the smaller half of its effect",
"verdict": "REFUTED for the drug, CONFIRMED for the genotype - they act in different strata",
"NT_VTP_over_DMSO_by_stratum": {
"TOP_0.1pct": 0.7554,
"D1": 0.9937,
"D2": 0.9909,
"D3": 0.993,
"D4": 1.0003,
"D5": 1.002,
"D6": 1.007,
"D7": 1.0107,
"D8": 1.0101,
"D9": 1.014,
"D10": 1.0158
},
"C1_VTP_over_DMSO_by_stratum": {
"TOP_0.1pct": 0.7655,
"D1": 1.0407,
"D2": 1.0051,
"D3": 0.997,
"D4": 0.9963,
"D5": 0.994,
"D6": 0.9934,
"D7": 0.9902,
"D8": 0.9873,
"D9": 0.9845,
"D10": 0.993
},
"absolute_share_lost_under_VTP_in_NT": {
"total": 0.005066,
"TOP_0.1pct": -0.002057,
"D1": -0.001215,
"D2": -0.001069,
"D3": -0.000724
},
"top_stratum_holds_pct_of_signal": 0.84,
"top_stratum_contributes_pct_of_drug_loss": 41,
"reading": "the drug acts at the PEAKS (top 0.1%, ~50-fold enriched per unit signal), PCGF1 loss acts in the BULK (D1 carries 0.0369 of the genotype deficit against 0.0014 at the top). Every beat in this thread was right to stratify on S_FLAG for the drug question."
}- PARTLY ANSWERED: the DRUG is confined to the top 0.1% (VTP/DMSO 0.7554 there against 0.9937 at D1), so S_FLAG was the right stratification for every beat that used it. The GENOTYPE effect is still in D1. What remains: no MARK has ever been measured at D1 - Q125's and Q130's gradients are all S_FLAG-only - so whether H2AK119ub, H3K27me3, H3K4me3 or H3K79me2 move in the bulk is unknown and computable from bins on disk.
- Is the monotone profile a scaling artefact? A ratio rising smoothly with falling signal is also what a dynamic-range difference produces. The inputs argue against it but have a much narrower range and may be unable to express the same artefact. A same-genotype ChIP replicate would settle it and none exists in this series.
- Does the flattening survive the drug? These are DMSO tracks only. The same profile under VTP-50469 is computable from the same directory and would say whether the drug acts on a flatter distribution or restores focus.
- What is in the 0.4% of signal outside the parental universe, and why does the knockout have more of it? It is a small fraction that breaks the mass balance, and it is the one part of the genome this stratification cannot see.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.analysis
Over the 1,174,321-bin parental FLAG DMSO universe cut into the top 0.1% plus ten deciles of the remainder, the PCGF1-knockout-over-control FLAG share ratio is 0.8374 at the top stratum and then 0.8095, 0.9503, 0.9902, 1.0209, 1.0374, 1.0612, 1.0811, 1.0993, 1.1203 and 1.1355 across deciles 1 to 10 - monotonically rising. The q125 input pair over the same strata gives 1.0967, 1.0568, 0.9891, 0.9822, 0.9764, 0.9702, 0.9821, 0.9821, 0.9911, 1.0054, 1.0256 and the q124 input pair gives 1.0360, 1.0132, 0.9967, 0.9958, 1.0012, 0.9934, 1.0007, 0.9988, 1.0002, 0.9965, 0.9931; neither is monotone. Absolute losses are 0.001368 at the top stratum, 0.036885 at D1, 0.005803 at D2 and 0.001015 at D3, totalling 0.045070, and the largest single gaining stratum takes 19% of that.
analyses/where-does-the-displaced-fusion-share-go/run.py and results.json, this repository; bins from C:/Temp/q117, C:/Temp/q124 and C:/Temp/q125, computed 2026-09-23 with zero downloads; FLAG tracks are GSM8123410 and GSM8123407, mislabelled in GEO as H2AK119ub rep1 (cycle 101) · Oncofusion knowledge graph, coordinator · internal analysis, reproducible from the repository 2026What this evidence cannot show
WITHIN-SAMPLE SHARE ONLY - GSE260742 has no usable spike-in, so a uniform global change is invisible and 'flattening' means the DISTRIBUTION of each sample's own signal is flatter, not that total fusion binding changed. SHARES ARE COMPOSITIONAL: the surplus is forced by the deficit and only its shape is informative, which is why the monotonicity and the input comparison carry the result rather than the surplus itself. n = 1 per genotype - there is no replicate of either FLAG track, so no stratum ratio has a confidence interval. The absolute changes sum to -3.89e-3 rather than 0 because roughly 0.4% of each track's signal lies outside the parental universe, differentially between genotypes. The PCGF1 knockout is one guide, one clone, no rescue, and the deposit contains no PCGF1 measurement. Strata are defined by PARENTAL signal from a third experiment, so a bin's stratum is fixed by the parental line rather than by either genotype being compared.
Evidence that argues against this (3)
- A monotone ratio across strata ordered by signal strength is the classic signature of a scaling or normalisation difference, not necessarily of biology. If the knockout track has a flatter dynamic range for a technical reason - different sequencing depth, different PCR duplication, a different signal ceiling in the TDF - the ratio would rise smoothly with decreasing signal exactly as observed. The inputs argue against it, but inputs have a much narrower dynamic range than a ChIP and may simply be unable to express the same artefact. A same-genotype ChIP replicate would settle it and none exists.
- Calling this 'losing focus' imports a mechanism the data does not contain. The measurement is a share profile across bins ranked by a third experiment's signal; it says the knockout's distribution is flatter, not that individual fusion molecules have become less able to find or hold their sites. A cell with fewer fusion molecules, or with a different proportion of cells expressing it, would produce the same flattening through a completely different route, and nothing here distinguishes population heterogeneity from per-molecule behaviour.
- I wrote that the thread may have been measuring the smaller half of its own effect, because D1 carries twenty-seven times the genotype deficit of the top 0.1%. TESTED BEFORE SHIPPING and it is REFUTED for the drug: VTP's eviction is 0.7554 at the top stratum and 0.9937 at D1, essentially confined to the peaks, with the top 0.1% contributing 41% of the drug-induced loss while holding 0.84% of the signal. So S_FLAG was the right stratification for every beat that used it. What survives is sharper than what I claimed: the DRUG acts at the peaks and the GENOTYPE acts in the bulk, and no entry had separated them. The residual worry is that D1 still holds 19.4% of the signal and nobody has looked at any MARK there - only at FLAG.
answers that entry's open question - the displaced share spreads monotonically into the bulk rather than relocating
PCGF1 loss flattens the fusion distribution, acting in the bulk while the drug acts at the peaks
- related to →KMT2A / MLL1 — the gene that breaks
the genomic distribution of KMT2A-fusion occupancy, stratified by parental signal
shows the drug eviction is confined to the top 0.1% of sites, which vindicates every beat in this thread stratifying on S_FLAG
answers that entry's closing gap - the first measurement of any mark outside the top 0.1%
CORRECTED in cycle 110: share and absolute profiles have identical shape, so that entry's monotone gradient is NOT undermined as a shape claim. What a share cannot support is any statement that a stratum above 1 gained signal rather than lost less than average.
restores that entry's SHAPE claim, which cycle 109 wrongly undermined, while leaving its redistribution reading exposed as an absolute-level claim a share cannot make