start / graph / ana-q135-the-same-condition-h2ak119ub-floor-spans-20-fold-across-arms-of-one-deposit

AnalysismoderateQ135-the-clean-pair-and-the-two-uncomputed-mark-channels

The same-condition H2AK119ub noise floor in GSE260742 spans twentyfold across the deposit's three experimental arms - D1 |R-1| of 1.89% in the PCGF1-KO clone, 10.49% in the non-targeting control and 37.04% in the MLLAF9 Control arm - at one mark, one lab, one pairing shape and one external stratifier. A single same-condition floor for this deposit does not exist.

also called the floor is arm-dependent · 20x across arms · no single floor exists

When the same experiment is repeated, the two runs never match exactly, and how badly they disagree is the yardstick for deciding whether a real effect is real. In one public dataset that yardstick turns out to be twenty times longer in one part of the experiment than in another, so borrowing one number for the whole dataset silently changes what counts as a finding.

The same thing, for a scientist

**Measured on the deposit's own repeated samples, with the pairing shape held fixed.** Each pairing is a `.filt.bam.tdf` track against a bare `.tdf` track from the same submission batch - the exact comparison cycle 107 made - scored with the identical statistic (within-sample shares, `R(S) = sum share_A / sum share_B`, floor = `|R-1|`) and the identical EXTERNAL stratifier, `q117_bin_FLAG_DMSO.tsv` (GSM8123359), which is no pair's member. | arm | n conditions | TOP 0.1% median | **D1 median** | |---|---:|---:|---:| | NTvKO **PCGF1-KO clone C1** | 3 | 15.97% | **1.89%** | | NTvKO **non-targeting control** | 3 | 19.49% | **10.49%** | | **MLLAF9 Control** (this beat) | 3 | 37.37% | **37.04%** | | *cycle 107 as published, both NTvKO genotypes pooled, 5 of 6 conditions* | 5 | *15.97%* | ***3.03%*** | **19.6-fold from the tightest arm to the widest, at D1.** The MLLAF9 Control arm's floor at D1 (37.04%) is larger than cycle 107's floor at the TOP 0.1% stratum (15.97%). **This is not an implementation difference.** The same script re-scored cycle 107's own NTvKO pair end to end and reproduced all five of its published per-condition values **exactly** - 22.37/10.49, 5.90/3.03, 15.97/1.89, 29.23/9.06, 3.58/0.40 for TOP/D1 - and additionally recovered the sixth condition cycle 107's universe gate had dropped (NT_SGC, 19.49/10.66). A difference between two arms measured by one script that reproduces the reference bit-for-bit is a difference between two arms. **Why it matters operationally.** Cycles 100-107 score claims against a floor. Which floor is chosen decides the verdict, and the choice has been made by which arm happened to have been binned, not by which arm the claim is about.

Go deeper

The per-condition numbers, so the medians can be checked

MLLAF9 Control arm, rep1 (.filt.bam.tdf) against rep2 (bare .tdf), same Mar-2024 submission batch:

condition	rep1	rep2	TOP	D1
DMSO	GSM8123365	GSM8123377	37.37%	37.04%
SGC0946 5 uM	GSM8123366	GSM8123378	62.63%	45.72%
VTP50469 500 nM	GSM8123367	GSM8123379	13.70%	22.25%

NTvKO arm, paired (.filt.bam.tdf) against rep2 (bare .tdf), re-scored here:

condition	TOP	D1	note
NT DMSO	22.37%	10.49%	reproduces cycle 107 exactly
NT SGC	19.49%	10.66%	recovered - cycle 107's gate dropped it
NT VTP	5.90%	3.03%	reproduces cycle 107 exactly
C1 DMSO	15.97%	1.89%	reproduces cycle 107 exactly
C1 SGC	29.23%	9.06%	reproduces cycle 107 exactly
C1 VTP	3.58%	0.40%	reproduces cycle 107 exactly

Three conditions per arm is three conditions. The spread is shown so the reader can see how stable each arm's floor is; it is not a confidence interval and none is computed from it.

What this does NOT establish

It does not establish that the MLLAF9 Control arm's data is worse. Rank concordance between the members of every pair scored here is moderate and similar across arms (Spearman 0.65-0.88 over the shared bin universe, NTvKO included), so the arms are not separated by an obvious quality gulf; what differs is how the share mass redistributes across strata. It also does not establish which arm's floor is 'right'. The claim is narrower and harder to escape: one number cannot serve all three.

THE RENORMALISATION BOUND, AND IT APPLIES TO EVERY NUMBER IN THIS ENTRY. Within-sample shares sum to 1 and GSE260742 has NO USABLE SPIKE-IN (Q117 verified zero hits for the entire spike-in vocabulary across all 419,824 characters of its 158 sample records, and no scale factor is deposited), so a purely local change leaks into every other stratum by arithmetic. A FLOOR COMPUTED THIS WAY CANNOT DISTINGUISH 'the bulk moved' from 'the peaks moved and the shares renormalised'. A TIGHTER NUMBER IS NOT THEREBY A CLEANER ONE.

Facts and figures

deposit	GSE260742
assembly	mm10
bin size bp	2000
statistic	within-sample share ratio R(S), floor = \|R-1\| in per cent
stratifier	q117_bin_FLAG_DMSO.tsv (GSM8123359), EXTERNAL to every pair
span d1 fold	19.6
n conditions per arm	3

arm floor d1 PCT

arm floor top PCT

cycle107 published

arm floor d1 PCT — the numbers behind the chart

{
  "NTvKO_PCGF1_KO_C1": 1.89,
  "NTvKO_non_targeting": 10.49,
  "MLLAF9_Control": 37.04
}

arm floor top PCT — the numbers behind the chart

{
  "NTvKO_PCGF1_KO_C1": 15.97,
  "NTvKO_non_targeting": 19.49,
  "MLLAF9_Control": 37.37
}

cycle107 published — the numbers behind the chart

{
  "TOP": 15.97,
  "D1": 3.03,
  "n_conditions": 5
}

Still open

Is the arm-to-arm spread driven by genotype, by culture batch, or by antibody lot? The deposit does not record an antibody lot (`antibod` occurs zero times in its 158 sample records), so this cannot be answered from GEO and needs the depositors.
Does the same 20-fold arm span appear in a deposit that DOES carry a spike-in, where abundance and redistribution can be separated?
Which arm's floor should a claim about the MLLAF9 Control arm be scored against - its own (37.04%, which would void most effects in this thread) or a pooled one?

Every source behind this entry (1)

E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
In GSE260742 (Mus musculus, mm10, murine MLL-AF9 myeloid cells), the same-condition H2AK119ub floor |R-1| at the D1 decile of an external FLAG-ChIP stratifier, measured from `.filt.bam.tdf` against bare `.tdf` pairs of one submission batch, is 1.89% (PCGF1-KO clone C1, n=3 conditions), 10.49% (non-targeting control, n=3) and 37.04% (MLLAF9 Control arm, n=3) - a 19.6-fold span. The same script reproduces all five of cycle 107's published per-condition values exactly.
Q135 re-analysis of GSE260742: the same-condition H2AK119ub floor decomposed into processing-marker and submission-batch terms across three experimental arms, plus first per-mark floors for H3K79me2 and H3K27me3. analyses/the-clean-pair-and-the-two-uncomputed-mark-channels/{PREREGISTRATION.md,run.py,results.json}, run 2026-09-23. 21 tracks downloaded one at a time, binned at 2 kb with Q117Bin, each raw .tdf deleted before the next fetch; 2,322,896,326 bytes transferred, 205,959,406 bytes peak raw disk. All 21 accessions re-verified live against the GEO record before use. · Iqbal Lab oncofusion knowledge graph 2026
What this evidence cannot show
SYSTEM: murine MLL-AF9 (KMT2A::MLLT3) myeloid blood cells, Mus musculus, mm10, one lab, one deposit - NOT the infant B-lineage KMT2A::AFF1 disease this graph is about. n = 3 conditions per arm, one cell system; the spread across conditions is reported so stability is visible and IS NOT A CONFIDENCE INTERVAL. THE RENORMALISATION BOUND, AND IT APPLIES TO EVERY NUMBER IN THIS ENTRY. Within-sample shares sum to 1 and GSE260742 has NO USABLE SPIKE-IN (Q117 verified zero hits for the entire spike-in vocabulary across all 419,824 characters of its 158 sample records, and no scale factor is deposited), so a purely local change leaks into every other stratum by arithmetic. A FLOOR COMPUTED THIS WAY CANNOT DISTINGUISH 'the bulk moved' from 'the peaks moved and the shares renormalised'. A TIGHTER NUMBER IS NOT THEREBY A CLEANER ONE. The three arms differ in genotype AND in which cultures were grown, so 'arm' is a composite nuisance and this entry does not decompose it further.
Evidence that argues against this (2)
- Genome Research 2012
- Cell Reports 2014

Connected to

← supportsScoring cycle 104's mark panel against a floor measured in its OWN arm: 9 of 10 mark-by-arm cells at D1 land inside the floor's own spread, so D1 is not resolvable from this deposit at all - and the same mark ratios give 0, 4 or 4 breaking marks depending only on which floor was imported, so cycle 104's pass, cycle 107's F1 and the GSE299100 beat's failure were all measuring the ruler rather than the marks
applies that beat's arm-level finding to the mark panel and shows the practical consequence: three published verdicts were measuring the ruler
← supportsHolding mark, arm and drug conditions fixed and varying only which confound the replicate pair carries: the floor ranges are DISJOINT at D1 and D10, a processing-confounded pair inflates D1 2.20x while a batch-confounded pair inflates D10 2.21x, and the direction reverses by stratum - so the floor is a property of the pair, not of the mark or the arm, and every floor this project has published came from the pair type that is wrong for D1
explains the arm-to-arm spread: different arms offer different pair types, and pair type moves the floor 2.2x with disjoint ranges at D1 and D10
related to →KMT2A::MLLT3 (MLL-AF9), t(9;11)(p21.3;q23.3)
murine MLL-AF9 is the fusion in every arm measured here
related to →PCGF1 — the PRC1.1 scaffold whose LOSS causes menin-inhibitor resistance, so it is a restraint on the leukaemia rather than a dependency of it
the PCGF1-KO clone is the arm with the TIGHTEST floor, 1.89% at D1
related to →Plain-English: chromatin and epigenetics — the sticky notes on the DNA, and which ones matter here
a noise floor for a histone-mark ChIP measurement
← supportsCycle 107's headline same-condition floor of 3.03% at D1 is a median over a MIXED population of two genotypes. Restricted to the non-targeting control genotype it is 10.49%, so the borrowed 10% floor cycle 107 declared 'too generous by 3.3x' was in fact about right there - cycle 107's own F2, not its F1.
the arm span is the general statement; this is its effect on the one number that has propagated
← supportsWithin one arm, swapping the processing-marker confound for a submission-batch confound takes the same-condition H2AK119ub floor from 37.04% to 16.84% at D1 (tighter, 0.45x) and from 37.37% to 61.89% at the TOP 0.1% stratum (worse, 1.66x). The brief's question - tighter, wider, or unchanged - has no single answer, and it is worse exactly where this thread scores its claims.
the within-arm pairing comparison behind the arm-span entry
← supportsEvery noise floor in this thread is computed from within-sample shares, which sum to one. A purely local change therefore leaks into every other stratum by arithmetic, so no floor measured this way can distinguish 'the bulk moved' from 'the peaks moved and the shares renormalised'. A tighter floor is not thereby a cleaner one.
the bound under which the arm-span measurement must be read
← supportsThe first same-condition H3K79me2 noise floor measured anywhere in this thread: 11.17% at D1 and 26.44% at the TOP 0.1% stratum, against 45.72% and 62.63% for H2AK119ub in the same arm, the same condition and the same pairing shape - 4.1x tighter. A mark floor transferred from H2AK119ub does not describe H3K79me2.
a second mark measured in the arm whose H2AK119ub floor is the widest
← related toAcross every same-condition H2AK119ub replicate pair in GSE260742 - nine pairs, three arms, both pairing shapes - rank concordance over the shared bin universe runs 0.6481 to 0.8796 and never reaches 0.88. The floors this thread scores against are all computed from pairs that are only moderately concordant.
shows the arm-to-arm floor span is not explained by one arm being noisier
← related toHypothesis: the same-condition floor measured this way is largely a property of where the EXTERNAL stratifier puts its top bins relative to each library's own coverage, not of how well the two replicates agree - which would explain a twentyfold arm span sitting on top of rank concordances that barely differ.
the candidate mechanism for the twentyfold arm span: if it holds, the span is a property of the stratum definition rather than of the replicate pairs
← testsRe-score all nine GSE260742 same-condition floors under two further external stratifiers already binned on disk - KDM2B DMSO and the Oct-2024 input - to decide whether the twentyfold arm span is a property of the replicate pairs or of the stratum definition. Zero downloads, one re-run of an existing script.
whether the twentyfold arm span survives a change of stratifier
← depends onEighteen libraries, no replicate: what the only absolute-scale chromatin dataset in this project is structurally unable to say, and why this project's own floors may not be borrowed to fill the gap
the measured 20-fold spread is why no floor may be imported into this deposit