start / graph / ana-q135-removing-the-processing-confound-makes-the-floor-tighter-in-the-bulk-and-worse-at-the-peaks

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

Within 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.

also called the split answer · tighter in the bulk, worse at the peaks · T4 within the arm

Two repeats of an experiment were compared in two ways, each removing a different bookkeeping difference between them. Removing one made the repeats agree better across most of the genome and noticeably worse at the small number of hottest spots - which are the spots every conclusion in this project depends on.

The same thing, for a scientist

**MLLAF9 Control arm, three conditions, arm held fixed, only the pairing changes:** | pairing | members | `.filt.bam` marker | submission batch | TOP 0.1% median | D1 median | |---|---|---|---|---:|---:| | `P_marker` | rep1 vs rep2 | **differs** | same (Mar 2024) | 37.37% | **37.04%** | | `P_batch` | rep2 vs rep3 | same (both bare) | **differs** (Mar vs Oct) | **61.89%** | 16.84% | | `P_both` | rep1 vs rep3 | **differs** | **differs** | *aborted* | *aborted* | `P_batch` is the pair the dispatch brief identified as removing the processing confound, and its six accessions were verified live and are exactly right. **But it does not isolate replicate variance**: rep3 is a different submission batch seven months later (see the companion entry). So the honest reading of the table is *one nuisance traded for another*, and the trade is not uniform across strata. **Against cycle 107's published 15.97% / 3.03%, `P_batch` is WIDER at both strata** (3.88x and 5.56x) - the pre-registered T2 outcome. **But that comparison is arm-confounded**: `P_marker`, which is cycle 107's exact pairing shape moved into this arm, is also far wider (2.3x and 12.2x). Most of the gap between this beat and cycle 107 is the arm, not the confound, and T2 must not be read as a statement about processing. **`P_both` was refused, under both the pre-registered and the amended universe gate** (90.43-91.38% genuine track agreement). Its refusal is reported, not worked around.

Go deeper

Why the TOP stratum is where this bites

The TOP 0.1% stratum is 1,100-1,200 bins of the external FLAG track's highest coverage - the fusion's own strongest sites, and the set every eviction, silencing and knockout claim in cycles 100-107 is scored on. A floor of 61.89% there means a pair of same-condition replicates of one culture disagree by more than half at the peaks. No effect measured at that stratum in this arm at less than ~62% is distinguishable from replicate-plus-batch variation by this instrument.

The D1 decile, by contrast, is ~110,000 bins and averages the noise down - which is why it moves the other way.

The pre-registered outcomes, scored

outcome	condition	fired?
T1 tighter	below cycle 107 at both strata	no
T2 wider	above cycle 107 at both strata	yes, and it is arm-confounded
T3 unchanged	within 1.5x at both	no
T4 split	opposite directions	yes, within the arm
D-BATCH / D-MARKER / D-ADDITIVE	needs all three pairings	unscoreable - `P_both` was refused

The decomposition the beat was designed around therefore did not complete. Two of its three legs stand; the third is missing because the data would not support it, and that is reported rather than patched.

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

arm	MLLAF9 Control
n conditions	3
preregistered outcome fired	T2 against cycle 107, T4 within the arm
decomposition status	INCOMPLETE - P_both refused, so D-BATCH/D-MARKER/D-ADDITIVE are unscoreable

P marker

{
  "members": "GSM8123365/66/67 vs GSM8123377/78/79",
  "TOP_pct": 37.37,
  "D1_pct": 37.04
}

P batch

{
  "members": "GSM8123377/78/79 vs GSM8549830/32/31",
  "TOP_pct": 61.89,
  "D1_pct": 16.84
}

P both

{
  "members": "GSM8123365/66/67 vs GSM8549830/32/31",
  "status": "REFUSED - track agreement 90.43-91.38%"
}

Still open

Is the TOP-stratum floor of 61.89% a property of the Oct-2024 batch's shallower libraries rather than of batch per se? Binning a second Oct-2024 H2AK119ub pair would separate them, and the deposit has only one per condition.
Can `P_both` be rescued by restricting to bins both batches cover at depth, and does the restriction change the other two pairings' floors?

Every source behind this entry (1)

E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
In GSE260742's MLLAF9 Control arm (n=3 conditions), the same-condition H2AK119ub floor |R-1| is 37.37% (TOP 0.1%) / 37.04% (D1) for the marker-differing pairing rep1-vs-rep2, and 61.89% / 16.84% for the marker-matched batch-differing pairing rep2-vs-rep3. The doubly-differing pairing rep1-vs-rep3 failed the universe gate at 90.43-91.38% track agreement in all three conditions and was refused.
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. `P_batch` is scored on a universe 3.5% smaller relative to its own tracks than cycle 107's was (stratifier retention 95.28-95.39%), and a floor measured on a smaller, better-covered universe is not automatically comparable to one measured on a larger one. `P_batch` trades a processing marker for a submission batch and is NOT a clean replicate floor.
Evidence that argues against this (2)
- Genome Research 2012
- Cell Reports 2014

Connected to

supports →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.
the within-arm pairing comparison behind the arm-span entry
related to →Revumenib (SNDX-5613, Revuforj) — first approved menin–KMT2A interface inhibitor
VTP50469 is the tool menin inhibitor in one of the three conditions scored; revumenib is its clinical successor
← supportsThe GSE260742 H2AK119ub pair that shares a processing state - rep2 against rep3, both bare `.tdf` - is not an unconfounded replicate pair: rep3 was submitted 2024-10-01 and rep2 2024-03-03, through a different file-naming pipeline. No pairing in this deposit isolates replicate variance alone.
the confound structure that makes the Task-1 answer a trade rather than a removal
← supportsThe universe-overlap gate this thread inherited from cycle 107 divides by a track and multiplies by the stratifier, so it also measures whether both libraries are as deep as the stratifier - a property of the stratifier, not of the pair. It aborted 13 of 16 comparisons here, and cycle 107's own pair passed it by 0.25 percentage points while losing more bins to the stratifier than any pair it refused.
the gate decides which pairings in the Task-1 decomposition are scoreable
← supportsGSM8123377 and GSM8123403 carry the same supplementary filename but for the `.filt.bam` marker, yet GEO titles one 'MLLAF9 Control - DMSO H2AK119ub rep2' and the other 'NTvKO Control - DMSO H2AK119ub paired'. Scored directly, they are DIFFERENT libraries (Spearman 0.83), so the collision is a naming accident and both arm titles stand.
the control that licenses Task 1's pairing as within-arm rather than cross-arm