A per-sample parameter nobody reads: MACS's fragment-size estimate varies 822-1112 bp across ten ChIPs from one sequencing run, and within the DOT1L arm peak count correlates +0.7 with fragment size and -0.8 with library depth - so the obvious confound is not the operating one, and the non-obvious one is
also called the fragment-size confound · GSE149183 instrument note
When comparing these experiments people usually worry about how deeply each was sequenced. In this dataset that worry points the wrong way - the number of hits tracks a different, rarely-checked setting that varies between samples for reasons nobody reports.
The same thing, for a scientist
A methodological finding from GSE149183 that generalises well beyond it. **The numbers.** Ten MEN1/DOT1L ChIPs, one sequencing run (200109), one shared input, identical MACS2 2.1.1 invocation (`-f BAMPE -g hs -q 0.01 --slocal 1500`). The only per-sample parameter that varies is the fragment size MACS derives from the read pairs [E1]: | sample | peaks | depth (fragments) | MACS fragment size | |---|---:|---:|---:| | DOT1L DMSO | 1,139 | 17,750,246 | 822 | | DOT1L compound 10 | 1,524 | 16,402,117 | 859 | | DOT1L VTP-50469 | 1,695 | 11,225,637 | 969 | | DOT1L EPZ5676 | 1,214 | 13,064,535 | 1,064 | | DOT1L compound 11 | 1,782 | 11,522,001 | 1,112 | | MEN1 DMSO | 2,362 | 13,980,511 | 851 | | MEN1 EPZ5676 | 1,626 | 15,172,064 | 841 | | MEN1 compound 10 | 1,779 | 13,577,277 | 870 | | MEN1 VTP-50469 | 1,101 | 13,844,955 | 906 | | MEN1 compound 11 | 1,273 | 12,441,785 | 926 | **Three things follow.** **1. The depth confound runs backwards here.** Across all ten, peak count versus depth is Spearman **-0.30**; within the DOT1L arm it is **-0.80**. The deepest DOT1L library (17.75 M) calls the *fewest* peaks (1,139). Anyone controlling only for depth would conclude the comparison is safe, and would be controlling for the wrong thing [E1]. **2. Fragment size tracks it instead, in the DOT1L arm.** Spearman **+0.70** against fragment size within that arm. Pileup scales with fragment length, so a 1.35-fold spread in estimated fragment size is a direct multiplier on the statistic the peak call is thresholded on [E1]. **3. It is arm-specific, and that is why one arm's result is trustworthy and the other's is not.** The MEN1 arm's fragment sizes span only **841-926 bp (1.10-fold)** and its depths only 12.44-15.17 M (1.22-fold); the DOT1L arm spans 822-1,112 bp (1.35-fold) and 11.23-17.75 M (1.58-fold). **The MEN1 arm is the tight one on both axes**, which is why this beat's headline claim rests on it and the DOT1L arm is reported as directional support [E1]. **The general rule.** BAMPE mode estimates fragment size per sample from the observed insert-size distribution. Sonication efficiency, chromatin state, library prep and drug treatment all move it, and **it is never reported in a methods section**. It sits in the MACS output header of every deposited `.xls` peak file, one grep away, and it should be pulled and tabulated beside depth whenever peak counts are compared across samples.
What this cost, concretely, in this beat
It is the reason hyp-q136-catalytically-inhibited-dot1l-is-retained-on-chromatin-at-fusion-target-genes is filed as a hypothesis rather than a finding. DOT1L's fold-enrichment at target loci rises under all three DOT1L inhibitors - PBX3 from 4.85 to 13.92 - and the rank order of that rise is the rank order of fragment size. The observation may be real and no statistic computable from the deposited files can say.
Why a drug-induced fragment-size shift is not purely a nuisance
A compound that alters chromatin compaction changes how the chromatin shears, so the fragment-size spread may partly BE the biology. That makes it worse, not better, as a confound: it is then correlated with treatment by construction and cannot be adjusted away. The only clean fix is to report the insert-size distribution per sample and, where peak counts are being compared, to re-call with a fixed extension size as a sensitivity check.
| script | analyses/menin-inhibitor-against-dot1l-inhibitors-in-one-deposit/run.py |
|---|---|
| status | run |
| geo | GSE149183 |
| where to find it | the '# fragment size is determined as N bps' line in any deposited MACS2 .xls/peaks file header |
| rows dropped by any filter | 0 |
fragment size bp — the numbers behind the chart
{
"DOT1L_DMSO": 822,
"DOT1L_comp10": 859,
"DOT1L_VTP50469": 969,
"DOT1L_EPZ5676": 1064,
"DOT1L_comp11": 1112,
"MEN1_EPZ5676": 841,
"MEN1_DMSO": 851,
"MEN1_comp10": 870,
"MEN1_VTP50469": 906,
"MEN1_comp11": 926
}depth fragments — the numbers behind the chart
{
"DOT1L_DMSO": 17750246,
"DOT1L_VTP50469": 11225637,
"DOT1L_EPZ5676": 13064535,
"DOT1L_comp10": 16402117,
"DOT1L_comp11": 11522001,
"MEN1_DMSO": 13980511,
"MEN1_VTP50469": 13844955,
"MEN1_EPZ5676": 15172064,
"MEN1_comp10": 13577277,
"MEN1_comp11": 12441785
}n peaks — the numbers behind the chart
{
"DOT1L_DMSO": 1139,
"DOT1L_VTP50469": 1695,
"DOT1L_EPZ5676": 1214,
"DOT1L_comp10": 1524,
"DOT1L_comp11": 1782,
"MEN1_DMSO": 2362,
"MEN1_VTP50469": 1101,
"MEN1_EPZ5676": 1626,
"MEN1_comp10": 1779,
"MEN1_comp11": 1273
}spearman npeaks vs depth — the numbers behind the chart
{
"all10": -0.297,
"MEN1": 0.3,
"DOT1L": -0.8
}spearman npeaks vs fragsize — the numbers behind the chart
{
"all10": 0.103,
"MEN1": -0.6,
"DOT1L": 0.7
}arm spread fragsize fold
{
"MEN1": 1.1,
"DOT1L": 1.35
}arm spread depth fold
{
"MEN1": 1.22,
"DOT1L": 1.58
}- A pharmacodynamic assay read out as peak count would report drug effects that are partly sonication behaviour. Any occupancy-based endpoint has to be locus-specified and fragment-size controlled, or it will move for reasons that have nothing to do with the drug.
- How many cross-sample ChIP comparisons in this corpus controlled for depth and not for fragment size? The parameter is in the header of every deposited MACS file and this is the first time this project has read it.
- Does the fragment-size spread reflect drug-induced chromatin compaction? If so it is a readout in its own right - a cheap, already-deposited measure of global chromatin state under menin and DOT1L inhibition - and nobody has looked at it that way.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Across the ten GSE149183 factor-ChIP MACS2 headers, fragment size spans 822-1112 bp and depth 11.23-17.75 M fragments; peak count correlates -0.30 with depth across all ten, -0.80 with depth and +0.70 with fragment size within the DOT1L arm, and +0.30 with depth and -0.60 with fragment size within the MEN1 arm.
analyses/menin-inhibitor-against-dot1l-inhibitors-in-one-deposit (PREREGISTRATION.md, run.py, results.json) - this repository, re-analysis of the ten deposited GSE149183 MACS2 peak tables · Oncofusion knowledge graph 2026What this evidence cannot show
Spearman over n = 10 and n = 5; these coefficients are descriptive and carry no meaningful p-value at that n. The claim is about rank agreement and about the SIGN of the depth correlation being negative, which is the part that overturns the default assumption.
Evidence that argues against this (1)
- Within the MEN1 arm the signs reverse (+0.30 depth, -0.60 fragment size), so neither variable is a universal driver. The honest statement is that peak count in this deposit is not a clean function of either, which is itself the reason peak count is not used as a readout anywhere in this beat. Oncofusion knowledge graph 2026
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
All ten peak files record the same MACS2 invocation and the same single control BAM, so the per-sample fragment size is the only varying analysis parameter.
NCBI GEO GSE149183 - MEN1, DOT1L and H3K79me2 ChIP-seq in MOLM13 and RS4;11 under DMSO, VTP-50469, EPZ5676, compound 10 and compound 11; 31 samples, Homo sapiens; SubSeries of GSE149185, BioProject PRJNA627612, SRA SRP258055 · NCBI Gene Expression Omnibus 2020What this evidence cannot show
Read from the `# Command line:` and `# control file =` lines of all ten deposited headers. The upstream alignment and duplicate-marking were also uniform per the series processing description.
Evidence that argues against this (1)
- Uniform parameters do not make uniform libraries: antibody efficiency, chromatin input amount and PCR cycle number (stated as 10-14 cycles) vary and are not recorded per sample. NCBI Gene Expression Omnibus 2020
the fragment-size spread whose rank order matches the rank order of the apparent gain
the MEN1 arm is the tight one on both depth (1.22x) and fragment size (1.10x), which is why the headline claim rests on it
the DOT1L arm spans 1.58x in depth and 1.35x in fragment size, which is why it is reported as directional support only
both entries are about the same failure: a peak-level statistic that moves for reasons other than occupancy