The MCF-7 arm, and the clause that saved it: SUV39H1 at the exon-11 anchor reaches the 96.3rd percentile of 56,171 same-cell-type CTCF anchors in one replicate - but all six non-anchor KMT2A intervals are elevated too (medians 73-87), so the pre-registered FIRES clause does not fire, the verdict is NEITHER, and four replicates of one antibody in one cell line span 51.8 to 96.3 at the same 147 bp
also called Q114 MCF-7 arm · the whole locus moved · the anchor is not picked out in MCF-7
In breast cancer cells the silencing enzyme does look present at the spot this project cares about - but it looks just as present everywhere else across the whole gene, so nothing about that spot is special. And four repeats of the same experiment in the same cells disagree wildly with each other, which is the most useful thing this arm produced.
The same thing, for a scientist
**The pre-registered verdict is UNINTERPRETABLE, and it was fixed before the data were read.** The MCF-7 arm's C4 instrument control failed - the ENCODE MCF-7 CTCF signal track puts the exon-11 window at the **72.95th** percentile of its own anchors against a required 90th - and §7 of the pre-registration says an arm failing C4 is uninterpretable [E2]. That is not being upgraded because the numbers came out interesting. **Descriptively, and this is where the design earned its keep.** Mean SUV39H1 signal over each interval, as a percentile of 56,171 **same-cell-type** CTCF summit windows [E1]: | interval | −E2 (857) | −E2 (858) | +E2 (859) | +E2 (860) | median | |---|---:|---:|---:|---:|---:| | **exon11 (147 bp)** | **90.81** | 69.05 | **96.29** | 51.78 | **79.93** | | ctcf_element | 86.08 | 75.45 | 91.38 | 34.46 | 80.77 | | ctcf_summit_1kb | 86.85 | 74.78 | 91.30 | 36.01 | 80.82 | | dcal_cut_1kb | 56.06 | 44.88 | 73.17 | 56.61 | 56.34 | | bcr_core | 82.71 | 81.25 | 82.66 | 72.76 | 81.96 | | kmt2a_tss_2kb | 78.01 | 90.74 | 89.69 | 69.73 | 83.85 | | kmt2a_gene (90 kb) | 85.51 | 85.31 | 86.79 | 81.65 | 85.41 | **FIRES requires the anchor above the 75th percentile in >= 60% of tracks AND the six non-anchor intervals' median below 50.** The first half is 2 of 4 (50%, short of 60%); **the second half fails outright** - the non-anchor median is 84.11, 78.35, 88.24 and 63.17 in the four tracks. **FAILS requires the anchor's median percentile inside [25, 75]; it is 79.93.** So the descriptive verdict is **NEITHER**, in those words. **And the reason is the most informative part: the entire 90 kb gene body is at the 81.7th-86.8th percentile of CTCF anchors in every one of the four tracks.** The anchor is not picked out; a 90 kb region is elevated, uniformly, in all four replicates and both conditions. A whole-locus elevation is what copy number, accessibility, mappability or an input-free normalisation produces - not what a 147 bp binding event produces. **CCLE calls *KMT2A* copy-number neutral in MCF-7** [E3], which weakens but does not eliminate the copy-number reading; the deposit has no input track, so the cleanest test is not available from it. **The replicate spread is a finding in its own right, and it is the one this graph should carry forward.** Four ChIP-seq samples, **one antibody, one cell line, one lab, one submission**, give the identical 147 bp window percentiles of **51.78, 69.05, 90.81 and 96.29** - a **44.5-percentile spread**, wider than the difference between any two of this project's published locus results. The two conditions do not separate: −E2 gives 90.81 and 69.05, +E2 gives 96.29 and 51.78.
What this says about single-track occupancy numbers everywhere in this corpus
This project has repeatedly drawn conclusions from one track per protein per cell type - Q35, Q37, Q110, Q112 and the DIPG arm of this very beat all do it, because that is usually all the archives hold.
Here four technical/biological replicates of the same antibody in the same line span 51.8 to 96.3 at the same 147 bp window. If that spread is typical, then a single-track percentile at a 147 bp element carries an uncertainty of tens of percentile points and no entry in this corpus states one. That does not invalidate the earlier results - a negative at the low end of such a spread is still a negative - but it does mean a single positive near the 75th-90th percentile is not distinguishable from noise, and several of this graph's borderline reads sit exactly there.
The honest generalisation is narrow: four samples, one deposit, no input, no clone stated, and the 'dsgf' token distinguishing two of them is unexplained. It is one deposit's spread, not a measured error model. But it is the only replicate estimate this corpus has, and it should be quoted as a floor on the uncertainty rather than ignored.
The clause that did the work, and why it should be standard
The FIRES rule was written with two halves: the anchor above the 75th and the six non-anchor intervals below the 50th. The second half is what stopped a 96.29th-percentile number becoming a positive.
Without it, this arm would have reported "SUV39H1 occupies the KMT2A exon-11 CTCF anchor at the 96th percentile of same-cell-type CTCF anchors" - a true sentence, a same-cell-type class-matched null, and a completely misleading result, because the promoter, the BCR and the whole 90 kb gene body are all at the 80th-87th in the same file.
Any future beat scoring an element in this graph should carry a non-target interval set from the same locus, not only a matched null from elsewhere in the genome. The two catch different failures: the matched null catches 'this class of site is just like this', and the non-target intervals catch 'this whole region is just like this'.
Controls in this arm
- C1 (dynamic range) PASSES cleanly and non-degenerately, unlike the DIPG arm: the random-tile median is 0.40-0.42 rather than 0, and the top 0.1% of anchor windows is 10.4x, 10.9x, 10.0x, 10.1x over it in the four tracks.
- C2 (chr19q13 KRAB-ZNF, 1 Mb) is 1.20-1.23x over the tile mean and ZNF274 3' 10 kb is 0.93-1.15x - i.e. no enrichment at the canonical heterochromatin address in the SUV39H1 tracks here either. The discriminating comparison is in this same deposit: the two MCF-7 H3K9me3 tracks pass the identical control at 3.05x and 4.15x (1 Mb) and 3.79x and 5.41x (ZNF274 3'). So the control works - for the mark. It fails for SUV39H1 in both deposits, which is the expected result rather than a defect: the KRAB-ZNF cluster is TRIM28/SETDB1's address, and SUV39H1's canonical address is pericentromeric satellite, which a unique-alignment pipeline cannot place. That also means the GSE225028 H3K9me3 failure of the same control is a statement about those CUT&RUN libraries, not about the control.
- C3 (chr2 gene desert) is 1.20-1.24x over the tile mean - i.e. above it, and about the same as the KRAB-ZNF cluster. In a track where a gene desert and a heterochromatin domain read alike, a 1.1-1.2x elevation over a gene body is not evidence of anything.
- The NaN probe behaved differently here and the per-file check caught it: in GSM4133860 the
chr11:1-1000window has zero raw records andpybigtoolsreturned NaN by default and0.0withfillna=0, whereas in the GEO-supplied DIPG bigWigs all three modes agreed because those files carry explicit zero records. The same code on two deposits needed two different fill behaviours, and only a per-file probe would have shown it.
| script | analyses/q114-suv39h1-at-the-kmt2a-locus/run_mcf7.py |
|---|---|
| status | run |
| run date | 2026-09-22 |
| run by | Q114-suv39h1-at-the-locus |
| result summary | Pre-registered verdict UNINTERPRETABLE (C4 failed at 72.95 against 90). Descriptive verdict NEITHER: exon-11 percentiles 90.81/69.05/96.29/51.78 (median 79.93) but all six non-anchor intervals elevated too (non-anchor medians 84.11/78.35/88.24/63.17) and the whole 90 kb gene body at the 81.7-86.8th percentile in all four tracks. |
| verdict question B class matched | UNINTERPRETABLE by C4; descriptively NEITHER - not FIRES (2 of 4 tracks above the 75th, and the six non-anchor intervals are not below the 50th) and not FAILS (median 79.93 is outside [25,75]) |
| verdict question A occupancy | NOT RESOLVED by this arm. exon11 reaches 2.46x and 2.57x the random-tile mean in two tracks and 1.03x and 1.71x in the other two, but the 90 kb gene body is at 1.17-1.19x in all four and the chr2 gene desert is at 1.20-1.24x, so the locus is not separable from the track's own background elevation. |
| exon11 replicate spread percentile points | 44.51 |
| n class matched anchors | 56171 |
| genome build | GRCh38, established from each bedGraph's OWN maximum coordinate per chromosome in both directions (hg19 excluded); the deposit also states Genome_build: hg38 per sample, which a corpus entry wrongly recorded as NOT STATED |
| nan probe | differs between deposits: GSM4133860 chr11:1-1000 has zero raw records and returns NaN by default, 0.0 with fillna=0; the GEO-supplied DIPG bigWigs carry explicit zero records and all three modes agree |
| cell type | MCF-7, ER-positive luminal breast adenocarcinoma |
| cell state | hormone-deprived (5% charcoal-stripped FBS >=3 days) and 100 nM 17-beta-estradiol 40 min |
| lineage | breast epithelium - NOT haematopoietic, NOT KMT2A-rearranged |
| shown for fusion | NO - nobody has looked in any KMT2A fusion system, and this is a breast line |
| copy number check | CCLE ccle_broad_2019_cna, MCF7_BREAST: KMT2A 0 (neutral), SUV39H1 0, ATM -2 |
inputs
[ "GSM4133857", "GSM4133858", "GSM4133859", "GSM4133860", "ENCFF942TCG", "ENCFF516HJR" ]
exon11 percentile vs same cell ctcf anchors — the numbers behind the chart
{
"GSM4133857_minusE2": 90.81,
"GSM4133858_minusE2": 69.05,
"GSM4133859_plusE2": 96.29,
"GSM4133860_plusE2": 51.78
}exon11 fold over random tile mean — the numbers behind the chart
{
"GSM4133857": 2.456,
"GSM4133858": 1.705,
"GSM4133859": 2.572,
"GSM4133860": 1.03
}kmt2a gene percentile all four
[ 85.51, 85.31, 86.79, 81.65 ]
non anchor median percentile per track — the numbers behind the chart
{
"GSM4133857": 84.11,
"GSM4133858": 78.35,
"GSM4133859": 88.24,
"GSM4133860": 63.17
}c4
{
"file": "ENCFF516HJR",
"exon11_mean": 62.3949,
"percentile": 72.954,
"threshold": 90,
"passes": false
}c1 ratio over random tile median — the numbers behind the chart
{
"GSM4133857": 10.38,
"GSM4133858": 10.91,
"GSM4133859": 10.01,
"GSM4133860": 10.08
}c2 znf chr19q13 1Mb over tile mean
[ 1.21, 1.23, 1.2, 1.21 ]
c2 znf274 3prime over tile mean
[ 0.93, 1.15, 1.04, 1.11 ]
c3 chr2 desert over tile mean
[ 1.23, 1.24, 1.2, 1.23 ]
bedgraph lines
{
"GSM4133857": 49840662,
"GSM4133860": 44303344
}- The tempting headline here - SUV39H1 at the 96th percentile of same-cell CTCF anchors at the KMT2A anchor - is true, class-matched, and wrong, because the promoter, the breakpoint cluster and the whole gene body are at the 80th-87th in the same file. Anyone quoting the anchor number without the gene-body number is making the exact error this beat's pre-registration was written to prevent.
- A 44.5-percentile spread across four replicates means that a therapeutic argument resting on a single occupancy track - in this graph, most of them - is resting on a number whose uncertainty nobody has stated.
- Is the locus-wide elevation copy number, accessibility, or normalisation? The ENCODE MCF-7 input/control track read over the same seven intervals would settle it in one run and was not done. CCLE calls KMT2A copy-neutral in MCF-7, which weakens the copy-number reading but cannot exclude a sub-threshold segmental gain.
- What does 'dsgf' mean in GSM4133857 and GSM4133859? Those two are the high tracks (90.81 and 96.29) and the two without the token are the low ones (69.05 and 51.78). If dsgf denotes a crosslinker or a protocol variant, the 44-percentile spread is a protocol effect rather than replicate noise - and that would be a much more useful fact. The deposit does not say and the corresponding author has not been asked.
- Does the 44.5-percentile replicate spread generalise? It is the only replicate estimate of single-track percentile uncertainty this corpus has. Two more deposits with >=3 replicates of one factor in one cell type would say whether tens of percentile points is typical, and that would recalibrate every borderline occupancy read in this graph.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
GSE139196's four SUV39H1 ChIP-seq bedGraphs in MCF-7 (two minus estradiol, two at 40 min estradiol) were converted to bigWig and read over the seven stored KMT2A intervals against 56,171 same-cell-type CTCF summit windows and 200,000 random tiles; the build was established from each file's own maximum coordinate per chromosome as GRCh38, with hg19 excluded.
GSE139196: SUV39H1 (KMT1A) ChIP-seq in MCF-7 with and without estradiol, GSM4133857-60, HOMER bedGraph on hg38 · Yang F, et al. · NCBI GEO 2019What this evidence cannot show
The deposit names no antibody clone, contains no input or IgG track identified by this beat, and the four samples differ in depth (44.3-49.8 million bedGraph intervals). The two 'dsgf' samples (GSM4133857, GSM4133859) and the two plain ones are described identically in the deposit apart from that token, whose meaning is not given.
Evidence that argues against this (1)
- Without an input track the fold and percentile statistics are against a genomic null rather than against matched chromatin, so any locus-wide elevation could be accessibility, copy number or sonication bias rather than occupancy. The four-track replicate spread reported here is the internal check on that and it is large. source
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
The pre-registered C4 instrument control FAILED for MCF-7: the ENCODE MCF-7 CTCF signal track places the exon-11 window at the 72.95th percentile of its own 56,171 summit windows, against a required 90th, so the whole arm is UNINTERPRETABLE by the rule written before the data were read.
ENCSR000AHD / ENCFF516HJR (MCF-7 CTCF signal p-value, GRCh38, preferred_default) read against ENCFF942TCG summit windows · Richard Myers, HudsonAlpha; ENCODE Project Consortium · ENCODE Portal 2026What this evidence cannot show
The threshold was not moved after the fact. The failure is a property of the element, not of MCF-7: Q112 measured the same statistic in six other ENCODE cell types and got a median of 70.61 with 0 of 6 above the 90th.
Evidence that argues against this (1)
- A control that fails in seven of seven cell types is arguably the wrong control rather than a real blindness finding. That reading is available and is recorded in genome-q114-the-kmt2a-exon-11-ctcf-summit-lands-within-12-bp-in-two-more-cell-types - but it cannot be adopted retroactively for this arm without turning a pre-registration into a suggestion. source
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database
CCLE's discrete copy-number call for KMT2A in MCF-7 is 0 (neutral), so a simple amplification does not explain the locus-wide elevation seen in these tracks - while ATM at 11q22.3 in the same line is called -2, a deep deletion.
Cancer Cell Line Encyclopedia (Broad, 2019), ccle_broad_2019_cna, sample MCF7_BREAST: KMT2A (Entrez 4297) = 0, SUV39H1 (6839) = 0, ATM (472) = -2 · Cancer Cell Line Encyclopedia (Broad Institute), served by cBioPortal - study ccle_broad_2019 · cBioPortal for Cancer Genomics 2019What this evidence cannot show
Fetched live from the cBioPortal REST API on 2026-09-22 (molecular-profiles/ccle_broad_2019_cna/molecular-data/fetch, sample MCF7_BREAST). The originating CCLE publication was NOT fetched and is deliberately not named here. A GISTIC-style discrete call is thresholded; a gain of the size that would produce a 1.1-1.2x signal elevation is exactly the size a discrete caller reports as 0. This evidence weakens the copy-number explanation, it does not eliminate it, and the segment-level log2 ratio was not retrieved.
Evidence that argues against this (1)
- MCF-7 is heavily aneuploid and structurally rearranged, and a discrete per-gene call cannot describe a local segment. A proper test would read the ENCODE MCF-7 input/control track over the same seven intervals, which this beat did not do. source
first reading of the four SUV39H1 samples that entry describes
SUV39H1 occupancy at seven KMT2A intervals in MCF-7, four replicates
- related to →KMT2A / MLL1 — the gene that breaks
the whole 90 kb gene body is elevated, not the anchor
the MCF-7 CTCF peak set supplies both the null and the failed C4 control
the second arm of the same pre-registration, in a different cell type and assay
the deposit census that identifies these four samples and their conditions
the reproducibility measurement behind that arm's 44.5-percentile spread
the writer's locus-wide elevation is not producing mark: over the same 90 kb the mark is at the 10th-16th percentile
the same locus-wide elevation that entry found for SUV39H1 is shared by every protein antigen in the deposit - seven antigens, 25 tracks, six-non-anchor medians 49.79 to 72.37
that entry could not tell whether SUV39H1's locus-wide elevation was the protein or the locus; eight antigens in the same deposit say it is the locus
the signature that entry found for SUV39H1 replicates in KDM4B (79.69) and KDM4C (76.62) against a 27-track panel median of 61.39