Oncofusion

start / graph / ana-q37-the-kmt2a-exon-11-anchor-carries-a-cbx3-and-a-brca1-peak-that-idr-removes

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q37-archive-sweep

The KMT2A exon-11 CTCF anchor carries a CBX3 peak and a BRCA1 peak in three of the very ENCODE experiments Q35 called empty - the peaks are in ENCODE's own pre-IDR files, present in both replicates, and deleted by the IDR threshold at global IDR 0.227 and 0.059

also called the IDR threshold deleted the KMT2A HP1-gamma peak · CBX3 and BRCA1 at the KMT2A exon-11 CTCF element · Q35's thirteen-of-thirteen negative re-read without IDR

Last cycle this graph concluded that two DNA-repair proteins are simply not present on the KMT2A gene, because a strict peak-calling filter found nothing in thirteen public datasets. Re-reading the same datasets with the filter turned down shows the proteins are there, sitting on one small 147-base-pair element, and one of them missed the cut-off by nine thousandths.

The same thing, for a scientist

**Q35 (cycle 63) read 13 released ENCODE BRCA1 / CBX3 / CBX1 ChIP-seq experiments at seven *KMT2A* intervals and reported ZERO called IDR peaks in 13 of 13.** That measurement is correct. Its conclusion - *no occupancy above background at the KMT2A locus* - is not. **Three of those thirteen experiments call a peak essentially ON the exon-11 CTCF element when the same reads are processed without IDR** [E1]. ChIP-Atlas re-processes SRA with one uniform MACS2 pipeline at q < 1e-5: | ChIP-Atlas SRX | = ENCODE | target | cell | peak (GRCh38) | fold | -log10 q | summit | |---|---|---|---|---|---:|---:|---:| | SRX190188 | **ENCSR000BRT** | CBX3 (HP1-gamma) | **K-562** | chr11:118,488,610-118,488,767 | 8.13 | 19.99 | 118,488,695 | | SRX190214 | **ENCSR000BUH** | CBX3 | HCT 116 | chr11:118,488,596-118,488,758 | 12.50 | 33.30 | 118,488,668 | | SRX150631 | **ENCSR000EDB** | BRCA1 | HeLa | chr11:118,488,641-118,488,772 | 8.19 | 14.25 | 118,488,703 | The exon-11 interval is chr11:118,488,614-118,488,760 (147 bp). The CTCF summit this graph has carried for several cycles is 118,488,694. **The CBX3 K-562 summit is 118,488,695 - one base away.** **And the peaks are inside ENCODE's own files, so this is not a claim about a foreign pipeline** [E2]. For **ENCSR000BRT** the optimal-IDR file Q35 read (ENCFF951BQB, 26,789 peaks) has nothing there; the pre-threshold `IDR ranked peaks` file ENCFF271ZAZ has chr11:118,488,467-118,488,897 at signalValue 22.39, **rep1 14.01 and rep2 8.49 - present in both replicates** - with **globalIDR -log10 = 0.6448, i.e. IDR 0.227**, against the 0.05 cut. For **ENCSR000EDB** the conservative-IDR file Q35 read (ENCFF668JZI, 9,452 peaks) has nothing; ENCFF457RFT has chr11:118,488,543-118,488,843 at signalValue 15.66, rep1 9.21 and rep2 7.13, **globalIDR -log10 = 1.2302, i.e. IDR 0.0588 - it misses the threshold by 0.009.** **The instrument was controlled both ways** [E3]. Positive: a ChIP-Atlas CTCF track calls the element in GM12878 (FE 22.5, summit 118,488,695) and HCT116. The decisive control is the **auxin-degron pair**: SRX11361143 (CTCF auxin-, 29,841 peaks) calls it and SRX11361144 (CTCF auxin+, CTCF acutely degraded) does not. So the interval is not a hyper-ChIPable artefact of this code path; the signal is CTCF-dependent. Negative: a 7,466 bp gene desert on chr2 returns zero peaks in **all 180 tracks**. The element is not in the ENCODE hg38 blacklist v2. **And the class-matched null, which is the part that keeps Q35 half right** [E4]. Against CTCF anchors of the same cell type - the null `concept-a-genome-wide-null-flatters-any-interval-that-belongs-to-an-enriched-class` demands - **30.2% of 44,722 K562 CTCF anchors carry a CBX3 peak** and 66.9% of HCT116 anchors do. CBX3 at this anchor is typical of its class. **BRCA1 is different: only 10.3% of 41,050 HeLa-S3 CTCF anchors carry a BRCA1 peak, and the KMT2A anchor is one of them.** **So the correct statement is narrower than either the old one or the obvious new one.** The proteins are demonstrably ON this element. For CBX3 that is unremarkable given where it sits generally; for BRCA1 it puts the *KMT2A* anchor in the minority tenth of CTCF anchors in that cell. What does **not** survive is *"no BRCA1 or CBX3 occupancy exists above background at the KMT2A locus"*.

Facts and figures
scriptanalyses/q37-archive-sweep-at-the-kmt2a-locus/run.py
designanalyses/q37-archive-sweep-at-the-kmt2a-locus/design.py (pre-registered, written before any peak file was opened)
statusrun
run date2026-09-22
run byQ37-archive-sweep
result summary180 ChIP-Atlas hg38 peak sets over 12 antigens, plus ENCODE's own pre-IDR files for three experiments. Three tracks call a peak on the KMT2A exon-11 CTCF element; the two whose ENCODE pipeline agrees fail IDR at 0.227 and 0.0588.
n tracks180
n tracks failed0
relationship to existing entriesThis is a direct re-read of the same experiments as `ana-q35-brca1-cbx3-and-cbx1-are-not-at-the-kmt2a-breakpoint-cluster-in-any-encode-cell`, at peak level only, with the IDR step removed. It does not re-measure the signal-level folds Q35 reported and does not contradict them.
inputs
[
  "ChIP-Atlas hg38 eachData/bed05 MACS2 q<1e-5 peak sets, 180 SRX, fetched 2026-09-22",
  "ENCODE files ENCFF951BQB, ENCFF275XNU, ENCFF469CQE, ENCFF271ZAZ (ENCSR000BRT); ENCFF885KVX, ENCFF991DCE (ENCSR000BUH); ENCFF668JZI, ENCFF457RFT (ENCSR000EDB)",
  "ENCODE K562 CTCF optimal IDR ENCFF843VHC (44,722 anchors) and HeLa-S3 CTCF optimal IDR ENCFF154APE (41,050 anchors)",
  "ENCODE hg38 blacklist v2, 636 regions"
]
intervals GRCh38 1based inclusive
{
  "kmt2a_gene": "chr11:118,436,456-118,526,832",
  "bcr_core": "chr11:118,482,422-118,489,887",
  "ctcf_element": "chr11:118,488,159-118,489,206",
  "exon11": "chr11:118,488,614-118,488,760",
  "ctcf_summit_1kb": "chr11:118,488,194-118,489,194",
  "kmt2a_tss_2kb": "chr11:118,435,456-118,437,456",
  "dcal_cut_1kb": "chr11:118,484,362-118,485,361"
}
the three positive tracks
{
  "SRX190188 = ENCSR000BRT = GSM1010732": "CBX3, K-562, 52,281 peaks genome-wide, peak chr11:118,488,610-118,488,767, FE 8.134, -log10q 19.991, summit 118,488,695, 81st percentile of its own track by fold enrichment",
  "SRX190214 = ENCSR000BUH = GSM1010758": "CBX3, HCT 116, 67,623 peaks, peak chr11:118,488,596-118,488,758, FE 12.501, -log10q 33.301, summit 118,488,668, 91st percentile",
  "SRX150631 = ENCSR000EDB = GSM935552": "BRCA1, HeLa, 16,300 peaks, peak chr11:118,488,641-118,488,772, FE 8.186, -log10q 14.254, summit 118,488,703, 65th percentile"
}
inside encodes own pipeline
{
  "ENCSR000BRT": "ENCFF271ZAZ IDR-ranked: chr11:118,488,467-118,488,897, signalValue 22.390, localIDR -log10 0.2821, globalIDR -log10 0.6448 (IDR 0.2266), rep1 signal 14.007, rep2 signal 8.488. ENCFF469CQE relaxed MACS2 (300,000 peaks): chr11:118,488,488-118,488,918, signalValue 15.840.",
  "ENCSR000EDB": "ENCFF457RFT IDR-ranked: chr11:118,488,543-118,488,843, signalValue 15.662, localIDR -log10 0.7402, globalIDR -log10 1.2302 (IDR 0.0588), rep1 9.214, rep2 7.128.",
  "ENCSR000BUH": "ENCFF991DCE IDR-ranked (65,155 peaks) has NO peak at the element. ENCODE's own pipeline does not see it in HCT116 even pre-threshold; only ChIP-Atlas's independent alignment and calling does. Recorded as a pipeline difference, not as corroboration."
}
controls
{
  "C1_positive": "CTCF SRX038504 GM12878 30,154 peaks -> chr11:118,488,547-118,488,848 FE 22.512 -log10q 73.371 summit 118,488,695. CTCF SRX11361143 HCT116 auxin- 29,841 peaks -> chr11:118,488,378-118,488,960 FE 12.458 summit 118,488,697.",
  "C1_the_degron_control": "SRX11361144, the SAME HCT116 line with auxin added and CTCF acutely degraded, 9,347 peaks, calls NOTHING at the element. The signal is CTCF-dependent, which is the strongest available argument against a hyper-ChIPable-region explanation.",
  "C1_two_failures_reported_not_dropped": "ERX1149057 and ERX1149058, BLUEPRINT CTCF in K-562 with only 4,134 and 7,570 peaks, MISS the element. A positive control can fail on depth and these two did.",
  "C2_negative": "chr2:35,000,000-35,007,465, a 7,466 bp gene desert of the same width as bcr_core: zero peaks in 180 of 180 tracks.",
  "blacklist": "the element is not among the 636 regions of the ENCODE hg38 blacklist v2 (Boyle-Lab, checked 2026-09-22)"
}
class matched null
{
  "CBX3_K562_vs_K562_CTCF_anchors": "13,524 of 44,722 (30.2%) ENCODE K562 CTCF optimal-IDR anchors carry a called CBX3 peak in SRX190188",
  "CBX3_HCT116_vs_HCT116_CTCF_anchors": "19,960 of 29,841 (66.9%)",
  "BRCA1_HeLa_vs_HeLaS3_CTCF_anchors": "4,222 of 41,050 (10.3%) - the KMT2A anchor is in the minority tenth",
  "the_anchor_itself_in_HeLa": "ENCFF154APE contains chr11:118,488,634-118,488,703, score 1000, signalValue 69.261 - the CTCF anchor is robustly present in HeLa as well as in the blood lines"
}
what this does and does not refute
{
  "SURVIVES": "Q35's literal peak-level statement (zero called IDR peaks in 13 of 13); Q35's class-matched finding that the element is not enriched above its own class for CBX3; Q35's signal-level fold values, which were not re-measured here.",
  "DOES_NOT_SURVIVE": "the conclusion that no BRCA1 or CBX3 occupancy exists above background at the KMT2A locus, and the downstream reclassification of `absence-no-resection-factor-localised-at-the-kmt2a-locus` to `never-measured` on the ground that the public data was 'spent'."
}
How this could bite us back
  • If HP1-gamma and BRCA1 genuinely sit on the exon-11 CTCF anchor, then an H3K9-axis inhibitor given to protect against KMT2A rearrangement would be removing a repair licence at the very element it is meant to protect - the risk already recorded in `risk-an-h3k9-axis-inhibitor-given-with-a-top2-poison-removes-a-repair-licence`, now with an occupancy measurement behind it rather than an inference.
  • The opposite error is cheaper and more likely: converting a peak at IDR 0.227 into a mechanism. An intervention designed around this element on the strength of two sub-threshold peaks in non-leukaemic lines would be built on the weakest evidence in this entry, not the strongest.
Still open
  1. What is the fold enrichment of the exon-11 element in these three tracks at SIGNAL level, against the housekeeping-promoter and CTCF-anchor nulls Q35 built? This entry is peak-level only and the signal-level question is the one that decides whether the occupancy is worth a wet experiment.
  2. Does the CBX3 K-562 peak survive in a KMT2A-rearranged blood line? K-562 is CML blast crisis and carries no KMT2A rearrangement; SEM, RS4;11 and MV4-11 have no CBX3 ChIP-seq in any archive.
  3. How many other bounded absences in this corpus rest on an IDR-thresholded peak set? A sweep of every entry citing 'zero called peaks' against the corresponding pre-IDR file is a one-afternoon audit and has never been run.
Every source behind this entry (4)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    In ChIP-Atlas's uniform hg38 MACS2 re-processing (q < 1e-5), three human ChIP-seq experiments call a peak on the KMT2A exon-11 CTCF element: CBX3 in K-562 (SRX190188, chr11:118,488,610-118,488,767, fold 8.13, -log10q 19.99, summit 118,488,695), CBX3 in HCT 116 (SRX190214, fold 12.50, -log10q 33.30) and BRCA1 in HeLa (SRX150631, fold 8.19, -log10q 14.25). These SRX correspond to ENCODE experiments ENCSR000BRT, ENCSR000BUH and ENCSR000EDB via GEO accessions GSM1010732, GSM1010758 and GSM935552, which are three of the thirteen experiments Q35 read.

    PMID 42051034DOI
    What this evidence cannot show

    Peak-level only: a sub-threshold enrichment is invisible to this design and no signal value was re-read. ChIP-Atlas's alignment and peak calling differ from ENCODE's in more than the IDR step (read trimming, duplicate handling, single-replicate calling), so the two pipelines are not identical apart from thresholding, and the HCT116 case is called by ChIP-Atlas alone. The three positive tracks are among the deepest peak sets in the sweep (52k, 68k, 16k peaks), and peak count tracks sensitivity, so the finding is partly a sensitivity finding.

    Evidence that argues against this (1)
    • CBX3 occupies 30.2% of K562 CTCF anchors and 66.9% of HCT116 CTCF anchors in these same tracks, so a CBX3 peak on a CTCF anchor is the ordinary case and not evidence that the KMT2A anchor is special. Q35's class-matched conclusion is unaffected by this entry, and a reader must not convert 'occupied' into 'enriched'. Q35-encode-reanalysis 2026
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    The peak is present in ENCODE's own pre-threshold files for two of the three experiments, in both biological replicates, and is removed by the IDR step. ENCSR000BRT (CBX3, K562): ENCFF271ZAZ gives chr11:118,488,467-118,488,897 at signalValue 22.390 with rep1 14.007 and rep2 8.488, localIDR -log10 0.2821 and globalIDR -log10 0.6448, i.e. IDR 0.227; the optimal-IDR file ENCFF951BQB that Q35 read (26,789 peaks) contains nothing there. ENCSR000EDB (BRCA1, HeLa-S3): ENCFF457RFT gives chr11:118,488,543-118,488,843 at signalValue 15.662 with rep1 9.214 and rep2 7.128, globalIDR -log10 1.2302, i.e. IDR 0.0588 against a 0.05 threshold; the conservative-IDR file ENCFF668JZI that Q35 read (9,452 peaks) contains nothing there.

    What this evidence cannot show

    IDR 0.227 is a genuinely weak reproducibility score and the CBX3 K562 call should not be read as a confident peak; the BRCA1 HeLa call at 0.0588 is the stronger of the two and is one hundredth from the standard cut. The columns read are the ENCODE narrowPeak-IDR local and global IDR fields (11 and 12) as -log10 values; they were converted, not read as probabilities. Neither experiment is in a KMT2A-rearranged cell and neither is a fetal or CD34+ progenitor.

    Evidence that argues against this (1)
    • ENCSR000BUH (CBX3, HCT116) is the case that does NOT corroborate: ENCODE's own IDR-ranked file ENCFF991DCE, with 65,155 peaks, has no peak at the element although ChIP-Atlas calls one at fold 12.5. So the two pipelines disagree in both directions and a reader should treat the HCT116 call as pipeline-dependent. ENCODE Project Consortium 2026
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    The instrument was controlled in both directions and the strongest control is a degron. A ChIP-Atlas CTCF track in GM12878 (SRX038504, 30,154 peaks) calls the element at fold 22.51 with summit 118,488,695, and in HCT116 the auxin-minus CTCF track (SRX11361143, 29,841 peaks) calls it at fold 12.46 while the auxin-plus track from the same degron line (SRX11361144, 9,347 peaks) calls nothing there. A 7,466 bp gene desert on chr2 returns zero peaks in 180 of 180 tracks, and the element is not in the ENCODE hg38 blacklist v2.

    PMID 42051034DOI
    What this evidence cannot show

    The degron pair differs in peak depth as well as in CTCF status (29,841 against 9,347), so loss of the element in the auxin-plus arm is confounded with lower sensitivity; it is suggestive, not decisive. Two BLUEPRINT CTCF tracks in K-562 with 4,134 and 7,570 peaks also miss the element, which shows the same confound from the other side and is reported rather than dropped.

    Evidence that argues against this (1)
    • Two of the six CTCF positive controls FAILED. If depth alone decides whether this element is called, then the difference between Q35's thirteen zeros and this entry's three positives may be depth and threshold together rather than threshold alone, and the honest framing is that the element is at the detection boundary of this assay class rather than that it is occupied or unoccupied. Zou Z et al. 2026
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    Against a class-matched null, BRCA1 and CBX3 behave differently at this anchor. Only 4,222 of 41,050 HeLa-S3 CTCF optimal-IDR anchors (10.3%) carry a called BRCA1 peak in SRX150631, and the KMT2A exon-11 anchor is one of them; by contrast 13,524 of 44,722 K562 CTCF anchors (30.2%) and 19,960 of 29,841 HCT116 CTCF anchors (66.9%) carry a CBX3 peak. The ENCODE HeLa-S3 CTCF anchor set independently contains the element at chr11:118,488,634-118,488,703 with signalValue 69.26.

    What this evidence cannot show

    The overlap fractions are sensitive to the anchor set chosen and to the peak-caller threshold on both sides, and no permutation test was run, so 10.3% against 30.2% is a description and not a tested difference. The HCT116 comparison uses a ChIP-Atlas CTCF track rather than an ENCODE anchor set, so the three rows are not strictly comparable to each other. Cross-cell-type comparison is confounded: HeLa, K562 and HCT116 differ in CTCF anchor number, ploidy and copy number at 11q23.

    Evidence that argues against this (1)
    • A 10.3% base rate still means roughly one in ten CTCF anchors carries a BRCA1 peak in this track, so a single anchor being in that set is weak evidence about that anchor specifically. The claim this supports is only that the anchor is occupied, not that it is preferentially occupied; asserting the latter would need the permutation this entry did not run. Oncofusion knowledge graph coordinator 2026
Connected to