Oncofusion

start / graph / ana-q116-trim28-comes-closer-to-firing-at-the-exon-11-anchor-than-anything-yet-and-the-specificity-clause-blocks-it

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q116-h3k9-axis-at-the-locus

NEITHER, and it is the closest call this project has had: KAP1/TRIM28 reads the 90.62nd, 81.85th and 89.45th percentile of MCF-7 CTCF anchors at the exon-11 element in three of four tracks - satisfying the first clause of the pre-registered PRESENT rule - and the six-non-anchor-interval clause blocks it at 65.87

also called the TRIM28 near-miss · the specificity clause earns its keep a second time · three of four tracks above the 81st percentile and still not a positive

One protein came closer than any before it to being found at the exact spot where KMT2A breaks - in three of four measurements. It still fails, because everything else nearby is elevated too, so the spot is not being singled out.

The same thing, for a scientist

**The numbers, and they are the highest any protein has reached at this element.** At the 146 bp exon-11 CTCF anchor, KAP1/TRIM28 reads the **90.62nd, 0.00th, 81.85th and 89.45th** percentile of **56,171 same-cell-type MCF-7 CTCF summit windows** - **three of four tracks above the 81st**, which is **75% and clears the pre-registered 60% clause of the PRESENT rule.** No other protein in five cell types has done that [E1]. **It fails on the second clause, which is exactly what that clause is for.** The PRESENT rule requires *both* `percentile(exon11) > 75` in ≥ 60% of tracks **AND** the median of the six non-anchor intervals **< 50**. TRIM28's six-non-anchor median is **65.87**. The whole locus is elevated: promoter **99.68-99.91** in all four tracks, gene body 49.90-75.12, BCR core elevated too. **The anchor is not picked out; the locus is up.** Verdict: **NEITHER**, reported in those words [E1]. **This is the second time in two cycles that this clause has stopped a true, class-matched and thoroughly misleading sentence.** Q114 recorded exactly the same rescue for SUV39H1, whose anchor reached the 96.29th percentile in one MCF-7 replicate while the whole 90 kb gene body sat at the 82nd-87th. The clause was added to the Q112 design almost as bookkeeping and it has now blocked two different proteins in the same cell line. **And the panel makes the diagnosis rather than leaving it a shrug.** Across **28 C1-passing tracks of eight antigens** in this one deposit, the six-non-anchor-interval median is **49.79 to 72.37 for every protein antigen** and **15.41 for H3K9me3**. **Every protein ChIP in this deposit reads the whole *KMT2A* locus up.** That is a property of the locus, the library or the input-free normalisation - not of eight unrelated proteins [E2]. **The one zero is the shallowest library in the deposit.** GSM4133830 is **61,995,812 bytes**, 2.2-6x smaller than every other file in the panel, and it is the track that reads 0.00. So the honest statement is **three tracks at 81.85-90.62 and one uninformative**, and the median of 85.65 is a better summary than the range suggests - which makes the specificity clause the *only* thing standing between this and a published positive.

Go deeper
The full decision, clause by clause
clauserequirementmeasuredpasses?
percentile(exon11) > 75 in ≥ 60% of C1-passing tracks0.600.75 (3 of 4: 90.62, 81.85, 89.45)YES
median of the six non-anchor intervals < 50< 5065.87NO
verdictNEITHER

All seven intervals, four tracks, percentile of width-matched MCF-7 CTCF summit windows:

intervalGSM4133828 (-E2)GSM4133829 (-E2)GSM4133830 (+E2)GSM4133831 (+E2)
exon1190.6281.850.0089.45
kmt2a_tss_2kb99.7099.8999.9199.68
kmt2a_gene61.8075.1249.9071.04
C1 dynamic range11.5x16.8x11.1x11.7x

Note the contrast in reproducibility: the 2 kb promoter window agrees to 0.23 percentile points across all four tracks; the 146 bp anchor window spans 90.6 points. Same files, same day, same antibody.

GSM4133828 and GSM4133829 are two replicates of the same condition and read 90.62 and 81.85 - so the high anchor value is replicated within a condition, which is more than Q114's SUV39H1 near-miss had.

What would turn this into a positive, and what it would take

The PRESENT rule fails only on locus-wide elevation, and locus-wide elevation in a deposit with no input track is exactly what an input track would resolve. Three concrete routes, in order of cost:

  1. An input or IgG track for GSE139196. If TRIM28's exon-11 signal survives input normalisation while the rest of the locus does not, the verdict inverts. Neither GSE139196 nor GSE225028 deposits one; the submitters generated input for the ChIRC-seq arm by their own processing notes and did not deposit it for the ChIP arm. This is a one-email question.
  2. The 636 MB and 684 MB KAP1 bedGraphs of GSM5660804 and GSM5660805, out of this run's disk budget. They are the deepest KAP1 tracks in the deposit and would give two more independent anchor readings - and their peak files already call zero peaks within 52 kb of the anchor, which cuts against.
  3. A same-cell CTCF track re-centred on MCF-7's own summit (118,488,689). The C4 instrument control has failed in seven of seven cell types for a geometric reason: a fixed external 146 bp window scored against each cell's own summit-centred windows. Until C4 can pass somewhere, no positive at this element is interpretable anyway.

And the thing that cuts hardest against it: TRIM28's own peak-level data, from two independent samples of the same cell line, call zero peaks in the exon-11 window, zero in the CTCF element and zero in the 7,465 bp BCR core, with the nearest chr11 peak 52,230 bp away. A signal percentile of 90 with no called peak at a point-source-callable target is the signature of diffuse local elevation, not of a binding site.

Facts and figures
scriptanalyses/q116-h3k9-axis-at-the-locus/run116.py and score.py
prereganalyses/q116-h3k9-axis-at-the-locus/PREREGISTRATION.md, section 8
statusrun
run date2026-09-23
run byQ116-h3k9-axis-at-the-locus
complete runtrue
verdictNEITHER
exon11 median85.65
kmt2a tss 2kb spread percentile points0.23
shallowest track is the zeroGSM4133830, 61,995,812 bytes, 2.2-6x smaller than every other file in the panel
second time this clause has blocked a near positiveQ114 recorded the same rescue for SUV39H1: 96.29th percentile at the anchor in one MCF-7 replicate, blocked because the whole 90 kb gene body sat at the 82nd-87th. The clause originated in analyses/q112-signal-level-reread/PREREGISTRATION.md.
localizationchromatin at chr11q23.3 - the 146 bp exon-11 CTCF anchor, the six other stored intervals, and the KMT2A promoter
subnuclear localizationnot measured
cell typeMCF-7 (ER+ breast adenocarcinoma). NOT haematopoietic, NOT KMT2A-rearranged.
cell statehormone-deprived culture and 40 min 17-beta-oestradiol
epigenetic readoutoccupancy of KAP1/TRIM28, the H3K9me3 corepressor scaffold, at the KMT2A exon-11 CTCF anchor
shown for fusionNO - MCF-7 carries no KMT2A rearrangement
absence classmeasured-not-analysed
absence class noteThe four tracks were public since 2019. What is missing and cannot be closed by re-analysis is the input arm, which is never-measured for this deposit.
panel median provenanceC5/C5b PANEL MEDIANS ARE QUOTED AT THE FINAL 28-TRACK COUNT (2026-09-23). Earlier drafts of this entry quoted them at intermediate counts (12, 18 or 25 tracks) while the run was still going. The gene-body panel median moved 61.80 -> 61.39; the PROMOTER panel median moved 82.36 -> 96.87, i.e. UP, and the direction matters: at 96.87 the CTCF-anchor percentile at the KMT2A promoter is even less informative than the first draft said, and the promoter claim rests ENTIRELY on the active-promoter null of deviation D9.
results
[
  "analyses/q116-h3k9-axis-at-the-locus/results.json",
  "analyses/q116-h3k9-axis-at-the-locus/results_scored.json"
]
clause 1 percentile above 75 in at least 60 percent of tracks
{
  "required": 0.6,
  "measured": 0.75,
  "passes": true
}
clause 2 six nonanchor interval median below 50
{
  "required": 50,
  "measured": 65.87,
  "passes": false
}
exon11 percentiles
[
  90.62,
  0,
  81.85,
  89.45
]
exon11 spread percentile points
{
  "across_all_four_tracks_max_minus_min": 90.62,
  "within_condition_minusE2_pair": 8.77,
  "within_condition_plusE2_pair": 89.45,
  "note": "the two numbers answer different questions; only the within-condition pair spreads are replicate noise. The +E2 pair contains GSM4133830, the shallowest file in the deposit."
}
exon11 within condition replicates minusE2
[
  90.62,
  81.85
]
kmt2a tss 2kb percentiles
[
  99.7,
  99.89,
  99.91,
  99.68
]
kmt2a gene percentiles
[
  61.8,
  75.12,
  49.9,
  71.04
]
c1 ratios
[
  11.5,
  16.8,
  11.1,
  11.7
]
panel baseline C5
{
  "n_C1_passing_tracks": 28,
  "panel_median_kmt2a_gene_percentile": 61.8,
  "six_nonanchor_interval_median_per_antigen": {
    "CBX5": 51.76,
    "CBX5_mut4A": 49.79,
    "CBX5_mut4E": 72.37,
    "EHMT2": 52.56,
    "KDM4B": 70.31,
    "KDM4C": 72.05,
    "TRIM28": 65.87,
    "H3K9me3": 15.41
  },
  "reading": "every PROTEIN antigen in this deposit reads the whole KMT2A locus elevated; only the histone mark does not"
}
what would invert it
[
  "an input or IgG track for GSE139196 (one email)",
  "the two 636-684 MB KAP1 bedGraphs of GSM5660804/805, out of this run's disk budget",
  "a same-cell CTCF track re-centred on MCF-7's own summit 118,488,689, so C4 can pass"
]
C5 final 28 track values
{
  "panel_median_kmt2a_gene_percentile": 61.39,
  "panel_median_kmt2a_promoter_percentile": 96.87,
  "per_antigen_median_promoter_percentile_vs_CTCF_anchors": {
    "TRIM28": 99.8,
    "CBX5_mut4E": 99.75,
    "KDM4B": 99.37,
    "KDM4C": 98.47,
    "CBX5_mut4A": 96.76,
    "CBX5": 96.65,
    "EHMT2": 95.83,
    "H3K9me3": 29.43
  },
  "reading": "every PROTEIN antigen reads 95.8-99.8 at the KMT2A promoter against CTCF anchors and only the histone mark (29.43) does not - so that statistic cannot discriminate between proteins at all, and every promoter claim in this beat rests on the ACTIVE-PROMOTER null instead"
}
How this could bite us back
  • If TRIM28 really were at the exon-11 anchor, the therapeutic reading would be serious rather than academic: KAP1 is phosphorylated at S824 by DNA-PK during the double-strand-break response, and ATM and DNA-PK inhibitors are in clinical trials. An agent that changes KAP1 behaviour at a break-prone anchor during a TOP2-poison exposure is the combination this graph's risk entries already worry about, and the sign is not predictable from occupancy.
  • The verdict here is NEITHER and it must not be quoted as a positive. Three tracks at 81.85-90.62 are a real measurement and they do not survive the specificity clause, and the same protein's own peak-level data from the same cell line call zero peaks within 52 kb. Anyone who wants to act on this needs the input track first, not a re-reading of these four numbers.
Still open
  1. Does an input track exist for GSE139196? The submitters used input for the ChIRC-seq arm by their own processing notes. If they will share the ChIP input, this verdict can be recomputed the same day and it is the only cheap route to inverting it.
  2. Is the whole-locus elevation copy number? CCLE calls KMT2A copy-neutral in MCF-7, but a discrete caller reports 0 for exactly the size of gain that would produce a 1.1-1.3x elevation. The segment-level log2 ratio at 11q23 was not retrieved by Q114 and has not been retrieved here.
  3. Would the two deep 2021 KAP1 bedGraphs agree with the four 2019 tracks at the anchor? They are the deepest KAP1 data in the deposit, their peak files call zero peaks within 52 kb, and their signal was never read.
  4. Is TRIM28 at this element in a haematopoietic cell? Q110's single K562 TRIM28 peak set says zero peaks in the gene or within 50 kb. That is one peak set in one blood line, and it is the only blood answer there is.
Every source behind this entry (2)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    In four MCF-7 KAP1/TRIM28 ChIP-seq tracks, the 146 bp KMT2A exon-11 window reads the 90.62nd, 0.00th, 81.85th and 89.45th percentile of 56,171 same-cell-type CTCF summit windows - 75% of tracks above the 75th percentile, clearing the pre-registered 60% clause - while the median of the six non-anchor intervals is 65.87 against a required < 50, so the pre-registered verdict is NEITHER. The 2 kb promoter window reads 99.70, 99.89, 99.91 and 99.68 in the same four tracks.

    GSE139196: ChIP-seq on KAP1 (TRIM28) in MCF-7 with and without estradiol, GSM4133828-31, hg38 bedGraph · Yang F, Tanasa B, Micheletti R, Ohgi KA, Aggarwal AK, Rosenfeld MG · NCBI GEO 2019
    What this evidence cannot show

    MCF-7, an ER+ breast adenocarcinoma, not KMT2A-rearranged and not haematopoietic. No input, no IgG, no antibody clone. GSM4133830 is 61,995,812 bytes, the shallowest file in the panel by a factor of 2.2-6, and it is the track that reads 0.00 - so the range understates and the median of 85.65 overstates. The C4 instrument control was declared FAILED in advance for this cell type (72.97th percentile against a required 90th, Q114), so no verdict here is labelled a class-matched positive in any case. 32 of 56,203 CTCF peaks were dropped for falling on non-primary scaffolds (0.057%), counted not silent.

    Evidence that argues against this (1)
    • TRIM28's OWN PEAK-LEVEL DATA FROM THE SAME CELL LINE CUT AGAINST THIS. Two independent HOMER peak files (GSM5660804, GSM5660805) call zero peaks in the exon-11 window, zero in the 1,047 bp CTCF element and zero in the 7,465 bp BCR core, with the nearest chr11 peak 52,230 and 52,149 bp from the summit. A signal percentile of 81-91 with no called peak at a point-source-callable target is diffuse local elevation, not a binding site - and that reading is consistent with the locus-wide elevation that already blocked the verdict. source
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    Across 28 C1-passing tracks of eight antigens in GSE139196, the median of the six non-anchor KMT2A intervals is 49.79-72.37 for every protein antigen (CBX5 51.76, CBX5-4A 49.79, CBX5-4E 72.37, EHMT2 52.56, KDM4B 70.31, KDM4C 72.05, TRIM28 65.87) and 15.41 for H3K9me3 - so every protein ChIP in this deposit reads the whole KMT2A locus elevated, and the panel median kmt2a_gene percentile is 61.80.

    What this evidence cannot show

    One deposit, one submitter, one HOMER pipeline, one cell line. The panel controls for that pipeline's behaviour at this locus and not for ChIP-seq in general. The panel median moves as tracks are added and is quoted at the track count recorded in results_scored.json. All eight antigens are H3K9-axis proteins, so they are not an unbiased sample of chromatin proteins - a shared true biology at this locus cannot be excluded by the panel alone, which is the whole reason an input track is the control that is missing.

    Evidence that argues against this (1)
    • The panel is eight H3K9-axis antigens, not eight random chromatin proteins. If the KMT2A locus genuinely carries a shared H3K9-machinery platform, a panel of H3K9 proteins would look exactly like this and the 'locus-level artefact' reading would be wrong. The panel cannot distinguish the two; an input track can, and does not exist. This is the single largest uncertainty in this beat. source
Connected to