start / graph / ana-q112-the-idr-free-signal-level-reread-run-the-pre-registered-prior-held

AnalysisstrongQ112-signal-level-reread

RUN: the IDR-free, signal-level re-read of the KMT2A exon-11 anchor against a class-matched CTCF-anchor census - the pre-registered prior of NEGATIVE HELD, both arms score FAILS on the pre-registered rule (BRCA1 median percentile 62.8 of 6 tracks, CBX3 55.9 of 3), and the one thing the prior did not predict is that 2 of 6 BRCA1 tracks sit above the 90th percentile and both are the non-blood cells

also called Q112 signal-level re-read · the 147 bp summit-matched class null, run

Two repair-related proteins were measured at one small spot inside the KMT2A gene in every public dataset that allows a fair comparison, and compared not to the whole genome but to tens of thousands of similar spots in the very same cells. On that fair comparison the spot is unremarkable, which is what had been predicted in writing before the data were opened - except in two non-blood cell types, where one of the proteins really is among the top few per cent.

The same thing, for a scientist

**The pre-registration this answers.** `exp-idr-free-signal-level-reread-of-the-kmt2a-anchor-across-every-archived-cbx3-and-brca1-track` (cycle 64) recorded, before any signal was read, the prior **"NEGATIVE"** and a decision rule: the anchor **FIRES** if its class-matched percentile exceeds the 75th in >= 60% of interpretable tracks of one protein while the six non-anchor intervals stay below the 50th, and **FAILS** if the median percentile across interpretable tracks lies in [25, 75]. Both are quoted verbatim in `analyses/q112-signal-level-reread/PREREGISTRATION.md`, which also discloses the single value read at the locus before that file was written (ENCFF690UED at chr11:118,488,614-760, mean 4.360008, a connectivity smoke test) [E1]. **THE ANSWER, and the prior was right.** | arm | interpretable tracks | excluded by C1 | median percentile | matched-null median | > 75th pct | six non-anchor intervals, median | **verdict** | |---|---:|---:|---:|---:|---:|---:|---| | **BRCA1** | 6 | 1 (ENCFF431CIQ, C1 4.00) | **62.84** | 69.90 | 2 of 6 (33%) | 47.58 | **FAILS** | | **CBX3** | 3 | 1 (ENCFF196UXL, C1 4.50) | **55.90** | 41.42 | 0 of 3 | 31.51 | **FAILS** | | CBX1 | 2 | 0 | 16.94 | 17.86 | 0 of 2 | 18.61 | NEITHER (below 25 - depleted, not ordinary) | Both named arms land inside the pre-registered FAILS band. **The anchor is an ordinary CTCF anchor for BRCA1 and for CBX3, and the prior recorded before the data were opened was correct.** [E1] **What the prior did not predict, and it is not noise.** The BRCA1 distribution is not a cloud around the median; it is bimodal and it splits by cell type: | track | cell | fold over anchor median | percentile | width+TSS-matched percentile | |---|---|---:|---:|---:| | ENCFF690UED | HeLa-S3 | 2.754 | **93.69** | 95.17 | | ENCFF620MRE | H1 | 2.146 | **92.28** | 95.66 | | ENCFF133FPU | HepG2 | 1.325 | 70.61 | 75.20 | | ENCFF191NNV | HepG2 | 1.089 | 55.08 | 64.61 | | ENCFF493XJG | K562 | 0.921 | 45.11 | 36.58 | | ENCFF017KXT | GM12878 | 0.754 | 34.01 | 26.23 | **The two above the 90th percentile are the two non-blood, non-hepatic cells, and the two blood lines are the two lowest.** Restricting to anchors matched within +/-25% of this anchor's peak width and within a factor of two of its distance to the nearest Ensembl TSS moves H1 to the 95.66th and HeLa-S3 to the 95.17th - the matched null makes the split *sharper*, not weaker [E1]. **This resolves, in the opposite direction to the guess, an open question the corpus has been carrying.** `ana-q35-brca1-cbx3-and-cbx1-are-not-at-the-kmt2a-breakpoint-cluster-in-any-encode-cell` asked: *"Is the exon-11 BRCA1 signal in H1 and HeLa-S3 (6.95x and 9.16x at 147 bp, absent from both blood lines) a CTCF-proximal artefact or a genuine cell-type-restricted event? Ranking that 147 bp window against 147 bp windows centred on CTCF summits, rather than against the element, would settle it in one."* **That is exactly the statistic computed here, and it does not dissolve.** At 147 bp against 147 bp summit-centred windows the two tracks read 92.28 and 93.69; at Q35's 1,048 bp geometry the same two tracks read 70.83 and 66.85. **The window geometry is worth ~25 percentile points at this element, and Q35 measured the geometry that hides it** [E1][E2]. **Reproduction of Q35 first, because a new number that cannot recover the old one is not a new number.** Recomputing Q35's own statistic - the 1,048 bp element against 1,048 bp windows centred on peak centres - gives 8.46 / 39.38 / 26.06 / 47.27 / 22.86 / 30.01 / 18.72 / 64.57 / 70.83 / 66.79 / 32.16 / 3.26 / 2.87 against Q35's published 9.97 / 40.2 / 22.72 / 48.87 / 26.13 / 26.43 / 14.67 / 64.95 / 70.07 / 68.57 / 19.87 / 5.02 / 4.93 (same file order). Eleven of thirteen agree within 6 percentile points; the largest gap is 12.3 (ENCFF845YDA, CBX3 HCT116), and the two runs differ in anchor sampling (a 42,947-54,654-anchor census here, a 6,000-anchor sample there). The C1 gate reproduces its exclusions exactly: ENCFF196UXL at 4.50 here against Q35's 4.57, ENCFF431CIQ at 4.00 against 3.93, the same two tracks excluded [E1][E2]. **Controls.** C1 ratios 7.95x to 57.23x over the 11 passing tracks. C2: the KMT2A promoter over the median of four live-fetched housekeeping promoters is 0.61-1.74 across the target tracks, reproducing Q35's 0.63-1.80 - the *KMT2A* promoter is an ordinary promoter, even though against the CTCF-anchor class it reads at the 96.05th percentile, which is a reminder that a promoter compared to loop anchors is not a matched comparison. C3: a live-fetched gene-free interval on chr2 reads 0.36-0.88. Coverage fraction is 1.000 at the exon-11 interval in 13 of 13 files and the modal raw-interval run length is 1-5 bp, so 147 bp is tens of independent values and never one bin. **C4 fails in 6 of 6 cell types and the reason is a false premise, not a blind pipeline** - see `genome-kmt2a-exon-11-ctcf-anchor-is-a-mid-rank-ctcf-anchor-in-encode-cells` [E1]. **The internal negative that makes the positive readable.** The DCAL cut point (chr11:118,484,362-118,485,361), 3.3 kb from the anchor and inside the same breakpoint cluster, reads at the **6.14th** percentile median across the 11 interpretable tracks. Whatever is at the anchor is at the anchor and not spread over the breakpoint region [E1].

Facts and figures

script	analyses/q112-signal-level-reread/main.py
pre registration	analyses/q112-signal-level-reread/PREREGISTRATION.md
results	analyses/q112-signal-level-reread/results.json
status	run
run date	2026-09-22
complete run	true
n tracks read	19
arm	ENCODE fold-change-over-control bigWigs only; the ChIP-Atlas read-depth arm is reported separately and is NOT pooled, because the pre-registration forbids pooling two normalisations
prior recorded before running	NEGATIVE
prior held	true
window geometry effect	147 bp centred on the narrowPeak point-source summit versus 1,048 bp centred on the peak centre changes the BRCA1 percentile by +22.9 points in HeLa-S3 (66.85 -> 93.69) and +21.5 in H1 (70.83 -> 92.28)
what this cannot close	the resting state in six non-haematopoietic-progenitor cell types, and nothing more. No CBX3 or BRCA1 track exists in a KMT2A-rearranged line, a CD34+ progenitor or a fetal haematopoietic cell, and no damage-state measurement exists at this locus in any cell type.

verdicts

per track

per interval median percentile over 11 interpretable tracks

verdicts — the numbers behind the chart

{
  "BRCA1": {
    "n_interpretable": 6,
    "n_excluded_C1": 1,
    "excluded": [
      "ENCFF431CIQ"
    ],
    "median_percentile": 62.84,
    "matched_median_percentile": 69.9,
    "n_above_75": 2,
    "frac_above_75": 0.333,
    "median_percentile_six_non_anchor_intervals": 47.58,
    "VERDICT": "FAILS",
    "percentiles": [
      34.01,
      45.11,
      55.08,
      70.61,
      92.28,
      93.69
    ],
    "folds": [
      0.754,
      0.921,
      1.089,
      1.325,
      2.146,
      2.754
    ]
  },
  "CBX3": {
    "n_interpretable": 3,
    "n_excluded_C1": 1,
    "excluded": [
      "ENCFF196UXL"
    ],
    "median_percentile": 55.9,
    "matched_median_percentile": 41.42,
    "n_above_75": 0,
    "frac_above_75": 0,
    "median_percentile_six_non_anchor_intervals": 31.51,
    "VERDICT": "FAILS",
    "percentiles": [
      12.58,
      55.9,
      64.26
    ],
    "folds": [
      0.259,
      1.072,
      1.196
    ]
  },
  "CBX1": {
    "n_interpretable": 2,
    "n_excluded_C1": 0,
    "median_percentile": 16.94,
    "matched_median_percentile": 17.86,
    "n_above_75": 0,
    "median_percentile_six_non_anchor_intervals": 18.61,
    "VERDICT": "NEITHER",
    "percentiles": [
      6.61,
      27.28
    ],
    "folds": [
      0.256,
      0.684
    ]
  }
}

per track — the numbers behind the chart

{
  "ENCFF690UED": {
    "target": "BRCA1",
    "cell": "HeLa-S3",
    "C1": 57.23,
    "C1_pass": true,
    "exon11_pct": 93.69,
    "fold": 2.754,
    "matched_pct": 95.17,
    "elem1048_pct": 66.85,
    "modal_run_bp": 1,
    "covered_fraction": 1
  },
  "ENCFF620MRE": {
    "target": "BRCA1",
    "cell": "H1",
    "C1": 32.25,
    "C1_pass": true,
    "exon11_pct": 92.28,
    "fold": 2.146,
    "matched_pct": 95.66,
    "elem1048_pct": 70.83,
    "modal_run_bp": 1,
    "covered_fraction": 1
  },
  "ENCFF133FPU": {
    "target": "BRCA1",
    "cell": "HepG2",
    "C1": 7.96,
    "C1_pass": true,
    "exon11_pct": 70.61,
    "fold": 1.325,
    "matched_pct": 75.2,
    "elem1048_pct": 47.27,
    "modal_run_bp": 4,
    "covered_fraction": 1
  },
  "ENCFF191NNV": {
    "target": "BRCA1",
    "cell": "HepG2",
    "C1": 25.16,
    "C1_pass": true,
    "exon11_pct": 55.08,
    "fold": 1.089,
    "matched_pct": 64.61,
    "elem1048_pct": 22.86,
    "modal_run_bp": 1,
    "covered_fraction": 1
  },
  "ENCFF493XJG": {
    "target": "BRCA1",
    "cell": "K562",
    "C1": 14.54,
    "C1_pass": true,
    "exon11_pct": 45.11,
    "fold": 0.921,
    "matched_pct": 36.58,
    "elem1048_pct": 64.57,
    "modal_run_bp": 1,
    "covered_fraction": 1
  },
  "ENCFF017KXT": {
    "target": "BRCA1",
    "cell": "GM12878",
    "C1": 32.46,
    "C1_pass": true,
    "exon11_pct": 34.01,
    "fold": 0.754,
    "matched_pct": 26.23,
    "elem1048_pct": 8.44,
    "modal_run_bp": 2,
    "covered_fraction": 1
  },
  "ENCFF431CIQ": {
    "target": "BRCA1",
    "cell": "K562",
    "C1": 4,
    "C1_pass": false,
    "exon11_pct": 14.04,
    "fold": 0.402,
    "matched_pct": 14.19,
    "elem1048_pct": 18.75,
    "modal_run_bp": 1,
    "covered_fraction": 1
  },
  "ENCFF051QOQ": {
    "target": "CBX3",
    "cell": "K562",
    "C1": 25.61,
    "C1_pass": true,
    "exon11_pct": 64.26,
    "fold": 1.196,
    "matched_pct": 59.97,
    "elem1048_pct": 39.37,
    "modal_run_bp": 1,
    "covered_fraction": 1
  },
  "ENCFF845YDA": {
    "target": "CBX3",
    "cell": "HCT116",
    "C1": 7.95,
    "C1_pass": true,
    "exon11_pct": 55.9,
    "fold": 1.072,
    "matched_pct": 41.42,
    "elem1048_pct": 32.18,
    "modal_run_bp": 1,
    "covered_fraction": 1
  },
  "ENCFF888LWP": {
    "target": "CBX3",
    "cell": "K562",
    "C1": 8.86,
    "C1_pass": true,
    "exon11_pct": 12.58,
    "fold": 0.259,
    "matched_pct": 11.28,
    "elem1048_pct": 3.26,
    "modal_run_bp": 1,
    "covered_fraction": 1
  },
  "ENCFF196UXL": {
    "target": "CBX3",
    "cell": "GM12878",
    "C1": 4.5,
    "C1_pass": false,
    "exon11_pct": 43.58,
    "fold": 0.902,
    "matched_pct": 44.11,
    "elem1048_pct": 30,
    "modal_run_bp": 1,
    "covered_fraction": 1
  },
  "ENCFF058NFI": {
    "target": "CBX1",
    "cell": "K562",
    "C1": 14.53,
    "C1_pass": true,
    "exon11_pct": 27.28,
    "fold": 0.684,
    "matched_pct": 26.67,
    "elem1048_pct": 26.05,
    "modal_run_bp": 1,
    "covered_fraction": 1
  },
  "ENCFF895JUC": {
    "target": "CBX1",
    "cell": "HepG2",
    "C1": 16.32,
    "C1_pass": true,
    "exon11_pct": 6.61,
    "fold": 0.256,
    "matched_pct": 9.04,
    "elem1048_pct": 2.87,
    "modal_run_bp": 5,
    "covered_fraction": 1
  }
}

per interval median percentile over 11 interpretable tracks — the numbers behind the chart

{
  "exon11_147bp": 55.08,
  "ctcf_element_1048bp": 32.18,
  "ctcf_summit_1kb": 30.84,
  "dcal_cut_1kb": 6.14,
  "bcr_core_7466bp": 23.88,
  "kmt2a_tss_2kb": 96.05,
  "kmt2a_gene_90377bp": 50.24
}

q35 reproduction 1048bp centre geometry

{
  "this_run": [
    8.46,
    39.38,
    26.06,
    47.27,
    22.86,
    30.01,
    18.72,
    64.57,
    70.83,
    66.79,
    32.16,
    3.26,
    2.87
  ],
  "q35_published": [
    9.97,
    40.2,
    22.72,
    48.87,
    26.13,
    26.43,
    14.67,
    64.95,
    70.07,
    68.57,
    19.87,
    5.02,
    4.93
  ],
  "file_order": [
    "ENCFF017KXT",
    "ENCFF051QOQ",
    "ENCFF058NFI",
    "ENCFF133FPU",
    "ENCFF191NNV",
    "ENCFF196UXL",
    "ENCFF431CIQ",
    "ENCFF493XJG",
    "ENCFF620MRE",
    "ENCFF690UED",
    "ENCFF845YDA",
    "ENCFF888LWP",
    "ENCFF895JUC"
  ],
  "max_abs_difference_percentile_points": 12.3,
  "c1_exclusions_reproduced": "ENCFF196UXL 4.50 here vs 4.57 there; ENCFF431CIQ 4.00 here vs 3.93 there; same two tracks excluded"
}

controls

{
  "C1": "track's own top-500 IDR peaks / median of 20,000 random 1,048 bp tiles; >= 5x required; 11 of 13 pass, ratios 7.95-57.23",
  "C2_housekeeping": "KMT2A TSS +/-1 kb over the median of ACTB, GAPDH, RPLP0, EEF1A1 promoters (Ensembl GRCh38, fetched live 2026-09-22): 0.61-1.74 across target tracks",
  "C3_negative": "chr2:52,961,485-53,470,446, the largest non-pericentromeric Ensembl-gene-free interval on chr2, computed live from a BioMart census of 5,687 chr2 genes: 0.36-0.88",
  "C3_note": "the LARGEST chr2 gene-free interval (chr2:92,035,001-94,185,554, 2.15 Mb) is pericentromeric and therefore also a mappability desert, so it cannot separate absent-protein from unmappable; it is recorded but the non-pericentromeric one is the interpretable negative",
  "C4": "FAILS in 6 of 6 cell types; the premise that the anchor is a top-decile CTCF anchor is false"
}

traps cleared not assumed

[
  "build read from each FILE's own chrom table across chr1, chr2, chr11, chr19 and chrX against live Ensembl GRCh38 lengths - 19 of 19 GRCh38, nothing lifted, the +129,285 bp offset never used",
  "pybigtools 0.3.0 NaN default: every interval read twice, fillna=None for coverage and fillna=0 for the statistic; the deprecation warning confirming the flipped default was observed on the first call",
  "bins=1 passed explicitly on every summary read",
  "HTTP HEAD Content-Length compared against the portal file_size for every bigWig, and md5 recomputed on download for every peak file",
  "modal raw-interval run length measured per file over chr11:118,480,000-118,500,000: 1-5 bp, so the 147 bp interval is never one bin"
]

Still open

Does the H1 and HeLa-S3 BRCA1 enrichment reproduce in the individual biological replicate fold-change files, rather than only in the pooled track? ENCSR000EDB and ENCSR000EBX each serve per-replicate fold-change bigWigs and this run read only the pooled ones, so a two-replicate agreement at the 90th percentile has not been demonstrated at signal level even though it was demonstrated at peak level in the pre-IDR file.
Is the 22-to-23-point gap between the 147 bp summit geometry and the 1,048 bp centre geometry specific to this element, or is it what any sharp peak does? Recomputing both geometries for 1,000 random CTCF anchors in the same tracks would give the null for the geometry effect itself, and it has not been done.
Why is CBX1 DEPLETED at this anchor - 6.61st and 27.28th percentile, fold 0.256 and 0.684 - when CBX3 is ordinary there? Depletion of one HP1 paralogue below the CTCF-anchor median in both available tracks is a more specific observation than either protein being ordinary, and no entry in this graph predicts it.

Every source behind this entry (2)

E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Against a census of every conservative-IDR CTCF anchor of the same cell type, the KMT2A exon-11 147 bp interval sits at a median percentile of 62.84 across six interpretable ENCODE BRCA1 tracks and 55.90 across three CBX3 tracks - inside the pre-registered FAILS band of [25, 75] for both - while two BRCA1 tracks, H1 and HeLa-S3, reach the 92.28th and 93.69th percentiles and both blood lines sit lowest.
ENCODE BRCA1, CBX3 and CBX1 ChIP-seq fold-change-over-control bigWigs and IDR peak sets, GRCh38, read 2026-09-22: ENCFF690UED, ENCFF620MRE, ENCFF133FPU, ENCFF191NNV, ENCFF493XJG, ENCFF017KXT, ENCFF431CIQ, ENCFF051QOQ, ENCFF845YDA, ENCFF888LWP, ENCFF196UXL, ENCFF058NFI, ENCFF895JUC, ranked against ENCFF041ODC, ENCFF796WRU, ENCFF497ZGV, ENCFF533TKY, ENCFF695QSJ and ENCFF850PKJ · ENCODE Project Consortium · ENCODE Portal 2026
What this evidence cannot show
Six cell types, none of them haematopoietic progenitor and none KMT2A-rearranged. The CBX3 arm has only three interpretable tracks, below the pre-registration's own n >= 10 requirement, so its FAILS verdict is scored on a sample the design declared too small; the ChIP-Atlas arm was run separately to address this and cannot be pooled with it. The two HepG2 BRCA1 tracks and the two K562 BRCA1 tracks are not independent cell types, so six tracks are four cell types. A percentile among conservative-IDR CTCF anchors is a rank statistic and says nothing about absolute occupancy.
Evidence that argues against this (1)
- ENCODE's own pre-IDR peak files place a called peak at this element in both biological replicates of two experiments - ENCFF457RFT (BRCA1 HeLa-S3, global IDR 0.0588, missing the 0.05 cut by 0.009) and ENCFF271ZAZ (CBX3 K562, global IDR 0.227) - so the peak-level evidence is positive at a threshold the optimal and conservative files discard. A median class-matched percentile of 62.8 does not contradict that; it says the element is not unusual among its own class, which is a different claim from 'nothing is bound there'. ENCODE Project Consortium 2026
E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
The same pipeline reproduces Q35's published 1,048 bp class-matched percentiles within 6 points in 11 of 13 tracks and reproduces its two C1 exclusions exactly, so the 147 bp summit-centred numbers are a change of window geometry and not a change of instrument.
ana-q35-brca1-cbx3-and-cbx1-are-not-at-the-kmt2a-breakpoint-cluster-in-any-encode-cell - kmt2a_element_percentile_among_anchors and c1_gate, analyses/q35-brca1-cbx3-encode-at-the-kmt2a-locus/results.json · Q35-encode-reanalysis · this graph 2026
What this evidence cannot show
Agreement was checked on the 1,048 bp centre geometry only; Q35 did not compute a 147 bp summit geometry, so there is nothing to reproduce there and the +21 to +23 percentile-point shift is unreplicated by any second party.
Evidence that argues against this (1)
- A rank statistic that moves 23 percentile points on a change of window width and centring is fragile by construction, and nothing in this design says which geometry is correct. A summit-centred narrow window is the geometry that favours a sharp peak, so choosing it after seeing that the element is a sharp peak would be circular - it is defensible here only because the corpus asked for exactly this geometry in writing, in an entry dated before the run. ENCODE Project Consortium 2012

Connected to

tests →Read all 76 archived CBX3 and BRCA1 tracks at the KMT2A exon-11 anchor at SIGNAL level against a CTCF-anchor-matched null, and score the anchor's rank among its own class in each - no wet work, no new data, and it decides whether the sub-IDR peak is a foothold or a threshold artefact
the run of the pre-registered signal-level re-read, reporting against the prior that was recorded before any data was read
depends on →Every peak file of all 13 ENCODE BRCA1, CBX3 and CBX1 experiments opened at the exon-11 element, not just the two Q37 reached: BRCA1 is there in BOTH pre-IDR replicates of THREE experiments - HeLa-S3, H1 and HepG2, global IDR 0.059, 0.577 and 0.482 - and it is a clean zero in both K562 experiments and in GM12878 against 300,000-peak pre-IDR lists
the peak-level census the signal-level reading is interpreted against
related to →Measured in six ENCODE cell types against every conservative-IDR CTCF anchor of the same cell: the KMT2A exon-11 anchor sits at the 70.6th percentile of CTCF occupancy (range 28.2-81.8, fold over the anchor median 0.69-2.37) - a called peak in 6 of 6 cells and an ORDINARY one, which fails the instrument control the corpus pre-registration wrote for this very experiment
the class-matched CTCF-anchor ranking that the signal-level result must be read against