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
also called the 13-experiment pre-IDR census · which ENCODE cells actually have a peak at the KMT2A anchor
Public datasets usually publish only the peaks that survive a strict reproducibility filter. Opening every unfiltered list instead shows the repair protein BRCA1 sitting on this one spot in both repeats of three different cell types - two more than anyone here had checked - and genuinely absent from both blood-cell datasets even in lists three hundred thousand entries long.
The same thing, for a scientist
**What was incomplete.** `ana-q37-the-kmt2a-exon-11-anchor-carries-a-cbx3-and-a-brca1-peak-that-idr-removes` opened ENCODE's own pre-IDR files for **two** experiments, ENCSR000BRT (CBX3 K562) and ENCSR000EDB (BRCA1 HeLa-S3), and reached the rest through ChIP-Atlas's re-processing. This is every released GRCh38 peak file of **all thirteen** experiments - **124 files** in total - queried at chr11:118,488,614-118,488,760 [E1]. | experiment | target | cell | element present in ENCODE's own files? | best global IDR | deepest list searched | |---|---|---|---|---:|---:| | ENCSR000EDB | BRCA1 | HeLa-S3 | **YES** - pooled + rep1 | **0.0588** | 67,003 | | ENCSR000EBX | BRCA1 | H1 | **YES - BOTH reps + pooled** | **0.577** | 300,000 | | ENCSR000EDY | BRCA1 | HepG2 | **YES - BOTH reps + pooled** | **0.482** | 300,000 | | ENCSR343RJR | BRCA1 | HepG2 | no | - | 108,511 | | ENCSR000DZS | BRCA1 | GM12878 | **no** | - | **300,000** | | ENCSR223MLH | BRCA1 | K562 | **no** | - | **300,000** | | ENCSR946BXO | BRCA1 | K562 | **no** | - | 124,863 | | ENCSR000BRT | CBX3 | K562 | **YES - BOTH reps + pooled** | **0.227** | 300,000 | | ENCSR000BUH | CBX3 | HCT116 | **rep2 ONLY** | 0.244 | 114,452 | | ENCSR000ATV | CBX3 | K562 | no (single-replicate experiment) | - | **300,000** | | ENCSR549NPZ | CBX3 | GM12878 | no | - | 63,082 | | ENCSR948QLZ | CBX1 | K562 | **no** | - | **300,000** | | ENCSR819WZE | CBX1 | HepG2 | no | - | 90,702 | **Three things this changes.** **1. The BRCA1 evidence doubles, and it is replicate-level.** H1 (ENCFF076ZNS rep1 signalValue 3.87, ENCFF024UWC rep2 4.13, ENCFF338YFB pooled 9.15, ENCFF887JQM IDR-ranked pooled 12.68 at global IDR 0.577) and HepG2 (ENCFF025YGD rep1 4.55, ENCFF383UGF rep2 6.06, ENCFF720ZDL pooled 9.53, global IDR 0.482 and 0.888) join HeLa-S3. **Three of seven BRCA1 experiments carry the element in both biological replicates before thresholding.** Neither H1 nor HepG2 was named in the cycle-64 entry or in the dispatch brief for this beat [E1]. **2. The CBX3 HCT116 call is weaker in ENCODE's own files than in ChIP-Atlas's.** ChIP-Atlas called it at fold 12.50, -log10 q 33.30. In ENCODE's pipeline it appears in **rep2 only** (ENCFF161HBO, 101,089 peaks, global IDR 0.244) and in neither the rep1 list (65,155) nor the pooled list (114,452). A one-replicate call in the deeper of two lists is exactly what IDR exists to discard, and here it is the honest description [E1]. **3. The blood-line zeros are now bounded, and bounded hard.** BRCA1 GM12878 and BRCA1 K562 (ENCSR223MLH) each serve **three 300,000-peak `peaks and background as input for IDR` files** - the most permissive product ENCODE publishes - and the element appears in none of them. CBX1 K562 likewise, across three 300,000-peak lists. **An absence against a 300,000-peak list is a different object from an absence against a 9,452-peak conservative-IDR list**, and it is the first time this graph has had one at this element [E1][E2]. **The concordance with signal level is the part worth keeping.** The two ENCODE cells where BRCA1 is a clean zero in a 300,000-peak list are the two where the signal-level class-matched percentile is lowest (GM12878 34.01, K562 45.11), and the two cells where BRCA1 reaches the 92nd-94th percentile at signal level are two of the three where it is present in both pre-IDR replicates (H1, HeLa-S3). **Two independent readouts of the same experiments agree on which cells have it.** The exception is honest and worth naming: HepG2 ENCSR000EDY is peak-positive in both reps yet sits at only the 55.08th percentile at signal level, while the peak-negative HepG2 experiment ENCSR343RJR sits at 70.61 [E1].
| script | analyses/q112-signal-level-reread/preidr_census.py |
|---|---|
| status | run |
| run date | 2026-09-22 |
| complete run | true |
| interval | chr11:118,488,614-118,488,760 (GRCh38, 147 bp) |
| n experiments | 13 |
| n peak files opened | 124 |
| col12 convention | narrowPeak column 12 in an ENCODE IDR-ranked file is -log10(globalIDR); globalIDR = 10^-col12 |
| persisted result | analyses/q112-signal-level-reread/preidr_census.json - every file, its peak count, its replicate membership, the hit row and the global IDR (rule 90: the members are persisted, not only the counts) |
| globalIDR reporting convention | Where an experiment released MORE THAN ONE IDR-ranked file, the figure quoted in this entry is the BEST (lowest) globalIDR among them, not an average and not the first file. HepG2 ENCSR000EDY is the case that makes this visible: ENCFF047KJJ gives 0.8876 and ENCFF989POH gives 0.4816 for the same interval and the same replicate pair, and 0.482 is what the title reports. HeLa-S3 likewise has three hitting files (0.0589, 0.0675, and 0.1454 for a single-replicate file). Stated here because quoting the best of several without saying so is exactly the shape of selective reporting, and a reader re-running this will open a different file and get a different number. |
positive
{
"ENCSR000EDB": {
"target": "BRCA1",
"cell": "HeLa-S3",
"files": {
"ENCFF457RFT": {
"n": 62322,
"reps": [
1,
2
],
"peak": "118488543-118488843",
"signalValue": 15.66241,
"globalIDR": 0.0588
},
"ENCFF580QYS": {
"n": 67003,
"reps": [
1,
2
],
"globalIDR": 0.0675
},
"ENCFF649XNX": {
"n": 39255,
"reps": [
1
],
"peak": "118488513-118488793",
"signalValue": 9.21362,
"globalIDR": 0.1454
}
},
"both_replicates": true
},
"ENCSR000EBX": {
"target": "BRCA1",
"cell": "H1",
"files": {
"ENCFF076ZNS": {
"n": 300000,
"reps": [
1
],
"peak": "118488488-118488778",
"signalValue": 3.8657
},
"ENCFF024UWC": {
"n": 300000,
"reps": [
2
],
"peak": "118488533-118488823",
"signalValue": 4.1342
},
"ENCFF338YFB": {
"n": 67776,
"reps": [
1,
2
],
"peak": "118488525-118488821",
"signalValue": 9.152
},
"ENCFF887JQM": {
"n": 28408,
"reps": [
1,
2
],
"peak": "118488527-118488827",
"signalValue": 12.67611,
"globalIDR": 0.5771
}
},
"both_replicates": true,
"NEW_in_this_beat": true
},
"ENCSR000EDY": {
"target": "BRCA1",
"cell": "HepG2",
"files": {
"ENCFF025YGD": {
"n": 300000,
"reps": [
1
],
"peak": "118488549-118488845",
"signalValue": 4.5543
},
"ENCFF383UGF": {
"n": 300000,
"reps": [
2
],
"peak": "118488402-118488706",
"signalValue": 6.0588
},
"ENCFF720ZDL": {
"n": 300000,
"reps": [
1,
2
],
"peak": "118488549-118488849",
"signalValue": 9.5283
},
"ENCFF047KJJ": {
"n": 52967,
"reps": [
1,
2
],
"globalIDR": 0.8875
},
"ENCFF989POH": {
"n": 51040,
"reps": [
1,
2
],
"globalIDR": 0.4816
}
},
"both_replicates": true,
"NEW_in_this_beat": true
},
"ENCSR000BRT": {
"target": "CBX3",
"cell": "K562",
"files": {
"ENCFF469CQE": {
"n": 300000,
"reps": [
1,
2
],
"peak": "118488488-118488918",
"signalValue": 15.8404
},
"ENCFF848QNF": {
"n": 300000,
"reps": [
1
],
"signalValue": 10.7715
},
"ENCFF725LHJ": {
"n": 300000,
"reps": [
2
],
"signalValue": 4.952
},
"ENCFF271ZAZ": {
"n": 133088,
"reps": [
1,
2
],
"globalIDR": 0.2266
},
"ENCFF844LFP": {
"n": 154990,
"reps": [
1,
2
],
"globalIDR": 0.4161
}
},
"both_replicates": true
},
"ENCSR000BUH": {
"target": "CBX3",
"cell": "HCT116",
"files": {
"ENCFF161HBO": {
"n": 101089,
"reps": [
2
],
"peak": "118488423-118488839",
"signalValue": 5.03375,
"globalIDR": 0.2438
},
"ENCFF991DCE": {
"n": 65155,
"reps": [
1
],
"hits": 0
},
"ENCFF093EQG": {
"n": 114452,
"reps": [
1,
2
],
"hits": 0
}
},
"both_replicates": false,
"note": "ONE replicate only in ENCODE's own pipeline, although ChIP-Atlas calls it at fold 12.50"
}
}negative and how deep — the numbers behind the chart
{
"ENCSR000DZS": {
"target": "BRCA1",
"cell": "GM12878",
"deepest_list": 300000,
"n_files": 7,
"pre_idr_300k_files": [
"ENCFF884PFA",
"ENCFF053BPC"
],
"hits": 0
},
"ENCSR223MLH": {
"target": "BRCA1",
"cell": "K562",
"deepest_list": 300000,
"n_files": 7,
"pre_idr_300k_files": [
"ENCFF748JQT",
"ENCFF686DEX",
"ENCFF673NII"
],
"hits": 0
},
"ENCSR946BXO": {
"target": "BRCA1",
"cell": "K562",
"deepest_list": 124863,
"n_files": 8,
"pre_idr_300k_files": [],
"hits": 0
},
"ENCSR343RJR": {
"target": "BRCA1",
"cell": "HepG2",
"deepest_list": 108511,
"n_files": 4,
"pre_idr_300k_files": [],
"hits": 0
},
"ENCSR000ATV": {
"target": "CBX3",
"cell": "K562",
"deepest_list": 300000,
"n_files": 4,
"pre_idr_300k_files": [
"ENCFF944XCG"
],
"hits": 0,
"note": "single-replicate experiment"
},
"ENCSR549NPZ": {
"target": "CBX3",
"cell": "GM12878",
"deepest_list": 63082,
"n_files": 8,
"pre_idr_300k_files": [],
"hits": 0
},
"ENCSR948QLZ": {
"target": "CBX1",
"cell": "K562",
"deepest_list": 300000,
"n_files": 13,
"pre_idr_300k_files": [
"ENCFF670HFC",
"ENCFF990SRA",
"ENCFF046QEQ"
],
"hits": 0
},
"ENCSR819WZE": {
"target": "CBX1",
"cell": "HepG2",
"deepest_list": 90702,
"n_files": 2,
"pre_idr_300k_files": [],
"hits": 0
}
}summary counts — the numbers behind the chart
{
"BRCA1_experiments_with_element_in_both_pre_IDR_replicates": 3,
"BRCA1_experiments_total": 7,
"CBX3_experiments_with_element_in_both_pre_IDR_replicates": 1,
"CBX3_experiments_with_element_in_one_replicate_only": 1,
"CBX3_experiments_total": 4,
"CBX1_experiments_with_element_anywhere": 0,
"CBX1_experiments_total": 2,
"experiments_where_a_300000_peak_pre_IDR_list_was_available_and_empty": 4
}concordance with signal level
{
"ENCSR000EDB_HeLa-S3_BRCA1": {
"peak": "both reps",
"signal_percentile": 93.69
},
"ENCSR000EBX_H1_BRCA1": {
"peak": "both reps",
"signal_percentile": 92.28
},
"ENCSR000EDY_HepG2_BRCA1": {
"peak": "both reps",
"signal_percentile": 55.08
},
"ENCSR343RJR_HepG2_BRCA1": {
"peak": "none",
"signal_percentile": 70.61
},
"ENCSR223MLH_K562_BRCA1": {
"peak": "none, 300k list",
"signal_percentile": 45.11
},
"ENCSR000DZS_GM12878_BRCA1": {
"peak": "none, 300k list",
"signal_percentile": 34.01
},
"discordant_pair": "the two HepG2 BRCA1 experiments invert - the peak-positive one is lower at signal level than the peak-negative one"
}n files per experiment — the numbers behind the chart
{
"ENCSR000EDB": 8,
"ENCSR000EBX": 13,
"ENCSR000EDY": 13,
"ENCSR343RJR": 8,
"ENCSR000DZS": 13,
"ENCSR223MLH": 13,
"ENCSR946BXO": 8,
"ENCSR000BRT": 13,
"ENCSR000BUH": 8,
"ENCSR000ATV": 4,
"ENCSR549NPZ": 8,
"ENCSR948QLZ": 13,
"ENCSR819WZE": 2
}- Is the H1 and HepG2 BRCA1 call reproducible in a fourth, independent BRCA1 experiment in either cell type? ENCODE holds only one BRCA1 experiment in H1 and two in HepG2, and the two HepG2 experiments disagree - one carries the element in both pre-IDR replicates and the other does not have it in a 108,511-peak list - so within-cell-type reproducibility is 1 of 2 where it can be tested at all.
- What global IDR value should a locus-specific query in this project accept? This census now provides the distribution to answer it: the element's global IDR across the five experiments that call it at all is 0.059, 0.227, 0.244, 0.482 and 0.577. `concept-a-peak-that-fails-idr-is-a-statement-about-replicate-rank-consistency-not-about-occupancy` posed this question and it is still open, but it is now answerable from data rather than convention.
- Does the same complete pre-IDR census at the AFF1, MLLT3 and MLLT1 breakpoint regions find the same pattern? The coordinates are in this graph and the 124 files are already identified; if BRCA1 sits on every recurrent translocation partner's breakpoint anchor this is a property of breakpoints, and if it sits only on KMT2A it is a property of KMT2A.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Across all 124 released GRCh38 peak files of the thirteen ENCODE BRCA1, CBX3 and CBX1 experiments, the KMT2A exon-11 element is present in both biological replicates before IDR thresholding in three BRCA1 experiments (HeLa-S3, H1, HepG2) and one CBX3 experiment (K562), in one replicate only in CBX3 HCT116, and in none of the eight remaining experiments - including four where a 300,000-peak pre-IDR list was available and empty.
What this evidence cannot show
A pre-IDR peak is a MACS2-style call before any reproducibility filter, so a 300,000-peak list is mostly noise by construction and presence in one is weak evidence on its own; the informative statements here are presence in BOTH replicates and the global IDR value, not presence in a permissive list. Three experiments (ENCSR343RJR, ENCSR946BXO, ENCSR819WZE) serve no 'peaks and background as input for IDR' product at all, so their zeros are bounded only at 90,702-124,863 peaks and are weaker absences than the four bounded at 300,000. The 147 bp query interval is a fixed window, not a peak-overlap test with a slop parameter, so a call whose interval stops just short would be missed - all recorded hits in fact span the whole interval.
Evidence that argues against this (1)
- Global IDR 0.577 (H1) and 0.482/0.888 (HepG2) are not near-misses in the way HeLa-S3's 0.0588 is. At those values the replicate ranks are close to independent, which is what IDR is designed to detect, and a reader could fairly say that only the HeLa-S3 call is a thresholding casualty while H1 and HepG2 are peaks IDR correctly rejected. This entry reports the values rather than arguing the point. Li Q et al. 2011
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
ENCODE serves, for many ChIP-seq experiments, a 'peaks and background as input for IDR' product truncated at 300,000 peaks per replicate, which makes an absence in that file a far tighter bound than an absence in a conservative-IDR file holding a few thousand peaks.
New developments on the Encyclopedia of DNA Elements (ENCODE) data portal. · Luo Y, Hitz BC, Gabdank I, et al. · Nucleic Acids Research 2020What this evidence cannot show
The 300,000 figure is the observed row count of these files, read this session, not a documented pipeline constant; every such file in this census had exactly 300,000 rows, which is consistent with a hard truncation but was not verified against the pipeline specification.
Evidence that argues against this (1)
- A truncated top-300,000 list is still a ranked list, so an element could in principle fall at rank 300,001. With ~3.1 Gb of genome and 300,000 peaks the list already covers roughly one call every 10 kb, so the residual risk is small but not zero, and no entry should read this as an exhaustive search of the reads. ENCODE Project Consortium 2012
the complete 124-file pre-IDR census is the measurement the pre-registered experiment specified, run at peak level across all 13 experiments
Q37 reached two experiments through ChIP-Atlas and ENCODE pre-IDR files and found the element in both; this census opened every released GRCh38 peak file of all 13 and reproduces both values (HeLa-S3 0.0589, CBX3 K562 0.2266)
124 files across 13 experiments, in which the element is present in pre-IDR and absent from the IDR-thresholded products of the same experiments
the per-experiment table this cell-type restriction is read from
the peak-level census the signal-level reading is interpreted against