KDM5A / JARID1A / RBP2 (P29375, 1,690 aa) - already in this graph as the 3' half of NUP98::KDM5A and never written as a gene: the H3K4 demethylase CTCF recruits to a break, the only ERASER in the licensing cascade, and the one cascade member with no blood-lineage ChIP-seq anywhere
also called KDM5A · JARID1A · RBBP2 · RBP2 · P29375 · KDM5A_HUMAN · Lysine-specific demethylase 5A · Histone demethylase JARID1A · Retinoblastoma-binding protein 2 · [histone H3]-trimethyl-L-lysine(4) demethylase 5A
KDM5A strips an activating chemical tag off DNA packaging near a break, and a 2026 paper shows that stripping is part of what lets the repair machinery arrive. This graph already treats KDM5A as a leukaemia fusion partner, so the same protein now sits on both sides of the story: a cause of one leukaemia and a repair licence in another.
The same thing, for a scientist
**Identity.** P29375, `KDM5A_HUMAN`, **1,690 aa**, primary gene name KDM5A, UniProt gene synonyms **JARID1A, RBBP2 and RBP2**, 46 PDB entries. JmjN 19-60, ARID 84-174, JmjC 437-603, PHD-type zinc fingers at 293-343, 1161-1218 and 1607-1661, a C5HC2 zinc finger 676-728, and a region *'Interaction with LMO2'* at 1623-1690 [E1]. One reviewed record, `gene_primary` reading back KDM5A. **It is the ERASER limb of the cascade, and the only one.** CTCF recruits HP1-gamma, KAP1, SUV39H1, SETDB1 **and KDM5A** to breaks, and the chromatin change the paper reports is symmetrical: *'loss of H3K9 methylation / H3K4 demethylation at DSBs'* on CTCF depletion. So the licence for BRCA1/BARD1 is not simply 'add H3K9me' - it is **add H3K9me and remove H3K4me at the same place**, and KDM5A is the enzyme doing the removing. The specificity was tested against the nearest paralogue: the paper depleted **KDM5A and KDM5B separately** and also raised antibodies against KDM5C, and it demonstrates a KDM5A-gammaH2AX proximity-ligation signal on micro-irradiated strips. Its KDM5 inhibitor is **CPI-455** [E2]. **Why that is more than a detail for this graph.** `mark-h3k4me3` and the WRAD/SET machinery are the centre of this project - KMT2A is an H3K4 methyltransferase. A cascade in which **H3K4 demethylation licenses repair** puts the mark this disease is about on the opposite side of a DSB decision from where the graph usually places it. And it makes a second prediction nobody has tested: a fusion protein that deposits H3K4me3 at its targets is, at those sites, doing the opposite of what this cascade needs. **And KDM5A is already here under another name.** `fusion-nup98-kdm5a` is a canonical id in this graph. Until now the gene had no entry of its own, so there was nowhere to record that the same protein is a DSB-repair licence factor. `KDM5A[tiab] AND KMT2A[tiab]` returns **14** records and `KDM5A[tiab] AND MLL[tiab]` returns 6 - a real literature, and title-checking is needed before any of it is read as being about the *KMT2A locus*, which none of it is. **Dependency: shallow.** Mean **-0.1185** over **1,258** screens, 6.4% below -0.5, none below -1.0; blood mean -0.1661 with 11.6% below -0.5 [E3]. **Archive: the one cascade member with no blood-lineage data.** ENCODE holds **3** KDM5A ChIP-seq experiments - A549, H1 and HepG2 - and **no blood line** [E4]. BRCA1 and CBX3 both have K562 and GM12878; KDM5A has neither. So for KDM5A the locus question in a blood cell is `never-measured` while for BRCA1 and CBX3 it is `measured-not-analysed`.
No three-dimensional model is available for this protein, so the map above is the whole picture. That is itself informative: AlphaFold does not model chains this long, and the parts that have been crystallised are held under the entries for the individual domains.
The demethylation limb is the half of this cascade that the graph should find most uncomfortable
Every existing argument in this graph about the CTCF/H3K9 axis is about adding a repressive mark. 2's phenotype is stated as a pair: CTCF depletion causes loss of H3K9 methylation AND loss of H3K4 demethylation at DSBs. Read carefully, the second half means H3K4 methylation is retained when the cascade fails, and retained H3K4me is part of what stops BRCA1/BARD1 loading.
Two consequences:
- An H3K9-only intervention is not the only way to break this licence. Blocking KDM5A does it too. The paper's reagent for that is CPI-455, and the graph currently has no entry for it.
- KMT2A is the opposing enzyme. KMT2A and its fusion products deposit H3K4me3. At a fusion-bound target promoter, the fusion is maintaining exactly the mark this cascade needs removed for HR. Nobody has tested whether a KMT2A fusion's H3K4me3 deposition impairs HR at the loci it occupies, and this graph's one controlled test of fusion effects on DSB repair -
concept-neither-t411-product-changes-dna-damage-repair- used genome-wide reporters in HEK293, which would not detect a locus-restricted effect at fusion-bound sites.
That second point is the sharpest new question this beat produces, and it is cheap: the DR-GFP reporter can be placed at a fusion-bound locus.
Two KDM5As: the fusion partner and the repair factor
This graph already carries fusion-nup98-kdm5a as a canonical id. That entry is about a chimeric protein in paediatric AML in which KDM5A contributes its PHD-type zinc fingers. This entry is about wild-type KDM5A as a demethylase recruited to double-strand breaks.
They are not the same claim and nothing here should be read as transferring between them. But the coincidence is worth recording, because it means a single protein appears in this graph as (a) a driver of one paediatric leukaemia and (b) a licence factor for accurate repair of the breaks that create another. No published work connects the two readings, and KDM5A[tiab] AND 11q23[tiab] returns 0.
| uniprot | P29375 |
|---|---|
| uniprot entry name | KDM5A_HUMAN |
| gene symbol | KDM5A |
| gene primary readback | KDM5A |
| protein name | Lysine-specific demethylase 5A |
| length AA | 1690 |
| n PDB entries | 46 |
| how the accession was resolved | gene_exact:KDM5A AND organism_id:9606 AND reviewed:true, fetched live 2026-09-22, returned EXACTLY ONE reviewed human record with gene_primary KDM5A; length asserted at 1,690 aa. No PDB polymer entity was fetched, so nothing is claimed about what the 46 deposits cover. |
| localization | Nucleus, nucleolus [ECO:0000269, PubMed:15949438 and 7935440] - the coded location is NUCLEOLAR, which is not what a DSB-repair factor's annotation usually looks like and is worth noting beside the cascade role. A second Nucleus assignment is ECO:0000250, i.e. inferred by similarity from mouse Q3UXZ9, not experimental. Rule 91: the two experimental citations are a 1994 and a 2005 paper, neither of which is about DNA damage. |
| subnuclear localization | NUCLEOLUS is the only coded subnuclear compartment on the UniProt record, with an uncoded note that KDM5A 'occupies promoters of genes involved in RNA metabolism and mitochondrial function'. The compartment this beat cares about - micro-irradiated chromatin at a break - comes only from [E2], which shows a KDM5A/gammaH2AX proximity-ligation signal on laser-induced strips in U2OS. Nothing places KDM5A at the KMT2A locus in any cell type, and the nucleolar annotation is a reminder that the resting-state address and the damage-state address are different questions. |
| localization shown for fusion | false |
| cell type | U2OS and HEK-293T for the cascade [E2]. ENCODE: A549 lung adenocarcinoma, H1 embryonic stem cells and HepG2 hepatoblastoma - NO blood-lineage cell [E4]. 1,258 DepMap lines including 147 haematopoietic and lymphoid [E3]. As a fusion partner KDM5A is established in paediatric AML through NUP98::KDM5A, which this graph already holds - a different disease context from the repair role and one where the relevant protein is a chimera, not wild-type KDM5A. |
| cell state | DNA DAMAGE is the state in which the cascade role appears - the KDM5A/gammaH2AX proximity signal is at micro-irradiated strips 5 min after damage [E2]. The resting-state address is nucleolar and promoter-associated [E1], which is a different compartment. As with every member of this cascade, occupancy at any locus is a function of damage state, and no damage-state measurement exists at the KMT2A locus in any cell type. |
| epigenetic readout | ERASER of H3K4me3/me2 (the systematic name is [histone H3]-trimethyl-L-lysine(4) demethylase 5A), and simultaneously a READER through three PHD-type zinc fingers. In this cascade it is the demethylase limb: CTCF depletion causes 'loss of H3K9 methylation / H3K4 DEMETHYLATION at DSBs', i.e. H3K4 methylation is inappropriately RETAINED when the cascade fails, and BRCA1/BARD1 does not load [E2]. This graph's central protein KMT2A is the opposing enzyme - an H3K4 methyltransferase - so the cascade puts KMT2A's own catalytic output on the side that must be removed for accurate repair. |
| mark erased | H3K4me3, H3K4me2 |
| eraser of that mark | mark-h3k4me3 |
| shown for fusion | YES but for a DIFFERENT function and a DIFFERENT fusion: KDM5A is the 3' partner of NUP98::KDM5A in paediatric AML, which this graph holds as `fusion-nup98-kdm5a`. NOTHING about the DSB/HR-licensing role has been shown in any fusion. In KMT2A-rearranged disease specifically: KDM5A[tiab] AND KMT2A[tiab] returns 14 records and KDM5A[tiab] AND MLL[tiab] returns 6, but KDM5A[tiab] AND 11q23[tiab] returns 0 with a non-empty querytranslation and 'No items found.', and no record places KDM5A at the KMT2A locus. |
| archive bound 2026 09 22 | ENCODE released TF ChIP-seq for KDM5A: 3 experiments, in A549, H1 and HepG2. NO blood-lineage biosample. This is the only member of the licensing cascade with released occupancy data in which no blood cell appears, which is why its absence class differs from BRCA1's and CBX3's. |
| absence class | never-measured |
| absence class reasoning | In a BLOOD cell the question cannot be asked of the archive at all - ENCODE's three KDM5A experiments are lung, embryonic stem and liver. Strictly, the class is measured-not-analysed for non-haematopoietic cells (three datasets exist and were never read at the KMT2A locus) and never-measured for the cell types this disease is about. The stricter reading is recorded as the class because the graph's question is about a haematopoietic cell. |
gene synonyms in uniprot
[ "JARID1A", "RBBP2", "RBP2" ]
identifier traps
{
"registry_row_that_already_exists": "The LMO2 row of data/symbol-collisions.json names KDM5A: it records that PMID 9129143, the only published LMO2-KDM5A interaction, publishes as 'rhombotin-2' and 'RBP2' and is unreachable from a title search for either current symbol. That row was found BEFORE this beat's first query, as rule 88 requires, and it is the reason the retired symbols were searched here.",
"retired_symbols_carry_the_older_literature": "KDM5A publishes as RBP2 and JARID1A across the whole pre-2010 retinoblastoma literature. Measured 2026-09-22: RBP2[tiab] AND MLL[tiab] returns 0 with a non-empty querytranslation and 'No items found.', while KDM5A[tiab] AND MLL[tiab] returns 6 - so on THIS pair the retired symbol adds nothing, and that is a measured result rather than an assumption.",
"the_paralogue_that_can_be_swapped": "KDM5B is the near neighbour and the source paper deliberately separates them with independent siRNAs; KDM5A 1,690 aa and KDM5B 1,544 aa are far enough apart that the length check discriminates, unlike the CBX family.",
"a_uniprot_feature_that_points_at_this_project": "the feature table annotates 'Interaction with LMO2' at 1623-1690, and LMO2 is a T-ALL oncogene already in this graph's orbit."
}domains — the numbers behind the chart
[
{
"name": "JmjN",
"start": 19,
"end": 60
},
{
"name": "ARID",
"start": 84,
"end": 174
},
{
"name": "PHD-type zinc finger 1",
"start": 293,
"end": 343
},
{
"name": "JmjC",
"start": 437,
"end": 603
},
{
"name": "C5HC2 zinc finger",
"start": 676,
"end": 728
},
{
"name": "PHD-type zinc finger 2",
"start": 1161,
"end": 1218
},
{
"name": "PHD-type zinc finger 3",
"start": 1607,
"end": 1661
},
{
"name": "Interaction with LMO2",
"start": 1623,
"end": 1690
}
]depmap
{
"n_screens_with_gene_effect": 1258,
"mean_gene_effect": -0.1185,
"median_gene_effect": -0.1195,
"min_gene_effect": -0.9972,
"max_gene_effect": 0.7462,
"frac_below_minus_0_5": 0.0644,
"frac_below_minus_1_0": 0,
"n_blood_screens": 147,
"blood_tissue_blocks": [
"hematopoietic tissue",
"lymphoid tissue"
],
"blood_mean_gene_effect": -0.1661,
"blood_frac_below_minus_0_5": 0.1156,
"n_tissue_blocks_flattened": 29,
"open_targets_data_version": {
"year": "26",
"month": "06",
"iteration": null
},
"depmap_release": null,
"rule_83": "no DepMap release identifier in the payload; the Open Targets data version is recorded and is not a substitute.",
"rule_84": "no median-split proportion is reported.",
"live_zero_screen_control": "POLR2A in the same call returned 0 tissue blocks and 0 screens with a passing approvedSymbol readback.",
"reading": "shallow, marginally deeper in blood than in the panel (-0.166 vs -0.119, 11.6% vs 6.4% below -0.5). KDM5B is a paralogue with overlapping substrate, so redundancy is the obvious explanation for a shallow score and the figure cannot discriminate dispensability from buffering."
}- CPI-455, the tool used for the demethylase limb, is a pan-KDM5 inhibitor, so any therapeutic reasoning about 'blocking KDM5A' in this cascade would in practice also block KDM5B and KDM5C - and KDM5B is the paralogue the source paper had to control against.
- If H3K4 demethylation licenses HR, then KDM5 inhibition during genotoxic therapy carries the same liability as H3K9 methyltransferase inhibition, and this graph has not previously counted KDM5 inhibitors as a repair-relevant class. Several are in clinical development for other indications.
- The opposite reading is also published - H3K4 methylation by SETD1A/BOD1L promotes RIF1-dependent NHEJ - so blocking the demethylase may not abolish repair but redirect it to end joining. At a translocation-prone locus, redirecting repair toward end joining is the worse of the two outcomes, not the milder one.
- Does a KMT2A fusion's own H3K4me3 deposition impair HR at the loci it occupies? The cascade requires H3K4 DEMETHYLATION for BRCA1/BARD1 loading, and a fusion that maintains H3K4me3 at its targets is holding the opposite state there. `concept-neither-t411-product-changes-dna-damage-repair` used genome-wide reporters in HEK293 and could not have seen a locus-restricted effect; a DR-GFP reporter integrated at a fusion-bound locus would.
- Is KDM5A present at the KMT2A exon-11 CTCF element (GRCh38 chr11:118,488,159-118,489,206) in any cell? ENCODE's three datasets are A549, H1 and HepG2 - so the resting-state question can be asked in three non-blood cells today and in no blood cell at all.
- Does ChIP-Atlas or GEO hold a KDM5A ChIP-seq in a haematopoietic line? Running that one query would move this absence from never-measured to measured-not-analysed, and it was not run in this beat.
- Is the DSB role of KDM5A retained in the NUP98::KDM5A chimera, which keeps KDM5A's PHD fingers? Nothing connects the two literatures and a single co-immunoprecipitation would start to.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
KDM5A is UniProt P29375, entry name KDM5A_HUMAN, 1,690 amino acids, primary gene name KDM5A with gene synonyms JARID1A, RBBP2 and RBP2, recommended protein name 'Lysine-specific demethylase 5A' and alternative names including 'Retinoblastoma-binding protein 2' and '[histone H3]-trimethyl-L-lysine(4) demethylase 5A'; the feature table annotates JmjN 19-60, ARID 84-174, JmjC 437-603, PHD-type zinc fingers at 293-343, 1161-1218 and 1607-1661, a C5HC2 zinc finger 676-728, and a region 'Interaction with LMO2' at 1623-1690; it cross-references 46 PDB entries; and its coded subcellular location is Nucleus, nucleolus (ECO:0000269, PubMed:15949438 and 7935440) with a second Nucleus assignment coded only ECO:0000250 by similarity to mouse Q3UXZ9, plus an uncoded note that it occupies promoters of genes involved in RNA metabolism and mitochondrial function.
UniProtKB P29375 (KDM5A_HUMAN) - Lysine-specific demethylase 5A, Homo sapiens · UniProt Consortium · UniProt Knowledgebase 2026What this evidence cannot show
A database record fetched 2026-09-22 with gene_exact:KDM5A AND organism_id:9606 AND reviewed:true returning exactly one reviewed record. Rule 91: the two ECO:0000269 citations behind the nucleolar call are from 1994 and 2005 and neither is a DNA-damage study, so the coded localisation is not about this beat's question. The ECO:0000250 Nucleus assignment is a curator inference from mouse, not a human measurement. No PDB polymer entity was fetched.
Evidence that argues against this (1)
- UniProt's only coded compartment for KDM5A is the nucleolus, and the cascade role requires it at euchromatic break sites. Either the database's compartment annotation is stale relative to the damage literature, or the damage-recruited pool is a minor fraction that a steady-state localisation study would miss - and nothing in the record distinguishes those. UniProt Consortium 2026
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
KDM5A is the H3K4-demethylase limb of the CTCF-directed HR licensing cascade. In U2OS and HEK-293T cells, CTCF recruits HP1-gamma, KAP1, SUV39H1, SETDB1 and KDM5A to DSBs, enabling rapid BRCA1/BARD1 accumulation; CTCF depletion abrogates their enrichment with loss of H3K9 methylation AND loss of H3K4 demethylation at DSBs, and defective rapid BRCA1/BARD1 recruitment; deregulating H3K9 methylation or H3K4 demethylation compromises the immediate accumulation of the other repressive proteins and of BRCA1/BARD1; and EXO1/DNA2 recruitment, end resection and HR fail as a consequence. Specificity for KDM5A over its paralogues was addressed directly: siKDM5A and siKDM5B cells were compared side by side after laser micro-irradiation, antibodies against KDM5A, KDM5B and KDM5C were all used, a KDM5A-gammaH2AX proximity-ligation signal was detected on micro-irradiated strips in more than 30 cells per experiment across at least three independent experiments, and CTCF co-immunoprecipitation with or without etoposide was probed for the cascade members. The KDM5-family inhibitor used is CPI-455.
CTCF primes repressive chromatin for rapid BRCA1/BARD1 recruitment to drive homologous recombination repair of DNA double-strand breaks. · Hwang SY, Choi D, Choi MJ, Jo HR, Jeong JH, Kang MA, Lee JS · Nucleic Acids Research 2026What this evidence cannot show
U2OS osteosarcoma reporter lines (AsiSI-ER, LacI-FokI-DD, DR-GFP, EJ5-GFP) and HEK-293T only; no haematopoietic cell, no fusion, no KMT2A locus. CPI-455 is a pan-KDM5 inhibitor, not KDM5A-selective, so the pharmacological arm cannot separate KDM5A from KDM5B or KDM5C - only the siRNA arm can, and siRNA specificity rests on the paper's own controls. Proximity ligation demonstrates proximity, not a direct contact. DOI taken from PubmedData/ArticleIdList only; the loose .//ArticleId path on this record returns 191 id elements, most from the reference list.
Evidence that argues against this (1)
- The opposite sign for H3K4 methylation at a break is also on record in this literature: H3K4 methylation by SETD1A/BOD1L facilitates RIF1-dependent NHEJ. If H3K4me promotes end joining and its removal promotes HR, then KDM5A is a pathway-choice factor rather than an HR licence per se, and an intervention that blocks it would shift repair toward NHEJ rather than simply failing - a materially different outcome for a translocation-prone locus. Bayley R et al. 2022
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
KDM5A is a shallow CRISPR dependency: flattening every tissue block gives 1,258 screens with a gene effect, mean -0.1185, median -0.1195, minimum -0.9972, maximum 0.7462, with 6.44% of lines below -0.5 and none below -1.0; in the 147 hematopoietic and lymphoid screens the mean is -0.1661 with 11.6% below -0.5.
Open Targets Platform GraphQL API, target.depMapEssentiality with all tissue blocks flattened, plus meta.dataVersion · Open Targets Platform; Broad Institute DepMap · Open Targets Platform 2026What this evidence cannot show
Rule 83: no DepMap release identifier in the payload; Open Targets data version year 26, month 06. Rule 84: no median-split proportion is reported. The live zero-screen control fired - POLR2A returned 0 blocks and 0 screens in the same call. KDM5B and KDM5C share substrate specificity with KDM5A, so a shallow single-gene score is the expected result for a redundant family and does not measure requirement.
Evidence that argues against this (1)
- A DepMap score for a demethylase says nothing about its role under genotoxic stress, because the screens carry no exogenous DNA-damaging agent. For a gene whose only claim here is damage-conditional, the unstressed screen is the least relevant condition available. Open Targets Platform; Broad Institute DepMap 2026
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
KDM5A is the only member of this licensing cascade with released ENCODE occupancy data and no blood-lineage biosample: ENCODE holds 3 released TF ChIP-seq experiments for target label KDM5A, in A549, H1 and HepG2. In the same census BRCA1 has 7 including K562 and GM12878, CBX3 has 4 including K562 and GM12878, CBX1 2 including K562, CBX5 4 including K562 and GM12878, TRIM28 5 including K562, SETDB1 5 including K562, CTCF 502, and BARD1, EXO1, DNA2, SUV39H1 and RBBP8 have no ENCODE target record at all.
ENCODE Portal experiment search, type=Experiment, assay_title=TF ChIP-seq, status=released, target.label=KDM5A · ENCODE Project Consortium · ENCODE Portal 2026What this evidence cannot show
Counted 2026-09-22 on the released TF ChIP-seq facet only; GEO individual depositions, ChIP-Atlas and ReMap were NOT swept, so a KDM5A ChIP-seq in a blood cell could exist outside ENCODE and would change the absence class. All three experiments are in undamaged cultured cells, so even a re-analysis at the KMT2A locus would test only the resting state of a damage-conditional function.
Evidence that argues against this (1)
- Bounding a cell-type gap on one consortium is the error this project has already made with GEO: three independent query forms returned false negatives for a SuperSeries that in fact had 79 SRA experiments. A ChIP-Atlas query, which aggregates GEO, was not run here and is the cheap way to overturn this. ChIP-Atlas 2026
- E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
KDM5A has a real KMT2A-adjacent literature that is not about the KMT2A locus, and the retired-symbol route adds nothing on this pair. Measured live 2026-09-22 with one [tiab] tag per term and every querytranslation read back: KDM5A[tiab] AND KMT2A[tiab] returns 14 records; KDM5A[tiab] AND MLL[tiab] returns 6; KDM5A[tiab] AND 11q23[tiab] returns 0 with the translation '"KDM5A"[Title/Abstract] AND "11q23"[Title/Abstract]' and warninglist 'No items found.'; and the retired symbol RBP2[tiab] AND MLL[tiab] returns 0, also with a non-empty translation and 'No items found.'
NCBI E-utilities esearch over PubMed, db=pubmed, querytranslation read back for every query · National Center for Biotechnology Information · NCBI Entrez Programming Utilities 2026What this evidence cannot show
Titles and abstracts only - rule 80 applies and a KDM5A track over KMT2A in a supporting figure would be invisible to all four queries. The 14 and 6 counts were NOT individually title-checked in this beat beyond confirming that none is a KMT2A-locus localisation study; most are expected to concern NUP98::KDM5A, which is a different question. Rule-67 positive controls through the identical SYMBOL[tiab] AND TOKEN[tiab] shape: CTCF[tiab] AND KMT2A[tiab] = 7 and RAD21[tiab] AND KMT2A[tiab] = 5, so the shape returns ones. Note that TOP2B[tiab] AND KMT2A[tiab] = 0 in the same shape although that literature plainly exists under other names - the instrument works and the vocabulary is the binding constraint.
Evidence that argues against this (1)
- The TOP2B control is the warning attached to every count in this entry: a symbol-pair query returned a confident zero for a relationship that is central to this disease, because the literature writes 'MLL' and 'topoisomerase II' rather than the current symbols. Any conclusion drawn from KDM5A[tiab] AND 11q23[tiab] = 0 inherits that limitation. National Center for Biotechnology Information 2026
the source of the transforming PHD finger, whose catalytic domain is lost in the fusion
CTCF recruits KDM5A to DSBs; siKDM5A separated from siKDM5B
erases H3K4me3/me2; H3K4 demethylation at the break is required for BRCA1/BARD1 loading
3' partner of the paediatric AML fusion already held in this graph
KDM5A is the eraser limb - H3K4 demethylation is half the licence
never-measured in a blood cell - ENCODE's three datasets are A549, H1 and HepG2
unchanged this cycle: 3 ENCODE experiments, none in a blood cell, not run here
KDM5A ChIP-seq exists in NB-4 (APL) and MM.1S myeloma, so 'never-measured in a blood cell' is wrong; and it is negative at the KMT2A breakpoint region in both