ARID1B (BAF250b) - the cBAF paralogue whose deletion PROMOTES MLL-AF9 leukaemogenesis, which makes it the one gene in this beat whose published KMT2A-fusion result points the wrong way for a drug target

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.

• ARID1B deletion PROMOTES MLL-AF9 leukaemogenesis in mouse. If that sign holds in KMT2A::AFF1, then any agent that reduces ARID1B function - including a non-selective cBAF-directed agent - would be expected to accelerate the disease, and it would do so through an on-target mechanism that a cytotoxicity screen would not surface.
• ARID1A/ARID1B double loss causes acute, fatal bone-marrow failure in mice from complete loss of BAF remodelling. A paralogue-directed ARID1B agent given to a patient whose blasts or normal progenitors carry an ARID1A lesion has a named, lethal, haematopoietic on-target toxicity, and the dose-limiting organ is the one the disease is in.
• The published authors of the dual-loss work explicitly advise caution against paralogue-directed therapies for ARID1-mutant cancer, because residual ARID1-less cBAF subcomplexes go on to disrupt pBAF. Partial inhibition of a scaffold subunit does not give partial loss of function - it gives a different, actively disruptive complex, which is a general warning for every scaffold subunit this graph reasons about, not just this one.

1. Does ARID1B's tumour-suppressor sign in mouse MLL-AF9 hold in human KMT2A::AFF1? If it does, cBAF is not this fusion's dependency and hyp-ncbaf-not-canonical-baf-is-the-kmt2a-dependency gains a clean subtractive argument. If it inverts - as Arid2's sign does between disease stages in the same paper - the whole ARID-subunit picture in KMT2A leukaemia is stage- and lineage-contingent and cannot support any therapeutic inference.
2. Is ARID1B a differential dependency in KMT2A-rearranged lines in DepMap? The query is public and cheap, was NOT run in this beat, and would need its release stamped per rule 83. It is the single most informative next measurement on this gene.
3. Paralogue-selective peptidic ligands for ARID1B were reported in 2024 (PMID 38655884, seen at title level only). Given that ARID1B loss PROMOTES MLL-AF9 leukaemia and ARID1A/ARID1B double loss causes fatal bone-marrow failure in mice, is there any scenario in which an ARID1B-selective agent is desirable in this disease, or is the correct conclusion that the chemical matter exists and should not be used here?
4. ARID1B is >90% intrinsically disordered with one 486-residue disordered stretch. Does it meet this graph's own localization-condensate-evidence-standards? Nobody has asked, and BAF-subunit condensate behaviour is an active literature elsewhere.
5. Should the symbol-collision registry carry an ARID1B row for SMARCF2? The alias resembles the SMARCA/B/C/D/E family names while naming a gene that is not in them, which is the exact shape of the MLL2/MLL4 trap rule 66 records.

1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

   ARID1B is Q8NFD5 (ARI1B_HUMAN), 2,319 aa, with an ARID domain at 1136-1227, LXXLL motifs at 502-506 and 2119-2123, a nuclear localisation signal at 1441-1460 and fourteen annotated disordered regions including one of 486 residues at 577-1062; it is nuclear, widely expressed with high levels in heart, skeletal muscle and kidney, is a component of SWI/SNF chromatin-remodelling complexes and of the neural-progenitor npBAF and neuron-specific nBAF complexes, binds DNA non-specifically, and germline loss causes Coffin-Siris syndrome 1.

   UniProtKB Q8NFD5 (ARI1B_HUMAN), AT-rich interactive domain-containing protein 1B · UniProt Consortium · UniProt 2026
   ▸What this evidence cannot show

   Fetched live 2026-09-23 via gene_exact:ARID1B AND organism_id:9606 AND reviewed:true, which returned exactly ONE reviewed record, so rule 62 does not fire. data/symbol-collisions.json (208 rows) word-boundary grepped for ARID1B before the first query: no row - although this gene has the widest alias surface in the beat (BAF250b, ELD/OSA1, OSA2, p250R, SMARCF2, DAN15, 6A3-5, KIAA1235) and SMARCF2 in particular is a name that invites confusion with the SMARC gene family it is not a member of. The npBAF/nBAF descriptions carry UniProt's 'By similarity' flag. Per rule 91 the papers UniProt cites were not opened. Note this entry is a GENE entry and deliberately carries no complex-level accession; rule 69b's prohibition on a bare uniprot field applies to complexes, and ARID1B is a subunit, so its own accession is correct here.

   ▸Evidence that argues against this (1)

   • UniProt's FUNCTION block for ARID1B is written almost entirely about the neural npBAF-to-nBAF switch, and its DISEASE section names only Coffin-Siris syndrome. The haematopoietic and leukaemic biology that makes this gene relevant here - an MLL-AF9 tumour-suppressor phenotype and an ARID1A synthetic-lethal relationship - is absent from the record entirely. Leukemia 2022
2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

   HGNC assigns ARID1B (HGNC:18040, 'AT-rich interaction domain 1B') the primary Ensembl gene id ENSG00000049618, cytoband 6q25.3 and UniProt Q8NFD5, with alias symbols KIAA1235, ELD/OSA1, p250R, BAF250b, DAN15, 6A3-5 and SMARCF2 and no previous symbol; Ensembl places the gene at GRCh37 chr6:157,099,063-157,531,913 and GRCh38 chr6:156,776,020-157,211,460, both on the plus strand.

   HGNC record HGNC:18040 (ARID1B) and Ensembl gene ENSG00000049618 on GRCh37 and GRCh38 · HUGO Gene Nomenclature Committee; Ensembl · genenames.org / Ensembl 2026
   ▸What this evidence cannot show

   Fetched live 2026-09-23 and resolved before any sentence using them was written. No previous official symbol, so rule 69's retired-symbol undercount does not apply; the alias arm BAF250b was run anyway (E7) and adds almost nothing. The alias SMARCF2 is a nomenclature hazard worth noting: it looks like a member of the SMARC families (SMARCA/B/C/D/E) and is not one.

   ▸Evidence that argues against this (1)

   • HGNC's alias list for ARID1B contains SMARCF2, a name that implies membership of the SMARCA/B/C/D/E families this gene is not in, and ELD/OSA1, which contains a slash that is query syntax in several search interfaces. An alias list is a set of strings with different failure modes, not a set of equivalent names, and this one has both a semantic trap and a syntactic one in it. HUGO Gene Nomenclature Committee 2026
3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   In conditional knockout mice, deleting Arid1b in either the initiation or the maintenance phase PROMOTES MLL-AF9 leukaemogenesis, whereas its PBAF counterpart Arid2 has dual roles - enhancing leukaemogenesis when deleted at initiation yet being required during maintenance; integrated transcriptomic and genomic binding analyses show Arid2 and Arid1b regulate largely distinct gene sets at different disease stages, converging on Etv4/Etv5 in an opposing manner while separately regulating Gata2, Tcf4, Six4, Irf4 and Hmgn3.

   Differential roles of BAF and PBAF subunits, Arid1b and Arid2, in MLL-AF9 leukemogenesis · Bluemn T, Schmitz J, Zheng Y, Burns R, Zheng S, DeJong J, et al. · Leukemia 2022
   PMID 35022500DOI
   ▸What this evidence cannot show

   Abstract-level read; the full text was not opened this session. MOUSE, MYELOID, MLL-AF9 (KMT2A::MLLT3) - not human, not B-lineage, not KMT2A::AFF1; the system is named in the claim itself per rule 73. 'Promotes leukemogenesis' is the abstract's wording and the magnitude, the latency and the penetrance are not recoverable from it. The paper's own closing emphasis is that cellular context and disease stage determine these subunits' functions, so a single sign should not be generalised across stages or lineages.

   ▸Evidence that argues against this (1)

   • The same paper shows Arid2 behaving in OPPOSITE directions at initiation and maintenance of the same leukaemia. A subunit whose sign flips with disease stage cannot support a stable therapeutic inference, and nothing establishes that Arid1b's sign is stable in a lineage and fusion the paper did not test. Leukemia 2022
4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   ARID1B deficiency is synthetically lethal with ARID1A mutation; ATAC-seq, genome-wide histone-modification mapping and expression analysis in colorectal cancer cells lacking one or both ARID proteins show that ARID1A has a dominant role in maintaining chromatin accessibility at enhancers while ARID1B's contribution is evident only in the context of ARID1A mutation, where accessibility changes predict expression changes and correlate with loss of H3K4me and H3K27ac, altered nucleosome spacing and altered transcription-factor binding, particularly at growth-pathway genes including MET; ARID1B knockdown in ARID1A-mutant ovarian cancer cells causes a similar loss of enhancer architecture.

   Chromatin accessibility underlies synthetic lethality of SWI/SNF subunits in ARID1A-mutant cancers · Kelso TWR, Porter DK, Amaral ML, Shokhirev MN, Benner C, Hargreaves DC · Elife 2017
   PMID 28967863DOI
   ▸What this evidence cannot show

   Abstract-level read; the full text was not opened this session. SOLID TUMOUR cell lines (colorectal, ovarian) - there is no haematopoietic or leukaemic arm. The synthetic lethality is conditional on ARID1A MUTATION, and ARID1A mutation is not a feature of KMT2A-rearranged infant ALL, which has among the lowest mutational burdens of any leukaemia - so the context that makes ARID1B a target in solid tumours is absent in this disease.

   ▸Evidence that argues against this (1)

   • In haematopoiesis the analogous double loss is not a therapeutic window but a lethal one: ARID1A/ARID1B double knockout mice succumb to acute bone-marrow failure from complete loss of BAF remodelling activity. The same synthetic-lethal relationship that is exploitable in an ARID1A-mutant epithelial tumour is catastrophic in the blood compartment. Blood Adv 2023
5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   In a mouse model with Arid1b deleted specifically in the haematopoietic compartment, Arid1b knockout mice show only a modest effect on steady-state haematopoiesis - unlike the extensive phenotype of ARID1A deficiency - but transplantation experiments show that reconstitution of myeloid cells in irradiated recipients is ARID1B-dependent; mice lacking both ARID1A and ARID1B in the haematopoietic compartment succumb to acute bone-marrow failure from complete loss of BAF-mediated chromatin-remodelling activity; and ATAC-seq shows that more than 80% of loci regulated by ARID1B are distinct from those regulated by ARID1A, with ARID1B controlling expression of genes crucial for myelopoiesis.

   Dissecting the role of SWI/SNF component ARID1B in steady-state hematopoiesis · Madan V, Shyamsunder P, Dakle P, Woon TW, Han L, Cao Z, et al. · Blood Adv 2023
   PMID 37611161DOI
   ▸What this evidence cannot show

   Abstract-level read; the full text was not opened this session. MOUSE, and the readouts are steady-state and post-transplant haematopoiesis, not leukaemia. 'More than 80% distinct loci' is an ATAC-seq overlap statistic whose denominator and thresholds are not recoverable from the abstract. The myeloid emphasis means the lymphoid - and specifically B-lineage - consequences of Arid1b loss are not reported here.

   ▸Evidence that argues against this (1)

   • A separate report attributes to ARID1B a role in the REGENERATIVE ability of haematopoietic stem cells in normal haematopoiesis, which is a stronger claim than 'modest effect at steady state'. The two are compatible - a regeneration-specific requirement is invisible at steady state - but they would support different statements about what losing ARID1B costs, and only one of them was read here. PLoS One 2024
6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   In double-knockout liver and skin, aggressive carcinogenesis follows de-differentiation and hyperproliferation, and in double-mutant endometrial cancer adding back either ARID1 protein induces senescence; biochemically, the residual cBAF subcomplexes resulting from loss of ARID1 scaffolding disrupt polybromo-containing pBAF function; 37 of 69 mutations observed in the conserved scaffolding domains of ARID1 proteins in human cancer cause complex disassembly; and the authors conclude that ARID1-less, cBAF-less states promote carcinogenesis across tissues and advise caution against paralogue-directed therapies for ARID1-mutant cancer.

   Dual ARID1A/ARID1B loss leads to rapid carcinogenesis and disruptive redistribution of BAF complexes · Wang Z, Chen K, Jia Y, Chuang JC, Sun X, Lin YH, et al. · Nat Cancer 2020
   PMID 34386776DOI
   ▸What this evidence cannot show

   Abstract-level read; the full text was not opened this session. LIVER, SKIN and ENDOMETRIUM - no haematopoietic arm. The 'caution against paralog-directed therapies' is the authors' own conclusion about ARID1-mutant cancer and is quoted as such; it is imported here as a warning applicable by analogy, since KMT2A-rearranged ALL is not an ARID1-mutant cancer and the specific therapeutic scenario is not the same one.

   ▸Evidence that argues against this (1)

   • This paper's conclusion is in direct tension with the synthetic-lethal literature it cites: ARID1B loss is exploitable in ARID1A-mutant cells according to one line of work and carcinogenic in ARID1A-null tissue according to this one. The reconciliation is presumably context and degree - acute loss in a transformed cell versus chronic loss in a normal tissue - but the field does not currently have one, and this entry records the disagreement rather than choosing. Elife 2017
7. E7moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

   Bounded literature counts, 2026-09-23: ARID1B[tiab] 509; ARID1B[tiab] AND KMT2A[tiab] 7 (all read at title level - syndromic-neurodevelopment, growth-chart methodology, lymphoma risk, tumour-mutational-burden and Cornelia-de-Lange papers; NONE concerns a fusion-target relationship); ARID1B[tiab] AND MLL[tiab] 4; ARID1B[tiab] AND "MLL-AF9"[tiab] 1 (the Leukemia 2022 conditional-knockout paper); ARID1B[tiab] AND leukemia[tiab] 18; ARID1B[tiab] AND "SWI/SNF"[tiab] 139; ARID1B[tiab] AND hematopoi*[tiab] 6; ARID1B[tiab] AND "MLL-AF4"[tiab] 0. On the alias: BAF250b[tiab] 10; BAF250b[tiab] AND leukemia[tiab] 0; "ELD/OSA1"[tiab] 1.

   PubMed esearch via scripts/pubmed_safe.py, 2026-09-23 · Q119-genes-the-measurement-named · NCBI PubMed 2026
   ▸What this evidence cannot show

   Bounded by TITLES AND ABSTRACTS only (rule 80). One tag per term, no parenthesised group (rule 94); every querytranslation was read back and each expanded to "TERM"[Title/Abstract]. Alias counts are reported SEPARATELY and not unioned. The KMT2A, leukemia and hematopoi* result sets were read at title level; the 139 SWI/SNF records were not. Note the shape of the KMT2A intersection here: seven records, all of them neurodevelopmental or incidental, because ARID1B and KMT2A co-occur in intellectual-disability gene panels - a co-citation artefact, not a biological link.

   ▸Evidence that argues against this (1)

   • The seven-record ARID1B-KMT2A intersection is a textbook example of a count that looks like evidence of a connection and is the opposite: every hit is a neurodevelopmental-genetics paper in which both genes appear on a panel. Reading the titles was what distinguished a co-citation artefact from a literature, and the counts alone could not have. NCBI PubMed 2026
8. E8moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

   In RS4;11 (KMT2A::AFF1) the hg19 2 kb bin chr6:157,100,000 reaches 75.8x the MLL1-N track's own genome median and is in the top thirty bins genome-wide; under VTP-50469 its MLL1-N within-sample share ratio is 0.236 and its MENIN share ratio is 0.072.

   Q118 re-analysis of GSE127507, instrument B: 2 kb genome-wide bins from the deposited .deduped.tdf files (RS4;11 MLL1n GSM3635695/96, MENIN GSM3635693/94, input GSM3635699); analyses/q118-menin-at-the-human-index-fusion/results-instrumentB.json, run 2026-09-23 · Iqbal Lab oncofusion knowledge graph 2026
   ▸What this evidence cannot show

   THE MEASUREMENT THAT NAMED THIS GENE IS NOT FUSION-SPECIFIC. The track is MLL1n, Bethyl A300-086A, an MLL1 N-TERMINUS antibody whose epitope is retained in KMT2A::AFF1, so it reads fusion and wild-type MLL1 together at a masking factor the same beat measured as 3.54-4.00x. n = 1 per antibody per condition - GSE127507 has no replicate - and the depositors measured a Drosophila spike-in and deliberately did not apply it, so cross-condition figures are within-sample shares. Both ratios here are unremarkable within the top thirty; ARID1B is in this beat because of its published MLL-fusion genetics, not because its bin stands out.

   ▸Evidence that argues against this (1)

   • A promoter bin for a broadly expressed chromatin-remodeller subunit is close to the least surprising place to find MLL1-reactive material, since wild-type MLL1 occupies CpG-island promoters of housekeeping and chromatin genes. Nothing about this bin argues for a fusion-specific relationship, and the entry's interest in ARID1B would survive unchanged if the bin did not exist. Iqbal Lab oncofusion knowledge graph 2026
9. E9strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

   On the GRCh37 primary assembly the bin centre chr6:157,100,000 lies 567 bp from the start of ENST00000414678 and 914-937 bp from the starts of ENST00000275248, ENST00000367148, ENST00000346085 and ENST00000350026, all plus-strand transcripts of ENSG00000049618 (ARID1B), so the bin is at the ARID1B promoter.

   Ensembl GRCh37 REST: lookup/id/ENSG00000049618?expand=1, per-transcript start coordinates and strand · Ensembl · Ensembl GRCh37 archive 2026
   ▸What this evidence cannot show

   Fetched live 2026-09-23, resolved from a stable gene id to transcript coordinates rather than by name lookup. Ensembl gene models, not experimentally mapped start sites in a leukaemic cell, and a 2 kb bin cannot resolve transcripts 370 bp apart. This check is run for all six genes in this beat as an independent cross-check of the source entry's RefGene-window assignment; for ARID1B the two agree.

   ▸Evidence that argues against this (1)

   • A promoter assignment from gene models says where transcription is ANNOTATED to start, not where it starts in RS4;11. ARID1B has five annotated starts within 370 bp of each other and the 2 kb bin cannot separate them; more importantly, none of this establishes that the promoter is the only ARID1B sequence in the bin, and a 2 kb window at a CpG-island promoter routinely contains regulatory sequence for the neighbouring gene as well. Ensembl GRCh37 archive 2026
10. E10weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.paper

    ARID1B is a partner in recurrent leukaemia gene fusions in its own right: ARID1B::ZNF384 occurs in B-cell acute lymphoblastic leukaemia as part of the ZNF384-rearranged subtype, and a cryptic t(6;21)(q25;q22) generating RUNX1::ARID1B has been reported in acute myeloid leukaemia.

    Frequency and outcome of pediatric acute lymphoblastic leukemia with ZNF384 gene rearrangements including a novel translocation resulting in an ARID1B/ZNF384 gene fusion · Pediatr Blood Cancer 2016
    PMID 27392123
    ▸What this evidence cannot show

    TITLE LEVEL ONLY - neither this abstract nor the RUNX1::ARID1B report (PMID 42283395, Genes Chromosomes Cancer) was fetched and read this session, so no frequency, cohort size or outcome is claimed and the confidence is weak for that reason. A third title-level record (PMID 37336773) reports an ARID1B::ZNF384 case simulating multifocal osteomyelitis. These are recorded as existence claims to place ARID1B in the leukaemia fusion landscape, and nothing is inferred from them.

    ▸Evidence that argues against this (1)

    • ZNF384-rearranged B-ALL and KMT2A-rearranged B-ALL are distinct entities with different biology and outcomes. An ARID1B observation in the ZNF384 subtype is not evidence about the index fusion, and pooling them would repeat the partner-pooling error this graph's epidemiology entries repeatedly identify. Iqbal Lab oncofusion knowledge graph 2026

• ← supportsDepMap agrees with the mouse genetics: ARID1B is not a FUSION-SELECTIVE dependency (+0.0145, p=0.80) and reads -0.08 in both groups - but CYCLE 81 SHOWED THAT FLATNESS IS PARALOGUE BUFFERING, not dispensability. ARID1A and ARID2 ARE dependencies and are equally so in both groups - so no BAF subunit tested is fusion-selective. And MEN1 (-1.26) is a harder and more selective dependency than KMT2A itself (-0.81)

  human CRISPR agrees with the murine genetics this entry reports: ARID1B is -0.0833 in KMT2A-rearranged and -0.0979 in other leukaemia screens, dispensable in all 57, with working positive controls

• ← supportsARID1B is not dispensable - it is paralogue-buffered. In ARID1A-mutant lines its CRISPR dependency goes from -0.0724 to -0.3709 (p=0.0002) while ARID1A's own falls away (+0.1925, p=0.044) and ARID2/SMARCA4 stay flat. Cycle 80's 'dispensable everywhere' framing was wrong and the caveat that entry carried was right

  ARID1B has a real, measurable cBAF dependency once its paralogue is disrupted - it was redundancy, not irrelevance

• ← supportsCycle 81 named lineage as the single biggest hole in the ARID1B paralogue-buffering result. Permuting ARID1A status only WITHIN lineage strata moves the estimate from -0.2985 to -0.2961 - a change of 0.0024 - with p = 0.00140, the instrument check surviving, both specificity controls flat and zero mutant lines dropped. Lineage explains none of it

  ARID1B's dependency in ARID1A-disrupted lines is not a lineage artefact - the effect is unchanged when only same-disease-label lines are compared with each other

• ← related toARID1B's buffering effect survives leave-one-out in 11 of 11 - even dropping the line whose own effect is -1.2269 leaves -0.2129 at p = 0.00465. But the lineage-concentration test says there was LITTLE confound to remove, cycle 84's headline confirms an absence rather than defeating a threat, and my own pre-registered TVD threshold fired on a measure that is large by construction at n=11

  the mutant group is HETEROGENEOUS - individual effects run +0.0550 to -1.2269 and roughly 5 of 11 sit near or above zero, so buffering is a statement about a subset of ARID1A-mutant lines

• ← supportsCycle 81's pre-registered PRIMARY stratum was underpowered at n=4 for five cycles. A different source gives n=124: ARID1B buffering is -0.2475 (p=5e-05) in truncating-ARID1A lines and +0.0117 (p=0.66) in missense-only ones - a complete dose-response. My own specificity control failed, and it failed because I chose it from inside the complex I was studying

  a complete dose-response on ARID1A functional status, not a graded one

• ← supportsRedone at 11x the power: at n=124 the truncating-ARID1A group IS materially lineage-concentrated (TVD permutation p = 0.00020, against p = 0.42 at n=11), so this time the lineage test had something real to remove - and the effect survives it at -0.2516. The missense null is tight, CI95 (-0.0242, +0.0473), excluding -0.10. And there are now 4 truncating-ARID1A leukaemia lines, one short of testable

  the dose-response is a THRESHOLD, not a gradient: the missense bootstrap interval is 0.0715 wide and excludes -0.10

• ← related toI withheld these four numbers last cycle to keep an underpowered contrast out of a narrative, and they point the OTHER WAY: the four truncating-ARID1A leukaemia lines mean -0.0401 against a leukaemia wild-type of -0.0923, a naive +0.0523 where pan-cancer gives -0.2475. And NONE of the four carries a KMT2A fusion, so the intersection this whole thread needs is EMPTY, not merely small

  the only leukaemia evidence that exists points the other way at n=4, and none of the four lines carries a KMT2A fusion

• ← related toIn St. Jude paediatric AML, 2 of the 6 ARID1A-mutant cases are KMT2A-rearranged and BOTH ARID1A mutations are TRUNCATING frameshifts in KMT2A::MLLT10 cases - so the exact combination cycle 86 showed the ARID1B dependency is confined to DOES occur in patients, and cycle 88's empty DepMap intersection is a cell-line sampling artefact rather than a fact about the disease. The paediatric ALL cohort returned zero ARID1A mutations, which is panel coverage and not biology

  supplies the patient population the eight-cycle paralogue-buffering thread had been inferring about without one

• ← related toThe two TARGET B-ALL studies in cBioPortal are 868 of the same patients assayed twice, so joining them DOES answer the question I had just written off as unanswerable: all four ARID1A-mutant B-ALL patients are definitively KMT2A-negative, three of them with truncating calls. The answer is zero of three - and at a 2.7% KMT2A-positive rate an empty overlap has probability 0.92, so it is an upper bound and not a finding

  the paralogue-buffering thread has no documented B-lymphoid population, and this bounds how large one could be

• ← refutesOnly 69 of the 529 entries carrying absence_class are absence entries, and every conceptually novel token authors reached for - partially-measured, measured-and-contradicted, not-worth-measuring - is on an interface entry, where the thing being graded is not an absence at all

  carries never-measured while its prose describes a measurement that exists for a comparator - one of only two such mislabels found

• related to →ARID1A (BAF250A) - required in the MLL-AF9 screen, present at the Myc enhancers, and flat in every selectivity measure

  mutually exclusive cBAF paralogues; ARID1B deficiency is synthetically lethal with ARID1A mutation and >80% of their regulated loci are distinct

• related to →ARID2 (BAF200) - the PBAF subunit whose loss ACCELERATES MLL-AF9 leukaemia, and which is nonetheless required to maintain it

  cBAF versus PBAF ARID subunits with different signs in MLL-AF9: Arid1b loss promotes leukaemogenesis at either stage, Arid2 enhances at initiation yet is required at maintenance

• confers resistance to →KMT2A::MLLT3 (MLL-AF9), t(9;11)(p21.3;q23.3)

  conditional Arid1b deletion at initiation OR maintenance PROMOTES MLL-AF9 leukaemogenesis in mouse - a tumour-suppressor sign, the wrong direction for a drug target

• supports →The SWI/SNF dependency in KMT2A-rearranged leukaemia belongs to ncBAF, not to canonical BAF - and the drug in the clinic is aimed at the wrong module

  if losing the cBAF-defining subunit makes MLL-AF9 worse, cBAF is not the fusion's dependency - a subtractive argument this hypothesis did not previously carry

• part of →"SWI/SNF" is three different machines: cBAF, PBAF and ncBAF - and only one of them looks KMT2A-selective

  one of the two mutually exclusive ARID1 subunits that DEFINE canonical BAF

• expressed in →Bone marrow haematopoietic stem cell — the postnatal comparator, and why the same fusion behaves differently here

  modest steady-state phenotype but ARID1B-dependent myeloid reconstitution on transplantation; ARID1A/ARID1B double loss causes acute bone-marrow failure

• supports →Instrument check: the top thirty RS4;11 MLL1-N 2 kb bins recovered from a deposited .tdf are JMJD1C (259x the genome median), ZMYND11, BAHCC1, CDK6, MYB, RUNX2, IGF2BP2, FOXP1, SATB1 and ARID1B - canonical KMT2A::AFF1 targets, annotated only against the series' own RefGene windows

  the entry that named this gene; its bin is independently confirmed here as the ARID1B promoter (567 bp)

• related to →Two published ways a fusion oncoprotein engages BAF - be a subunit, or recruit it with a disordered domain - and KMT2A has never been tested for either

  ARID1B is a BAF subunit that is itself a leukaemia fusion partner (ARID1B::ZNF384, RUNX1::ARID1B), the same shape as SS18::SSX in that analogy

• ← related toZMYND11 (BS69) - an H3.3K36me3 reader that restrains elongation, is itself the 5' partner of a chromatin-hijacking AML fusion driving HOXA/MEIS1/MYB, and whose promoter is the third-strongest MLL1-N bin in RS4;11

  both are chromatin proteins in this beat that are simultaneously subunits/readers with their own biology AND recurrent leukaemia fusion partners - a recurring architecture in this disease family