SMARCB1 (BAF47/INI1/SNF5) - zero entries, zero mentions and, on the evidence, correctly so: it is lost in primary AML rather than required

This is a prediction, not a measurement. No one has seen this protein in this shape. The colours say how much the software trusts itself, and that is worth reading before anything else — the orange stretches are places the model is effectively saying this part has no fixed shape.

• Any strategy that further destabilises cBAF in this disease risks the rhabdoid outcome: SMARCB1 loss is itself oncogenic in the same paediatric age band.
• SMARCB1's pan-essentiality of -0.85 means a hypothetical SMARCB1-directed agent has essentially no therapeutic index.
• Because SMARCB1 loss makes cells hypersensitive to BRG1 inhibition, a patient whose blasts have low SMARCB1 may respond to FHD-286 for a reason unrelated to the KMT2A rearrangement - which would confound any biomarker built on the rearrangement.

1. Do KMT2A-rearranged blasts retain SMARCB1 protein? The one primary-AML study reporting near-complete loss did not stratify by rearrangement, and if KMT2A-rearranged blasts are cBAF-perturbed the ncBAF synthetic-lethal logic would transfer directly.
2. Is the ncBAF dependency seen in KMT2A-rearranged lines (BRD9 rank 39) downstream of a cBAF perturbation, as it is in rhabdoid tumour and synovial sarcoma, or does it have an independent cause?
3. Does the 1999 report of hSNF5/INI1 deletion in CML replicate with modern copy-number data?

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

   SMARCB1 is UniProt Q12824, 385 aa, primary gene SMARCB1, synonyms BAF47 INI1 SNF5L1; entry name SNF5_HUMAN.

   UniProtKB Q12824 (SNF5_HUMAN) · UniProt Consortium · UniProtKB 2026
   ▸What this evidence cannot show

   gene_exact:SMARCB1 AND organism_id:9606 AND reviewed:true, run 2026-09-20; one reviewed row; gene_primary and length read back from the fetched record. Note UniProt lists SNF5L1, not bare SNF5, as the gene synonym - 'SNF5' appears only inside the protein-name string.

   ▸Evidence that argues against this (1)

   • The alias set BAF47/INI1/SNF5 is used interchangeably in three literatures (leukaemia, rhabdoid tumour, HIV integrase biology, where INI1 was named as integrase interactor 1), so a symbol-keyed search returns three different fields' papers under one name. Kadoch C et al. 2013
2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   SMARCB1 is present in cBAF and PBAF and absent from ncBAF; in cBAF-perturbed cancers including SMARCB1-deficient malignant rhabdoid tumour, ncBAF subunits are synthetic-lethal targets and BRD9 depletion attenuates proliferation.

   A non-canonical SWI/SNF complex is a synthetic lethal target in cancers driven by BAF complex perturbation. · Michel BC, D'Avino AR, Cassel SH, Mashtalir N, et al. · Nat Cell Biol 2018
   PMID 30397315DOI
   ▸What this evidence cannot show

   Demonstrated in synovial sarcoma and rhabdoid tumour lines, not in leukaemia. The synthetic lethality is conditional on cBAF perturbation, which KMT2A-rearranged leukaemia does not obviously have.

   ▸Evidence that argues against this (1)

   • If the ncBAF dependency of these tumours is strictly conditional on cBAF perturbation, then the KMT2A-rearranged BRD9 dependency must have a different cause, and the analogy explains nothing about this disease. Hohmann AF et al. 2016
3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

   SMARCB1 ranks 14,024 of 17,728 with difference +0.0415 (KMT2A-r -1.1953, other leukaemia -1.2368, pan-essentiality -0.847) and fails the lineage split in both halves (myeloid z -0.19, B-lineage +0.27).

   Genome-wide KMT2A-rearranged selectivity scan, 17,728 genes (DepMap Chronos via Open Targets) · Oncofusion knowledge graph, analyses/depmap-kmt2a-selective-scan-full/run.py · DepMap Chronos gene effect via Open Targets GraphQL 2026
   ▸What this evidence cannot show

   A pan-essentiality of -0.85 puts this gene near the assay floor, where index-versus-comparator differences are uninterpretable rather than small - this is an absence of measurable selectivity, not a demonstration that SMARCB1 is irrelevant.

   ▸Evidence that argues against this (1)

   • If primary AML blasts have already lost SMARCB1 protein, cell lines that retain it are not the right model for the question, and their dependency says little about patient disease. Chatterjee SS et al. 2018
4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

   Human primary AML cells exhibit near-complete loss of SMARCB1 (BAF47/SNF5/INI1) and SMARCD2 (BAF60B); the residual complex nucleates on SMARCC1 (BAF155), colocalises with H3K27ac at oncogenic loci, and loss of SMARCB1 increases recruitment of that complex and associated HATs, raising Rac-GEF expression, migration and survival - a tumour-suppressor role.

   SMARCB1 Deficiency Integrates Epigenetic Signals to Oncogenic Gene Expression Program Maintenance in Human Acute Myeloid Leukemia. · Chatterjee SS, Biswas M, Boila LD, Banerjee D, et al. · Mol Cancer Res 2018
   PMID 29483235DOI
   ▸What this evidence cannot show

   Primary AML cohort not selected for KMT2A rearrangement; 'near complete loss' is a protein-level observation in that cohort; the mechanistic arm is loss-of-function in lines. Whether KMT2A-rearranged blasts specifically show this has not been reported.

   ▸Evidence that argues against this (1)

   • SMARCB1 is one of the most depleted genes in every DepMap leukaemia line (-1.2), which is hard to square with primary blasts having lost the protein; one of the two observations is likely to be about a different property (protein level versus functional requirement). Oncofusion knowledge graph et al. 2026
5. E5weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.paper

   Frequent deletion of hSNF5/INI1 was reported in chronic myeloid leukaemia, establishing an early precedent for SMARCB1 loss rather than requirement in a myeloid malignancy.

   Role of SWI/SNF in acute leukemia maintenance and enhancer-mediated Myc regulation. · Shi J, Whyte WA, Zepeda-Mendoza CJ, Milazzo JP, et al. · Genes Dev 2013
   PMID 24285714DOI
   ▸What this evidence cannot show

   Cited here as a secondary reference inside Shi et al.'s introduction (Grand et al. 1999, Cancer Res 59:3870). This beat did NOT retrieve or read the primary report, so the claim is carried at weak confidence and the primary should be fetched before it is relied on.

   ▸Evidence that argues against this (1)

   • A 1999 deletion-mapping report predates modern copy-number calling and the cited frequency has not been re-examined here; treat it as historical context only. Chatterjee SS et al. 2018
6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

   `SMARCB1 AND (MLL OR KMT2A)` returns 11 PubMed records and title-checking finds none about SWI/SNF in KMT2A-rearranged leukaemia; the two most recent on-topic-looking hits are an ATRT CAR-T paper and a review of cognitive genetics.

   PubMed E-utilities esearch record counts: `SMARCB1 AND (MLL OR KMT2A)` = 11; aliases `SNF5 AND (MLL OR KMT2A)` = 4, `INI1 AND (MLL OR KMT2A)` = 4, `BAF47 AND (MLL OR KMT2A)` = 4; complex form `SWI/SNF AND (MLL OR KMT2A)` = 27 also run and title-checked per rule 64; operand controls `leukemia` = 405,093 and `SMARCB1` alone large; run 2026-09-20 · NCBI E-utilities, run for this beat on 2026-09-20 · PubMed 2026
   ▸What this evidence cannot show

   Bounded absence. Bounded to PubMed on 2026-09-20 over SMARCB1, SNF5, INI1 and BAF47. Two of the eleven were fetched and read in full (PMID 32341579, a B7-H3 CAR-T study in ATRT; PMID 33659785, a review of longitudinal cognitive genetics) and both are noise, which is what justifies calling the raw count inflated; the remaining nine were judged on titles only.

   ▸Evidence that argues against this (1)

   • The complex-level paper PMID 29483235 does carry a substantive SMARCB1 result in AML and is reachable from the symbol, so the literature is not empty - it is empty of a KMT2A-rearranged SMARCB1 result specifically. Chatterjee SS et al. 2018
7. E7established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

   The SS18-SSX fusion competes for assembly with wild-type SS18 and forms an altered BAF complex lacking the tumour suppressor BAF47 (hSNF5), which then activates SOX2 by reversing polycomb repression.

   Reversible disruption of mSWI/SNF (BAF) complexes by the SS18-SSX oncogenic fusion in synovial sarcoma. · Kadoch C, Crabtree GR · Cell 2013
   PMID 23540691DOI
   ▸What this evidence cannot show

   Synovial sarcoma, not leukaemia; the eviction is shown biochemically and by rescue with excess wild-type SS18.

   ▸Evidence that argues against this (1)

   • No KMT2A fusion partner is a BAF subunit, so the SS18 mechanism - a fusion that is itself part of the complex - has no direct counterpart here. Boulay G et al. 2017
8. E_q91_smarcb1_loc1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

   UniProtKB Q12824 (SNF5_HUMAN, primary gene SMARCB1, 385 aa, read back live 2026-09-21) carries exactly ONE subcellular location value, 'Nucleus', and its EVIDENCE LIST IS EMPTY - no ECO:0000269, no ECO:0000305, no PROSITE rule, no by-similarity source. There is no location note and no annotated NLS or NES in the portion of the feature table this beat retrieved, which is consistent with the primary literature's finding that SMARCB1 lacks a bona fide NLS.

   UniProtKB Q12824 SUBCELLULAR LOCATION comments, evidence codes and feature table, fetched live 2026-09-21 (a single location value with an empty evidence list) · UniProt Consortium · UniProtKB 2026
   ▸What this evidence cannot show

   An empty evidence list is not a claim that the annotation is wrong; it is a claim that the database offers the reader nothing to check. Tranche 3 found the same form on MAP3K14 and chose not to treat it as NOT APPLICABLE, and the same choice is made here.

   ▸Evidence that argues against this (1)

   • The bare word 'Nucleus' is not merely under-evidenced here, it is actively misleading for a fifth of the tumours in which this protein is clinically assessed: 19 of 102 primary ATRT show distinct CYTOPLASMIC SMARCB1 by immunohistochemistry, concentrated in C-terminal truncating variants. A pathologist or a curator reading UniProt's one word would conclude that cytoplasmic staining is an artefact. The absent evidence list and the absent NES annotation are the same omission seen twice, and the missing feature is the one that carries the therapeutic handle. Pathak R et al. 2021
9. E_q91_smarcb1_loc2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   In a series of 102 PRIMARY atypical teratoid/rhabdoid tumours, distinct CYTOPLASMIC SMARCB1 staining was found in 19 cases (19%), highly over-represented in tumours with variants truncating or mutating the C-terminal part of SMARCB1 (15/19 versus 4/83; chi-square 56.04, p = 1.0e-10) and in the ATRT-TYR molecular subgroup (16/36 versus 3/66; chi-square 24.47, p = 7.6e-7). Mechanism: SMARCB1 LACKS A BONA FIDE NUCLEAR LOCALISATION SIGNAL and harbours a MASKED NUCLEAR EXPORT SIGNAL that C-terminal truncation unmasks. GFP-SMARCB1(Q318X) localised to the cytoplasm; the double mutant GFP-SMARCB1(Q318X;L266A), in which the SMARCB1-NES/exportin-1 interaction is disrupted, localised to the nucleus. Cytoplasmic SMARCB1(Q318X) could not induce senescence by morphology or SA-beta-galactosidase while the nuclear double mutant regained that function, and SELINEXOR inhibited nuclear export of SMARCB1(Q318X) and caused rapid cell death in rhabdoid tumour cells.

   Inhibition of nuclear export restores nuclear localization and residual tumor suppressor function of truncated SMARCB1/INI1 protein in a molecular subset of atypical teratoid/rhabdoid tumors. · Pathak R, Zin F, Thomas C, Bens S et al. · Acta Neuropathol 2021
   PMID 34003336DOI
   ▸What this evidence cannot show

   Paediatric CNS tumour, not leukaemia; the mechanistic half is GFP-fusion over-expression in cell lines while the 19% figure is immunohistochemistry on archival tumour, and those are different assays supporting the two halves of one story. Read from the abstract, which is unusually complete. Rule 73: the system is primary human ATRT and ATRT cell lines.

   ▸Evidence that argues against this (1)

   • Transferring this to leukaemia would be wrong in a specific way worth writing down. In primary AML the reported lesion is near-complete LOSS of SMARCB1 PROTEIN with residual SWI/SNF nucleating on SMARCC1, not a mislocalised truncated protein - a different failure mode with a different therapeutic consequence, because selinexor can only relocate a protein that is still being made. The two findings are compatible and they point opposite ways for intervention, and this graph should not let the ATRT result imply an export-inhibitor rationale in AML. It also cuts against this entry's own second open question: the rhabdoid ncBAF synthetic-lethal logic assumes a cBAF perturbation of the ATRT type, which the AML data do not describe. Chatterjee SS et al. 2018
10. E_q91_smarcb1_loc3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

    Counted absence, 2026-09-22, querytranslation read back. SMARCB1 has no assigned SUBNUCLEAR compartment: SMARCB1[tiab] AND localization[tiab] = 20 records (operand control, a real localisation literature) and SMARCB1[tiab] AND nuclear[tiab] AND localization[tiab] = 6, all on the nucleocytoplasmic axis. The alternative symbol out-retrieves the current one in this frame: hSNF5[tiab] AND nuclear[tiab] = 32.

    Counted absence with alias control: SMARCB1 versus hSNF5 in the nuclear-localisation frame, querytranslation recorded, run 2026-09-22 · NCBI E-utilities esearch (db=pubmed), run by Q91 · PubMed E-utilities 2026
    ▸What this evidence cannot show

    Bounded by titles and abstracts (rule 80). Two of this protein's four common names were run; INI1 and BAF47 were not, and INI1 in particular indexes the HIV-1 integrase literature, where the protein is studied heavily under that name and where a nuclear-compartment observation could plausibly sit.

    ▸Evidence that argues against this (1)

    • The 32-versus-6 split argues that this absence is weaker than it reads, and in a direction the brief anticipates. SMARCB1 is the worst case in this tranche for symbol-keyed retrieval because it has FOUR names in active use - SMARCB1, hSNF5, INI1 and BAF47 - each dominant in a different subfield: the current symbol in cancer genomics, hSNF5 in chromatin biochemistry, INI1 in HIV integration, BAF47 in diagnostic pathology. A subnuclear observation made by the HIV field under INI1, which is where the multimerisation and DNA-binding biochemistry was done, would be invisible to every query run here. Das S et al. 2009

• ← bindsPPP1R15A / GADD34 (O75807) - the eIF2alpha phosphatase adaptor, and KMT2A fusion proteins were shown to BIND it and block its apoptosis in 1999, a paper this graph has never cited

  GADD34 binds human SNF5/INI1 (SMARCB1) - reported in the same 1999 paper as the HRX interaction, never followed up; a GADD34-SMARCB1-KMT2A triangle nobody has resolved

• ← related toSMARCD1 (BAF60A) - rank 59 of 17,728, survives the lineage split, and had never been named in this graph or studied in this disease

  Complex membership does not fix a compartment for BAF subunits: SMARCB1 is cytoplasmic in 19% of 102 primary ATRT through an unmasked NES (PMID 34003336), which is why SMARCD1's uninformative nuclear annotation leaves the question open rather than settled (Q91 2026-09-22)

• related to →XPO1 / exportin-1 (CRM1) - the cell's main nuclear export receptor, and a chromatin-bound transcriptional scaffold

  SMARCB1 has no bona fide NLS but a MASKED NES; C-terminal truncation unmasks it, exportin-1 exports the protein, and selinexor puts it back in the nucleus and kills rhabdoid cells (PMID 34003336) (Q91 2026-09-22)

• related to →Two pools of one protein: the chromatin-bound XPO1 that matters here has never been separated biochemically from the soluble export machine

  A worked case of XPO1 setting a chromatin regulator's compartment: cytoplasmic SMARCB1 in 19/102 primary ATRT, restored to the nucleus and to senescence competence by blocking the NES/exportin-1 contact (PMID 34003336) (Q91 2026-09-22)

• related to →SMARCC1 / BAF155 (Q92922) - the SWI/SNF core scaffold, in the pan-MLL-fusion core, and the missing gene entry in this graph's SWI/SNF beat

  In primary AML, SMARCB1 protein is near-completely lost and residual SWI/SNF nucleates on SMARCC1 (PMID 29483235) - a different failure mode from the ATRT mislocalisation, and the one relevant to leukaemia (Q91 2026-09-22)

• ← co-occurs withVIM::KMT2A - the one KMT2A rearrangement that keeps the SET domain and still throws away the menin-binding exon

  concurrent SMARCB1 mutation in both cases of the pan-sarcoma screen (n=2)

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

  cBAF and PBAF core; absent from ncBAF

• related to →BRD9 — dispensable in the average cell, rank 39 of 17,728 in KMT2A-rearranged lines, and with an on-target proof most epigenetic targets never get

  SMARCB1 loss is what makes ncBAF/BRD9 synthetic-lethal in rhabdoid tumour; the architecture this beat tests in leukaemia

• co-occurs with →SMARCD2 (BAF60B) - named in one KMT2A paper in the whole of PubMed, and the dependency it carries is myeloid identity, not the rearrangement

  both reported near-completely lost at protein level in primary AML blasts

• ← related toTwo 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

  BAF47 eviction by SS18-SSX; SMARCB1 loss as the rhabdoid entry point to the same architecture

• ← supportsSWI/SNF enumeration for KMT2A-rearranged leukaemia: every subunit searched, every count dated, and what was read and rejected

  SMARCB1's bounded absence: 11 records, title-checked to 0 on-beat