Oncofusion

start / graph / absence-q149-only-one-public-b-lineage-cohort-annotates-the-kmt2a-partner-beside-an-expression-matrix

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q149-irx1-expression-across-partners

Only one public B-lineage cohort annotates the KMT2A fusion PARTNER per sample beside an expression matrix - every other large paediatric leukaemia resource records only 'MLL yes/no' or the 11q23 band

also called q149 resource survey · no partner annotation in TARGET or MILE

To ask whether a gene behaves differently depending on which partner the KMT2A fusion has, you need a dataset that records the partner for each child. Almost none do - they only record whether a KMT2A rearrangement is present. We found exactly one that does, in the right disease.

The same thing, for a scientist

**The bound this beat rests on, stated as three categories and not two.** Eleven resources were checked for the conjunction *per-sample KMT2A fusion partner* AND *per-sample expression values*: | resource | n | partner annotation | expression | status | |---|---:|---|---|---| | **GSE68720** COG P9407 infant ALL | 97 | **YES** - `mll_par: AF4 / ENL / AF9 / EPS15 / ASAH3` | array, 8.3 MB series matrix | **VERIFIED-PRESENT** | | **aml_stjude_2024** paediatric AML | 887 | **YES** - per-sample structural-variant calls | mRNA-seq CPM | **VERIFIED-PRESENT** | | **ccle_broad_2019** cell lines | 1,739 | **YES** - per-line SV calls | RNA-seq RPKM | **VERIFIED-PRESENT** | | **aml_ohsu_2022** BeatAML, adult | 942 | **PARTIAL** - SVs present, only MLLT3 named; 30 records are bare `KMT2A rearrangement` | RNA-seq | **VERIFIED-PRESENT, partner incomplete** | | `all_phase2_target_2018_pub` TARGET ALL ph II | 1,978 | **NO** - `MLL_STATUS: Positive` (48) / `Negative` (1,464) / `Unknown` (431); `MOLECULAR_SUBTYPE: MLL-Rearranged` (29) | yes | **VERIFIED-ABSENT** | | `aml_target_2018_pub` TARGET AML | 1,025 | **NO** - `MLL: Yes` (145) / `No` (818) / `Unknown` (49); `PRIMARY_CYTOGENETIC_CODE: MLL` (171) | yes | **VERIFIED-ABSENT** | | `bll_target_gdc` | 989 | **NO** - 20 clinical attributes, no cytogenetic field at all | yes | **VERIFIED-ABSENT** | | `aml_target_gdc` | 2,766 | **NO** - same | yes | **VERIFIED-ABSENT** | | **GSE13159** MILE stage 1 | 2,096 | **NO** - `leukemia class: Pro-B-ALL with t(11q23)/MLL` (70), `AML with t(11q23)/MLL` (38): the band, never the partner | 886 MB matrix | **VERIFIED-ABSENT** | | **GSE49642 / GSE62190** Leucegene | 43 / 82 | **NO** - the entire characteristics set is `tissue:` and `cell type: acute myeloid leukemia` | no processed matrix in GEO | **VERIFIED-ABSENT** | | St Jude Cloud PeCan / ProteinPaint | - | - | - | **UNREACHABLE** | | DepMap portal expression | - | - | - | **UNREACHABLE** | **So the whole B-lineage partner-stratified expression question in this disease rests on one 97-patient 2012 array deposit**, and every large modern cohort that could answer it better has thrown the partner away at the point of curation. That is the finding, and it is a curation problem rather than a data problem - the partner was known to the treating institutions. **And it is invisible to the current gene symbol.** GSE68720 was found only via the retired symbol `AF4`: GEO DataSets `infant acute lymphoblastic leukemia AND AF4` returns it; `... AND KMT2A` does not. Its metadata never contains the string `AFF1`.

Go deeper
Two instrument failures found while establishing this bound

1. db=geoprofiles returns a zero that is NOT silent - read the ErrorList. Verified at this desk:

esearch db=geoprofiles term=IRX1[Gene Symbol] AND GSE68720[GEO Accession]
  -> Count 0
     QueryTranslation: IRX1[Gene Symbol] AND GSE68720[GEO Accession]
     ErrorList/PhraseNotFound: GSE68720[GEO Accession]
     WarningList/OutputMessage: No items found.
esearch db=geoprofiles term=IRX1[Gene Symbol]            -> Count 2799   (control passes)

geoprofiles indexes only curated GDS DataSets and GSE68720 is not one. The zero is accompanied by an explicit PhraseNotFound naming the failing term, so a reader who checks only QueryTranslation - which is non-empty and looks correct - would take this as a real negative. Trusting it would have made the one usable resource in this beat look like it lacked IRX1 entirely. The rule the corpus already holds is that an empty querytranslation means failure; this extends it: a NON-empty querytranslation with a non-empty ErrorList/PhraseNotFound also means failure.

2. "t(4;11)"[All Fields] fails in db=gds as it does in db=pubmed - the esearch response carried no <Count> and an empty <QueryTranslation>, which is a hard parser failure, not a zero. Quoting is not sufficient in E-utilities for a cytogenetic term in the GEO database.

Checked and rejected, with the reason, so nobody repeats the search

Bounded rejections, three lines each is the standard and one is enough here:

  • GSE228296 - genotype: KMT2A-AFF1 (17) / KMT2A-MLLT3 (8) / WT (2), 1.05 MB gene matrix. Partner IS annotated, but 21 of 27 samples are cell line: CRISPR-engineered and only 2 are Primary. A useful isogenic comparator; not a patient cohort. NEVER-CHECKED for IRX1 values.
  • GSE295919 - LIN28B knockdown in KMT2A::AFF1 models. Partner is constant by construction, so no between-partner contrast is possible.
  • GSE18400 (60 infant ALL) and GSE19671 (15) - right disease, methylation arrays, wrong assay.
  • GSE13164 MILE stage 2 - custom platform GPL7473 whose IRX1 coverage was not verified. NEVER-CHECKED.
  • GSE52656 Leucegene part 2 - rejected by inheritance from its two sibling parts, not by direct per-sample test. NEVER-CHECKED per-sample.
  • GSE270185 - a genuine paediatric BCP-ALL patient series carrying a partner string, surfaced but not opened. NEVER-CHECKED, and the cheapest remaining lead.
  • ArrayExpress / BioStudies - the API returns 29,190 fuzzy OR-matched hits with no usable ranking, and the top hits are E-GEOD-* mirrors of GEO. Rejected as a distinct route, not as a resource.
  • GSE276679 / GSE276676 / GSE276675 NUTM1-rearranged; GSE217126 circular RNA; GSE179733 / GSE147862 / GSE147860 single-cell n=1-22; GSE13189 n=6; GSE40103 ESC model; GSE251658 n=6 ontogeny model; GSE295391 type Other. All off-beat for the stated reason.
Facts and figures
absence classmeasured-not-curated
what is absentper-sample KMT2A fusion partner annotation beside an expression matrix, in B-lineage paediatric leukaemia, in any resource other than GSE68720
verified present
[
  "GSE68720",
  "aml_stjude_2024",
  "ccle_broad_2019",
  "aml_ohsu_2022 (partner incomplete)"
]
verified absent for partner
[
  "all_phase2_target_2018_pub",
  "aml_target_2018_pub",
  "bll_target_gdc",
  "aml_target_gdc",
  "GSE13159",
  "GSE49642",
  "GSE62190",
  "GSE295919"
]
unreachable
[
  "pecan.stjude.cloud (1,278-byte SPA shell; /api/gene/IRX1 -> HTTP 404)",
  "proteinpaint.stjude.org (HTTP 403)",
  "ppr.stjude.org (connection failed)",
  "viz.stjude.cloud/api/termdb (HTTP 404)",
  "depmap.org/portal/api/* (HTTP 200 with a 5,175-byte 'DepMap - Verification' bot wall)"
]
never checked
[
  "GSE270185",
  "GSE13164",
  "GSE52656 per-sample",
  "MILE IRX1 values (886 MB matrix not downloaded)",
  "TARGET free-text KARYOTYPE parse"
]
found only via retired symbol
{
  "GSE68720": "AF4; GEO DataSets 'infant acute lymphoblastic leukemia AND AF4' returns it, '... AND KMT2A' does not"
}
instrument failures found
{
  "geoprofiles_zero": "Count 0 with a NON-empty QueryTranslation but ErrorList/PhraseNotFound naming GSE68720[GEO Accession]; control IRX1[Gene Symbol] alone returns 2799",
  "gds_translocation_parse": "\"t(4;11)\"[All Fields] in db=gds returns no Count and an EMPTY QueryTranslation - a parser failure, not a zero"
}
measurements
[
  {
    "quantity": "count_resources_with_partner_and_expression",
    "subject": "public paediatric leukaemia expression resources, B-lineage",
    "value": 1,
    "units": "resources",
    "n": 11,
    "universe": "GSE68720, aml_stjude_2024, ccle_broad_2019, aml_ohsu_2022, 4 TARGET studies, MILE, 2 Leucegene parts; 'B-lineage' restricts the 1 to GSE68720",
    "null": null,
    "note": "3 of 11 carry partner+expression at all, but 2 of those 3 are myeloid or cell lines; TWO further resources are UNREACHABLE and five are NEVER-CHECKED"
  },
  {
    "quantity": "n_kmt2a_r_cases_with_no_partner_annotation",
    "subject": "TARGET ALL ph II + TARGET AML + MILE stage 1",
    "value": 279,
    "units": "cases",
    "n": 5099,
    "universe": "MLL_STATUS Positive 48 + PRIMARY_CYTOGENETIC_CODE MLL 171 + MILE t(11q23) 70 and 38 minus overlap unknown",
    "null": null,
    "note": "an upper-bound arithmetic sum across three non-overlapping cohorts; these are KMT2A-r children with public expression data whose partner was discarded at curation"
  }
]
How this could bite us back
  • Any partner-stratified biomarker claim in B-lineage KMT2A-r leukaemia currently has one 97-patient 2012 array cohort behind it. A clinical decision resting on partner-specific expression is resting on that, and nothing has replicated it because nothing else records the partner.
Still open
  1. Do the TARGET free-text KARYOTYPE strings contain recoverable partners? A t(9;11) or t(4;11) in that field would turn a 48-case MLL_STATUS-positive arm into a partner-stratified one, and it is a string parse rather than a data request.
  2. Does the MILE consortium hold the partner for its 108 t(11q23) cases? That would be the largest partner-stratified expression cohort in existence, on the same platform as GSE68720.
  3. What is in GSE270185? It is a paediatric BCP-ALL patient series carrying a partner string and it was surfaced and not opened - the cheapest remaining lead in this beat.
Every source behind this entry (2)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    In the cBioPortal public API, the TARGET paediatric ALL Phase II study records MLL_STATUS as Positive in 48 of 1,943 samples and MOLECULAR_SUBTYPE as MLL-Rearranged in 29 of 1,163, with no clinical attribute naming a fusion partner, and the TARGET paediatric AML study records MLL Yes in 145 of 1,012 with PRIMARY_CYTOGENETIC_CODE MLL in 171 of 886 and likewise no partner attribute.

    Integrative analysis of complex cancer genomics and clinical profiles using the cBioPortal · Gao J, Aksoy BA, Dogrusoz U, et al. · Sci Signal 6:pl1 2013
    PMID 23550210DOI
    What this evidence cannot show

    Bounds the cBioPortal-curated view of TARGET, not the TARGET primary submission. The underlying TARGET data files or dbGaP-controlled clinical tables may carry a partner field that cBioPortal's curation dropped - that is a plausible and untested possibility, so the correct claim is 'absent from the curated resource' and NOT 'never recorded'. Attribute value counts were read from the live API on 2026-09-24 and are a snapshot.

    Evidence that argues against this (1)
    • TARGET AML additionally carries OTHER_MLL (Yes in 12 of 330) and CYTOGENETIC_CODE_OTHER, and KARYOTYPE is present in the ALL study - a free-text karyotype can contain t(9;11) or t(4;11) and therefore an implicit partner. Those free-text fields were NOT parsed here, so the 'no partner' verdict is about structured attributes only and could be softened by a karyotype-string parse. That is the cheapest way to prove this entry partly wrong. Gao J et al. 2013
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    The MILE study's 2,096-sample stage-1 deposit annotates KMT2A-rearranged cases only by cytogenetic band, as 'Pro-B-ALL with t(11q23)/MLL' in 70 samples and 'AML with t(11q23)/MLL' in 38, with no per-sample fusion partner.

    GSE13159 - Microarray Innovations in Leukemia (MILE) study, stage 1 (per-sample GSM metadata) · GEO depositors, series GSE13159 · NCBI Gene Expression Omnibus 2010
    What this evidence cannot show

    Established from the GSM brief-format metadata (77,892 lines) by a worker in this beat and reported here with that provenance; I did not re-fetch it myself, so this specific row is second-hand within this session. The 886 MB series matrix was deliberately not downloaded (disk constraint), so IRX1 values in MILE are NEVER-CHECKED - which is a separate fact from the partner annotation being absent.

    Evidence that argues against this (1)
    • MILE is on GPL570, the same platform as GSE68720, and therefore DOES carry the IRX1 probe 230472_at. Its 108 KMT2A-rearranged cases are the largest such expression set in existence and are unusable here only for lack of partner annotation - if the MILE consortium's own records carry the partner, the best available cohort for this question exists and is one request away. GEO depositors et al. 2010
Connected to