aml_stjude_2024 - 887 paediatric AML cases with per-sample KMT2A fusion-partner structural-variant calls for all 180 KMT2A-rearranged cases, plus mRNA-seq and FAB morphology
also called St Jude paediatric AML 2024 · Umeda 2024 pAML · aml_stjude_2024
A public dataset of 887 children with myeloid leukaemia. For every one of the 180 with a KMT2A fusion it records which partner gene was involved, plus gene activity and the leukaemia's cell type - which is what let us ask whether IRX1 tracks the partner or the cell type.
The same thing, for a scientist
Umeda et al.'s 887-case paediatric AML genomic framework [E1], served through the cBioPortal public API as study `aml_stjude_2024`. **It is the largest partner-annotated KMT2A-rearranged expression resource this corpus has found: 180 KMT2A-rearranged cases with the partner named per sample by the study's own structural-variant calls**, not by recall and not by a cytogenetic band. **Partner tally from `aml_stjude_2024_structural_variants` (KMT2A entrez 4297), 237 SV records over 180 samples:** | partner | n samples | above n>=5? | |---|---:|---| | MLLT10 | 57 (incl. 3 with a second call) | yes | | MLLT3 | 55 (incl. 1) | yes | | AFDN | 22 | yes | | ELL | 17 (incl. 1) | yes | | MLLT1 | 8 | yes | | MLLT6 | 5 | at the boundary | | MLLT11 | 3 | **NAME, DO NOT COMPUTE** | | USP2 | 3 | **NAME, DO NOT COMPUTE** | | **AFF1** | **1** | **NAME, DO NOT COMPUTE** | | EPS15, CORO1A, DNAJC1, NEBL, SARNP | 1 each | **NAME, DO NOT COMPUTE** | | site2 blank (`UNSPECIFIED`) | 4 | **NAME, DO NOT COMPUTE** | **The count cross-checks exactly against the study's own curated label:** `MOLECULAR_CATEGORY: KMT2Ar` is assigned to **180** samples, the same 180 that carry a KMT2A structural variant. That is an agreement of totals, not of per-sample identity, and is recorded as such. **What makes it uniquely useful here:** alongside expression it carries **`FAB_MORPHOLOGY`** (787 of 887), `MOLECULAR_CATEGORY` (887), `DX_CYTOGENETICS` (837) and `ICC_SUBCATEGORY` (243). The FAB field is what turned a pooled fusion-versus-rest difference at p ~ 0 into a morphology composition artefact - see `concept-q149-a-fusion-versus-rest-expression-difference-can-be-a-morphology-composition-artefact`. **And what it is not:** it is **AML**, and the IRX1/HOXA biology this corpus cares about is infant **B-lineage** ALL. It contains **one** KMT2A::AFF1 case. Anything read from it about the index fusion is n=1.
Access, and the exact calls
POST https://www.cbioportal.org/api/structural-variant/fetch
{"entrezGeneIds":[4297],"molecularProfileIds":["aml_stjude_2024_structural_variants"]}
POST https://www.cbioportal.org/api/molecular-profiles/aml_stjude_2024_mrna_seq_cpm/molecular-data/fetch?projection=SUMMARY
{"entrezGeneIds":[...],"sampleListId":"aml_stjude_2024_all"}
GET https://www.cbioportal.org/api/studies/aml_stjude_2024/clinical-data?attributeId=FAB_MORPHOLOGY&clinicalDataType=SAMPLE&projection=SUMMARY&pageSize=10000
Partner extraction is not symmetric. In the SV records KMT2A appears sometimes as site1HugoSymbol and sometimes as site2HugoSymbol - 67 KMT2A,MLLT3 records and 11 MLLT3,KMT2A records, for example. Taking site2 as the partner loses a third of the calls; the partner is whichever site is not KMT2A. Five records are KMT2A with a blank site2.
Expression profiles available: aml_stjude_2024_mrna_seq_cpm (used here) and its all-sample z-scores. INS returns no value in this profile, so any fixed background gene panel must report how many members the profile actually served.
The profile is labelled only mRNA expression / RNA expression levels; values are log2-CPM-like (cohort IRX1 median -3.31, ACTB median 11.69, PTPRC median 9.14). No DepMap-style release version is served, so the provenance is the study, not a dated release.
The other two cohorts checked the same way, for comparison
ccle_broad_2019- 16 samples with a KMT2A SV, of which 11 haematopoietic with a known partner: AFF1 (MV4-11, NALM-19, RS4;11, SEM), MLLT3 (MOLM-13, MONO-MAC-1, MONO-MAC-6, NOMO-1, THP-1), MLLT1 (KOPN-8), AFDN (OCI-AML2). The only resource found that supplies a KMT2A::AFF1 arm of more than one - and seeabsence-q149-no-kmt2a-rearranged-cell-line-in-ccle-expresses-irx1-including-all-four-aff1-linesfor why it still cannot serve as the AFF1 reference.aml_ohsu_2022(BeatAML, adult) - 53 KMT2A SV records over 53 samples, of which only MLLT3 is ever named (23 records); 30 areKMT2A rearrangementwith a blank site2 andsite1Position: -1. Partner-incomplete, and adult. Its expression profile also serves only 2 of 10 of this beat's background panel genes.
| study ID | aml_stjude_2024 |
|---|---|
| n samples | 887 |
| n kmt2a rearranged | 180 |
| n sv records for kmt2a | 237 |
| expression profile | aml_stjude_2024_mrna_seq_cpm |
| sv profile | aml_stjude_2024_structural_variants |
| crosscheck | MOLECULAR_CATEGORY KMT2Ar = 180 = the number of samples with a KMT2A SV (totals agree; per-sample identity unchecked) |
| gotcha site1 vs site2 | KMT2A appears as site1 in some records and site2 in others; taking site2 as the partner loses ~a third of calls |
| pmid | 38212634 |
| api | https://www.cbioportal.org/api |
| script | analyses/q149-is-irx1-expressed-in-non-aff1-kmt2a-r/run149.py |
| absence class | measured-not-analysed |
partner counts — the numbers behind the chart
{
"MLLT10": 57,
"MLLT3": 55,
"AFDN": 22,
"ELL": 17,
"MLLT1": 8,
"MLLT6": 5,
"MLLT11": 3,
"USP2": 3,
"AFF1": 1,
"EPS15": 1,
"UNSPECIFIED": 4,
"not-a-known-partner (DNAJC1, CORO1A, NEBL, SARNP)": 4
}clinical attribute coverage — the numbers behind the chart
{
"FAB_MORPHOLOGY": 787,
"MOLECULAR_CATEGORY": 887,
"WHO5_SUBCATEGORY": 199,
"DX_CYTOGENETICS": 837,
"ICC_SUBCATEGORY": 243
}genes absent from the expression profile
[ "INS" ]
measurements
[
{
"quantity": "n_kmt2a_r_cases_with_named_partner",
"subject": "aml_stjude_2024 KMT2A-rearranged arm",
"value": 172,
"units": "samples",
"n": 180,
"universe": "180 samples carry a KMT2A structural variant; 8 have a blank site2 or a partner that is not a known KMT2A fusion partner",
"null": null,
"note": "171 of the 172 are non-AFF1; the single AFF1 case is SJAML040667_D1 and is NAMED, NOT COMPUTED"
},
{
"quantity": "share_kmt2a_sv_calls_unusable",
"subject": "aml_stjude_2024 KMT2A SV calls",
"value": 0.0444,
"units": null,
"n": 180,
"universe": "8 of 180: 4 UNSPECIFIED site2 plus DNAJC1, CORO1A, NEBL, SARNP",
"null": null,
"note": "a floor on how often these calls are ambiguous; all eight are excluded and named rather than assigned"
}
]- Its 180 KMT2A-rearranged cases are 56% FAB M5, so any 'KMT2A-rearranged AML' statistic computed from it is dominated by monocytic disease. Treating it as representative of KMT2A-rearranged leukaemia generally imports that skew silently.
- Do the 180 samples with a KMT2A structural variant and the 180 labelled MOLECULAR_CATEGORY KMT2Ar contain the same patients, or do the totals agree by coincidence? A per-sample join would settle it and has not been done.
- What are SJAML030987_D1 (DNAJC1), SJAML040620_D1 (CORO1A), SJAML058299_D1 (NEBL) and SJAML040578_D1 (SARNP)? Four paediatric AML cases with a KMT2A breakpoint to a partner this corpus has never recorded.
- Does this cohort's DX_CYTOGENETICS free text agree with the SV partner calls? It is populated for 837 of 887 and would give an independent check on the partner assignment that nothing has run.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
887 paediatric AML cases were systematically categorised into 23 mutually distinct molecular categories covering 91.4% of the cohort, with categories associated with unique expression profiles and mutational patterns, including categories characterised by specific HOXA or HOXB expression signatures.
A new genomic framework to categorize pediatric acute myeloid leukemia · Umeda M, Ma J, Westover T, et al. · Nat Genet 56:281-293 2024What this evidence cannot show
Read from the abstract. The partner-level structural-variant table used here is the cBioPortal rendering of their calls, not a field they report in the paper - so the per-sample partner is theirs but its packaging is cBioPortal's, and a curation error would be invisible from the publication. Umeda et al. do not report IRX1. The KMT2Ar molecular category and the KMT2A SV set agree in TOTAL (180 = 180); per-sample identity was not checked.
Evidence that argues against this (1)
- A structural-variant call is not a validated fusion transcript. Eight of the 180 KMT2A-SV samples have a partner that is not a known KMT2A fusion partner (DNAJC1, CORO1A, NEBL, SARNP) or a blank site2, which is a 4.4% rate of calls this beat could not use - and gives a floor on how often the calls are ambiguous. Those eight were excluded and named rather than assigned. Umeda M et al. 2024
the cohort this analysis is computed on
the FAB field in this study is what exposed the composition artefact
180 KMT2A-rearranged paediatric AML cases with partner-level SV calls
- related to →KMT2A / MLL1 — the gene that breaks
237 KMT2A structural-variant records over 180 samples