GSE68720 - 97 infant ALL patients from COG P9407 with the KMT2A partner named per sample (mll_par) and a GPL570 expression matrix, 74 partner-named KMT2A-r cases and 17 germline controls
also called COG P9407 infant ALL GEP · GSE68720 · Kang 2012 infant ALL array
A public dataset of 97 babies with leukaemia that records, for each child, both which partner gene the KMT2A fusion joined and the activity of every gene. It is the only such dataset we could find in this disease, and it is downloadable in one 8 MB file.
The same thing, for a scientist
GEO series **GSE68720**, the expression arm of Children's Oncology Group trial **P9407**, published as PMID 22210879 [E1]. **97 primary infant ALL patients** on Affymetrix HG-U133 Plus 2.0 (**GPL570**), one 8,330,634-byte `GSE68720_series_matrix.txt.gz` that carries the sample metadata **and** the data table in the same file - no join required. **Why it is load-bearing for this corpus:** it is the only public resource found that carries a per-sample KMT2A **fusion partner** beside per-sample expression in **B-lineage** disease. Fifteen characteristics keys per sample; the partner lives in three of them: - **`mll_par:`** - the authoritative field, populated for all 97. - `gene:` - a duplicate of `mll_par`. - `pg:` - populated for only 9 samples, the rare and unknown partners. - `karyotype:` - free text, an independent second handle (`46,XX,T(4;11)(Q21;Q23)` etc.), but `--` in 29 of 97. **The partner tally, summing exactly:** | `mll_par:` | current symbol | n | |---|---|---:| | AF4 | **AFF1** | 48 | | ENL | **MLLT1** | 16 | | AF9 | **MLLT3** | 6 | | EPS15 | EPS15 | 3 | | ASAH3 | retired symbol, **not resolved** | 1 | | `NOT AF4, AF9 OR ENL; PARTNER UNKNOWN` | - | 5 | | `NOT AF4; PARTNER UNKNOWN` | - | 1 | | `--` | KMT2A germline | 17 | **Cross-checked against the independent `mll_ind:` field: 80 `MLL_R` + 17 `MLL_G` = 97, and the seven rearranged categories sum to exactly 80.** The two fields agree and the 17 `--` are exactly the 17 `MLL_G`. **So the design is 74 partner-named KMT2A-r infants across five partners, 6 partner-unknown, and 17 KMT2A-germline infant ALL controls in the same cohort on the same array** - the control arm being the part that makes any rate interpretable. **It is invisible to the current gene symbol.** GEO DataSets `infant acute lymphoblastic leukemia AND AF4` returns it; `... AND KMT2A` does not, and the string `AFF1` appears nowhere in its metadata.
What it can and cannot support
Can: partner-stratified single-gene expression calls in infant B-lineage ALL with a within-cohort KMT2A-germline null; a 2x2 between two genes' thresholded calls; the AFF1-versus-non-AFF1 contrast at n = 48 versus 26.
Cannot:
- Isoform, allele or fusion-transcript resolution. One array, one probe per gene for IRX1 (
230472_at). - HOXA9 specifically. GPL570 has no single-gene HOXA9 probe; both are annotated
HOXA10-HOXA9///MIR196B///HOXA9. HOXA10 (213147_at,213150_at) is the available single-gene proxy. - Any occupancy, chromatin or protein statement. It is transcript only.
- Outcome analysis. The EFS variable that Kang et al. modelled is not in the GEO record, so the link between the IRX1-high subgroup and survival cannot be drawn from the deposit.
- Purity correction. No blast count or immunophenotype per sample, so a bulk expression call cannot be attributed to blasts.
- Groups below n = 5. EPS15 (3), ASAH3 (1) and the partner-unknown categories must be named, not computed.
ASAH3 is left unresolved deliberately. It is a retired symbol and resolving it from recall is exactly the error this project logs repeatedly; it is reported as the depositors wrote it, with n = 1.
How to use it, exactly
https://ftp.ncbi.nlm.nih.gov/geo/series/GSE68nnn/GSE68720/matrix/GSE68720_series_matrix.txt.gz 8,330,634 B
https://ftp.ncbi.nlm.nih.gov/geo/platforms/GPLnnn/GPL570/annot/GPL570.annot.gz 8,471,521 B
A trap in the annotation file, which cost this beat a run. GPL570.annot.gz is a SOFT file: it opens with ^Annotation, then !-prefixed and #-prefixed lines, and the real column header is the row whose first field is exactly ID (columns ID, Gene title, Gene symbol, Gene ID, UniGene title). Taking the first non-# line as the header yields a single-element header ['^Annotation'], an empty probe map, and a KeyError on IRX1 - which reads as IRX1 is not on this platform and is not. A failed check is a claim too.
Multi-gene probes must be dropped and counted. Twelve of the probes for this beat's gene set are multi-gene: eight ACTG1///ACTB, two HOXA10-HOXA9///MIR196B///HOXA9, MYH7///MYH6, HOXA10-AS///MIR196B. A probe annotated to several genes cannot be attributed to one.
Values are log2 (cohort ACTB median 13.47, PTPRC median 10.48, silent-tissue genes 3.46-6.20).
| accession | GSE68720 |
|---|---|
| platform | GPL570, Affymetrix HG-U133 Plus 2.0 |
| n samples | 97 |
| trial | Children's Oncology Group P9407 |
| pmid | 22210879 |
| bioproject | PRJNA283535 |
| matrix file | GSE68720_series_matrix.txt.gz |
| matrix bytes | 8330634 |
| annotation file | GPL570.annot.gz |
| annotation bytes | 8471521 |
| partner field | mll_par |
| irx1 probe | 230472_at |
| hoxa9 probe status | NO single-gene HOXA9 probe on GPL570; both are HOXA10-HOXA9///MIR196B///HOXA9 |
| found only via retired symbol | AF4 |
| soft header trap | the real column header is the row whose first field is exactly 'ID'; taking the first non-# line gives ['^Annotation'] and an empty probe map |
| script | analyses/q149-is-irx1-expressed-in-non-aff1-kmt2a-r/gse68720.py |
| absence class | measured-not-analysed |
partner fields secondary
[ "gene", "pg", "karyotype" ]
partner counts — the numbers behind the chart
{
"AF4": 48,
"ENL": 16,
"AF9": 6,
"EPS15": 3,
"ASAH3": 1,
"NOT AF4, AF9 OR ENL; PARTNER UNKNOWN": 5,
"NOT AF4; PARTNER UNKNOWN": 1,
"--": 17
}independent crosscheck
{
"mll_ind_MLL_R": 80,
"mll_ind_MLL_G": 17,
"sum": 97,
"rearranged_categories_sum": 80,
"agrees": true
}hoxa10 probes
[ "213147_at", "213150_at" ]
measurements
[
{
"quantity": "n_partner_named_kmt2a_r_cases",
"subject": "GSE68720 usable partner-stratified design",
"value": 74,
"units": "patients",
"n": 97,
"universe": "mll_par populated for all 97; 74 name a partner, 6 are partner-unknown, 17 are KMT2A germline",
"null": 17,
"note": "the 'null' column here is the within-cohort KMT2A-germline control arm size, not a rate"
},
{
"quantity": "matrix_size",
"subject": "GSE68720_series_matrix.txt.gz",
"value": 8330634,
"units": "bytes",
"n": 1,
"universe": "NCBI GEO FTP, fetched 2026-09-24",
"null": null,
"note": "carries metadata AND data table in one file; MILE's equivalent is 886 MB and was not downloaded"
}
]- Every partner-stratified expression claim this corpus now makes in B-lineage KMT2A-r disease traces to this one deposit. If its `mll_par` field is wrong for any sample, the claim moves with it, and no second cohort exists to catch that.
- Can the P9407 outcome variable be obtained, so the IRX1-high subgroup can be tested against event-free survival? Kang et al. modelled EFS on these exact patients and the deposit does not carry it.
- What is `ASAH3` in the mll_par field? One patient, a retired symbol, and resolving it from recall is the error this project logs most often.
- Do the 29 samples with `karyotype: --` differ systematically from the 68 with a karyotype? If karyotype availability tracks institution or era, it is a covariate nobody has checked.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
GSE68720 is the gene expression profiling of 97 infant ALL cases from Children's Oncology Group Trial P9407, from which a three-gene outcome predictor (FLT3, IRX2, TACC2) and a seven-gene MLL-AFF1 classifier were derived.
Gene expression profiles predictive of outcome and age in infant acute lymphoblastic leukemia: a Children's Oncology Group study · Kang H, Wilson CS, Harvey RC, et al. · Blood 119:1872-1881 2012What this evidence cannot show
Read from the abstract; the supplementary tables and the outcome variable were not retrieved. The deposit post-dates the paper by three years (GSE accession 2015, paper 2012), so the deposited matrix is a later release of the same data and the exact normalisation used in the publication was not confirmed against the deposit. 2012 array technology: no isoform resolution, single probe for IRX1, and a hybridisation background rather than a true zero.
Evidence that argues against this (1)
- P9407 is one trial from one era with one treatment backbone, so any subgroup frequency measured in it is a frequency in that trial's accrual, not in infant ALL generally. The 48 AFF1 / 16 MLLT1 / 6 MLLT3 partner mix is roughly the published partner distribution for infant ALL, which is reassuring, but 6 cases of MLLT3 and 3 of EPS15 cannot represent those partners. Kang H et al. 2012
the cohort this analysis is computed on
the cohort this 2x2 is computed on
97 infant ALL patients, COG P9407
- related to →KMT2A / MLL1 — the gene that breaks
per-sample KMT2A partner in the mll_par field, 74 partner-named cases
- related to →FLT3 in KMT2A-r leukemia — near-universal overexpression, but activating mutation in only ~3% of infants
the same Kang 2012 cohort this corpus already cites for FLT3; its IRX2 result and its GEO deposit were never used
the cohort this 2x2 is computed on
the cohort this control arm comes from
re-derived from the same cached GSE68720 series matrix and GPL570 annotation