GSE252378's 27 KMT2A-fusion samples are 24 independent patients: three pairs are one patient sampled twice
A public dataset's 27 leukaemia samples come from 24 children, because three pairs are the same child sampled twice as the disease changed. Counting them as 27 independent patients overstates the evidence.
The same thing, for a scientist
Every per-sample statistic in this deposit is affected by the fact that **three of the 27 sample ids are pair members from one patient each**, stated in the GEO description of each in the words *'paired with'* [E1]: | pair | partner | states | institution | |---|---|---|---| | p279 / p318 | KMT2A-AFF1 | infant B-ALL / infant AML lineage switch | Texas Children's | | 148752 / 152985 | KMT2A-ENL | infant B-ALL diagnosis / infant AML relapse on revumenib | Texas Children's | | R3595 / R3652 | KMT2A-EPS15 | infant B-ALL lineage switch pair | Seattle Children's | **27 samples, 3 pairs, 24 independent patients.** The brief that dispatched this beat said 25, and the figure had propagated; it is 24 [E1]. The institutions are four: Texas Children's, St. Jude, Fred Hutch, Seattle Children's [E1]. **And the treatment annotation is worse than a miscount.** GEO carries the string *'treated with the Menin inhibitor'* on all records of BOTH 148752 and 152985, and this corpus has already established from the paper that **148752 is the drug-naive diagnostic aliquot** and only 152985 is revumenib-exposed. This beat's own pre-registration excluded both as treated, which is conservative but wrong, and the correction is recorded here against the pre-registration rather than hidden [E2]. Including 148752 as drug-naive changes the ENL group from n = 5 to n = 6 and **changes no conclusion**: IRX1 remains unoccupied in all six (0.000-0.246x), and the gene-scale Spearman moves from +0.100 to +0.200 [E2].
What the pairing does to every n in this deposit
Correlations in this beat are computed on samples and re-run with each pair member dropped, which is the only honest treatment at these n. For KMT2A::AFF1 the pair-dropped values are stronger than the full set (-0.857 and -0.821 against -0.7143), so the non-independence is not what produces the AFF1 result.
It also means three of the six partner groups contain a pair, and for EPS15 the pair IS the group - its n of 2 is one patient. So KMT2A-EPS15 in this deposit is a single patient's B-ALL and AML aliquots, and any statement about EPS15 from GSE252378 is a case report.
The pre-registration error, stated plainly
PREREGISTRATION.md section 0 reads: 'Two ENL samples (148752, 152985) are MENIN-INHIBITOR TREATED - the manifest description says so in words.' The manifest does say so; the manifest is repeating a GEO field that is a property of the CASE, not of the aliquot. This corpus recorded that two cycles ago in ana-q144-geos-treated-with-the-menin-inhibitor-string-is-a-case-label-and-it-is-attached-to-the-untreated-diagnostic-aliquot and ana-q143-148752-and-152985-are-one-patient-before-and-during-revumenib-not-two-treated-patients, and I did not read them before writing the pre-registration.
The general lesson is not 'read the paper' - q144 already read the paper. It is that a manifest built from a deposit inherits the deposit's labelling errors, and the correction may already be in the graph rather than in the source. Checking the graph for an entry about your own input file is as cheap as checking the input file.
| n samples | 27 |
|---|---|
| n independent patients | 24 |
| n pairs | 3 |
| brief said | 25 |
| actual | 24 |
| enl n5 rho | 0.1 |
| enl n6 rho | 0.2 |
pairs
[
[
"p279",
"p318"
],
[
"148752",
"152985"
],
[
"R3595",
"R3652"
]
]institutions
[ "Texas Children's Hospital", "St. Jude Children's Hospital", "Fred Hutch Cancer Center", "Seattle Children's Hospital" ]
revumenib exposed
[ "152985" ]
drug naive mislabelled as treated
[ "148752" ]
analyses
[ "ana-q147-irx1-fusion-occupancy-is-confined-to-kmt2a-aff1" ]
- Are any of the remaining 21 samples from the same patient across institutions? Nothing in GSE252378 would reveal it, so 24 is an upper bound.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
GEO descriptions in GSE252378 name three sample pairs as one patient each - p279/p318 (KMT2A-AFF1, Texas Children's), 148752/152985 (KMT2A-ENL, Texas Children's) and R3595/R3652 (KMT2A-EPS15, Seattle Children's) - using the phrase 'paired with' in every case. The 27 KMT2A-fusion samples are therefore 24 independent patients, drawn from four institutions: Texas Children's Hospital, St. Jude Children's Hospital, Fred Hutch Cancer Center and Seattle Children's Hospital.
GSE252378 - AutoCUT&RUN and AutoCUT&Tag profiles of 36 MLL-rearranged leukemia samples (171 GSMs, hg38 bigWigs and fragment BEDs) · Janssens DH, Ahmad K, Henikoff S · NCBI GEO 2024What this evidence cannot show
The pairing is taken from the deposit's own free-text descriptions, not from a structured patient identifier, because GEO has none here. If any two samples from different institutions were also the same patient, nothing in the deposit would say so - so 24 is an upper bound on independence, not a verified count.
Evidence that argues against this (1)
- The deposit paper describes 36 MLL-rearranged samples while this analysis uses the 27 with a full KMT2A-N/C antibody panel; the remaining records are AutoCUT&Tag and cell-line samples. So '27 samples' is a property of the arm selected, not of the deposit, and a different selection gives a different patient count. Janssens DH et al. 2024
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
This beat's pre-registration excluded both 148752 and 152985 as menin-inhibitor-treated on the strength of the GEO string; only 152985 is drug-exposed. Re-running the KMT2A-ENL group with 148752 included as drug-naive gives n = 6, IRX1 0.000-0.246x in all six (0 occupied), and gene-scale Spearman(HOXA, IRX1) = +0.200 (exact two-sided p = 0.714) against +0.100 (p = 0.950) at n = 5. No conclusion changes.
q147 re-analysis of GSE252378: HOXA and IRX1 oncoprotein occupancy across all six KMT2A fusion partners, 27 samples, 133 libraries (analyses/q147-hoxa-irx1-across-partners) · Iqbal Lab oncofusion knowledge graph, beat q147 · internal analysis over deposited GSE252378 bigWigs 2026What this evidence cannot show
The sensitivity re-run is the honest form: the pre-registration's assignment was wrong, the analysis was re-run under the correct assignment, and both numbers are reported. 148752's inclusion is the version used in the headline entry. Every number is a WITHIN-SAMPLE share: an interval's mean coverage divided by that same library's mean coverage over all 19,404 hg38 RefSeq promoters, so 1.0 = an average promoter in that library. GSE252378 has no spike-in, so no between-library absolute statement is licensed; the only cross-sample quantity used is a ratio of two loci measured inside one library, where the unknown per-library scale cancels.
Evidence that argues against this (1)
- The GEO field itself disagrees with this assignment - it labels 148752 treated - so anyone re-deriving from the deposit alone will get the pre-registration's answer. The correction rests entirely on the paper's text, which says 46 oncoprotein-binding sites were detectable in the B-ALL sample at diagnosis and that ENL was enriched in 148752 prior to treatment. Janssens DH et al. 2024