GEO labels both aliquots of the lineage-switching patient 'treated with the Menin inhibitor', but the paper says 148752 is the B-ALL at diagnosis and only 152985 arose during treatment - the deposit's drug flag is a case label wearing an aliquot label's clothes
A public dataset marks two samples from one child as having been taken during drug treatment. The paper says the first was taken at diagnosis, before any drug. Anyone comparing the two as treated-versus-treated would be comparing before with after and not know it.
The same thing, for a scientist
**What the deposit says.** `!Sample_source_name_ch1` and `!Sample_description` carry the string *"treated with the Menin inhibitor"* on **all ten** GSMs of sample `148752` and **all ten** of `152985`, with no qualifier distinguishing them [E1]. **What the paper says.** *"We compared the B-ALL and AML patient-matched samples (designated 148752 and 152985, respectively)"*, and *"while 46 oncoprotein-binding sites were detectable in the B-ALL sample at diagnosis, treatment with the menin inhibitor reduced..."*, and *"the 152985 sample that lineage switched during treatment with the menin inhibitor"* [E2]. **148752 is the pre-treatment diagnostic aliquot.** **So the GEO string is a property of the CASE** - this child was treated with a menin inhibitor and lineage-switched on it - **and it has been written onto the aliquot that precedes the treatment.** **The direction of the error matters.** A re-analysis working from the deposit alone, as this beat's own pre-registration did when it was written, will classify 148752 as treated. Doing so does two things at once: it removes the only pre-treatment aliquot the deposit contains, and it adds an untreated sample to the treated group. A drug comparison built on that assignment is not merely underpowered - **its two groups are mislabelled in opposite directions**, which biases any difference towards zero. **And it is not visible from inside the deposit.** There is no field that disagrees, no date, no `treatment:` characteristic, and no vehicle arm to notice is missing. The only way to find it is to read the paper - which is the general lesson: **a per-sample treatment flag in GEO is free text, and free text written once per case propagates to every aliquot of that case.**
What this rescues
It also makes the deposit better than it looked. Read correctly, GSE252378 contains a matched pre- and post-menin-inhibitor pair from one patient, with KMT2A N1/N2/C1/C2, ENL, DOT1L and menin CUT&Tag on both - the only menin chromatin measurement in the deposit. That is not a vehicle-controlled arm, because the two aliquots differ in lineage (B-ALL vs AML) as well as in drug exposure, and the confound is total. But it is a great deal more than 'two treated samples with no control', which is what the deposit's own annotation says it is.
Corpus cross-check
A sibling beat in this corpus (ana-q143-menin-sits-with-the-fusion-n-terminus-and-not-with-the-wild-type-c-terminus-in-a-drug-naive-infant-b-all) already treats 148752 as "the diagnostic, pre-revumenib aliquot", reading the paper rather than the deposit, and is correct. This entry records the discrepancy so that the deposit's own string is not trusted the next time, and so the contradiction between two corpus entries that read different sources is on the record rather than latent.
| script | analyses/q144-unbounded-magnitude/run.py |
|---|---|
| status | run |
| result summary | GEO's menin-inhibitor string is case-level; 148752 is the pre-treatment diagnostic B-ALL aliquot and 152985 is the post-treatment AML. |
| n gsms affected | 20 |
affected samples
[ "148752", "152985" ]
- A pharmacodynamic comparison that puts an untreated diagnostic aliquot into the treated arm shrinks the measured drug effect towards zero, so a drug that works can be scored as not working - and nothing in the deposit signals the mislabel.
- How many other GEO deposits this corpus relies on carry a treatment string in !Sample_source_name_ch1 that is a case-level rather than an aliquot-level fact?
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
All twenty GSMs belonging to samples 148752 and 152985 in GSE252378 carry 'treated with the Menin inhibitor' in !Sample_source_name_ch1 and !Sample_description, with no field distinguishing pre- from post-treatment.
GSE252378 - KMT2A oncoproteins induce epigenetic resistance to targeted therapies (171 GSMs, AutoCUT&RUN / AutoCUT&Tag, hg38 fragment BEDs and bigWigs) · Janssens DH, Henikoff S (Fred Hutchinson Cancer Center / Van Andel Institute) · NCBI Gene Expression Omnibus 2024What this evidence cannot show
Read from every record of the series matrix. The deposit does carry the reciprocal 'paired with' statements that identify the two as one patient; it is only the treatment status that is undifferentiated.
Evidence that argues against this (1)
- A defender of the annotation could argue 'treated with the Menin inhibitor' is a correct description of the case and was never intended as an aliquot-level flag. That is exactly the reading this entry arrives at - but GEO's source-name field is per sample, so a per-case string placed there is indistinguishable from a per-sample one for any automated reader. Janssens DH et al. 2024
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The paper describes 148752 as the B-ALL sample at diagnosis and 152985 as the patient-matched AML that arose during menin-inhibitor treatment.
MLL oncoprotein levels influence leukemia lineage identities · Janssens DH, Duran M, Otto DJ, Wu W, Xu Y, Kirkey D, Mullighan CG, Yi JS, Meshinchi S, Sarthy JF, Ahmad K, Henikoff S · Nature Communications 15:9341 2024What this evidence cannot show
Quoted from the article text served by PubMed Central (PMC11522475). The interval between the 148752 draw and the first menin-inhibitor dose is not stated, so 'at diagnosis' is the paper's word and is not independently dated here.
Evidence that argues against this (1)
- The paper does not explicitly say 148752 was drawn before the first dose - 'at diagnosis' is a clinical descriptor, and diagnostic marrow is normally pre-therapy but the deposit's own string would be consistent with an on-therapy re-biopsy also being called diagnostic. The reading here is the natural one and is the same one an independent beat in this corpus reached, but it rests on a phrase rather than a recorded date. Janssens DH et al. 2024
the deposit whose treatment annotation this corrects
the menin inhibitor the case was treated with
confirms that beat's reading of 148752 as the drug-naive diagnostic aliquot against the deposit's own string
- related to →KMT2A::MLLT1 (MLL-ENL), t(11;19)(q23;p13.3)
the lineage-switching patient carries KMT2A::ENL
the corrected pre-/post-treatment labelling this result depends on
the corrected pre-/post-treatment labelling
independent confirmation from the paper text that 148752 is at diagnosis