GSE252378 labels one of its two SEM AFF1 libraries KMT2A-AFF2, a fusion that does not exist in the deposit and does not exist in SEM - a filter on that field silently halves the SEM data
One sample in a public dataset is tagged with the wrong fusion name because of a single-character typing error. Anyone filtering the data by that tag would lose half the cell-line measurements without noticing.
The same thing, for a scientist
`GSM8371032`, titled `SEM_AFF1_2`, carries `group: KMT2A-AFF2` in `!Sample_characteristics_ch1`. Its partner `GSM8371031` (`SEM_AFF1_1`) carries `group: KMT2A-AFF1`, and so does every other KMT2A::AFF1 library in the series [E1]. **SEM is a KMT2A::AFF1 (t(4;11)) line**; `KMT2A-AFF2` appears nowhere else in the 171 sample records and is not among the fusions the paper profiled [E2]. **Why it is worth an entry rather than a footnote.** A `group == "KMT2A-AFF1"` filter - the obvious way to select the index-fusion samples from this deposit - returns **one** of the two SEM AFF1 libraries. The two libraries are two different antibodies, and this beat's central reagent finding is that they disagree (Spearman 0.885 in SEM, 0.712-0.784 in patients). So the filter does not merely lose depth, it **silently selects one reagent's view of the cell line**, which is precisely the choice this beat argues must be made explicitly. **The same field also reads `pedatric` for `pediatric`** in both SEM records, so a free-text match on `pediatric` misses SEM entirely. **This is a defect report about a third-party deposit, not about this corpus.** It is recorded so that any future beat selecting samples from GSE252378 by metadata field is warned, and so that a sample count derived from that field is not trusted.
The general shape
A one-character error in a categorical metadata field behaves like a missing record, not like a corrupt one: the filter succeeds, returns fewer rows, and reports no problem. The defence is to count what a filter returned against what you expected before using the result - here, two SEM AFF1 libraries are documented in the supplementary file listing and one comes back from the field.
| script | analyses/q142-aff1-in-primary-infants/extract.py |
|---|---|
| status | run |
| result summary | 1 of 2 SEM AFF1 libraries mislabelled KMT2A-AFF2; a group-field filter returns half the SEM data and one reagent. |
- Should this corpus keep a standing list of known metadata defects in the deposits it reuses, as it does for symbol collisions?
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
GSM8371032 (SEM_AFF1_2) is annotated group: KMT2A-AFF2 in GSE252378 while GSM8371031 (SEM_AFF1_1) is annotated KMT2A-AFF1, and no other sample in the 171-record series uses KMT2A-AFF2.
GSE252378 - AutoCUT&RUN and AutoCUT&Tag profiles of 36 MLL-rearranged leukemia samples (171 GSMs, hg38) · Janssens DH, Henikoff S (Fred Hutchinson Cancer Center / Van Andel Institute) · NCBI Gene Expression Omnibus 2024What this evidence cannot show
Read from the series matrix transposed column-wise, so the value is attributed to the correct accession. Reported as observed; not communicated to the depositors by this beat.
Evidence that argues against this (1)
- AFF2 (FMR2) is a real AF4-family gene and KMT2A::AFF2 fusions have been described, so the label is not impossible on its face - but SEM's t(4;11) KMT2A::AFF1 identity is established and the sample's own title and description both say AFF1, which is why this is read as a typographical error rather than as a claim. Janssens DH et al. 2024
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The source paper profiled the SEM cell line as a KMT2A::AF4 (AFF1) model alongside three patient samples bearing the same translocation.
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
From the results section of PMC11522475.
Evidence that argues against this (1)
- The paper does not enumerate GSM accessions in its text, so the mapping from its SEM figures to these two accessions comes from GEO, the same source that carries the defect. Janssens DH et al. 2024
metadata defect in the source deposit
one of SEM's two AFF1 libraries is mislabelled
second labelling defect found in the same deposit, by a different beat