GSM8123405 is mm10 DATA in an hg38-DECLARED TDF - the exact mirror of Q122's mm10-declared / hg38-coordinate defect, in the same series - and the container silently clipped 72.8 Mb (2.76%) of chr13-chr19 and chrX at deposition; the declared extents name the container and max_end names the data
also called the hg38 container · GSM8123405 is clipped · the mirror of Q122 DEVIATION 1
One of the deposited mouse files was packaged as if it were human. The mouse data inside is genuine, but wherever a human chromosome is shorter than the mouse one, the packaging cut the end off - about 3% of the genome, silently. A check that reads only the label calls the file human; a check that reads where the data actually stops calls it mouse.
The same thing, for a scientist
**One of 34 files failed this beat's pre-registered build check.** `GSM8123405` - *NTvKO Control - SGC0946 5uM H2AK119ub paired*, a `Mus musculus` / taxid 10090 sample - carries per-chromosome declared extents matching **hg38 to 63 bp over 20 chromosomes** against 382,695,718 bp for mm10, and its self-declared `genomeId` is `hg38` [E1]. **The data inside is mouse.** On chr1-chr12, where hg38 is longer than mm10, coverage stops at the **mm10** extent - chr1 `max_end` 195,371,600 against mm10's 195,471,971 and hg38's 248,956,422. A human track would run to 248.9 Mb. Its Spearman over `U_m` against its own culture's mouse H2AK119ub DMSO track is **0.8698** [E1]. **On chr13-chr19 and chrX, where hg38 is SHORTER than mm10, coverage stops at exactly the hg38 length.** chr13 at 114,364,328, chr19 at 58,617,616, chrX at 156,040,895 - the hg38 numbers to the base. **The container clipped the data at deposition and 72,812,699 bp, 2.76% of mm10 chr1-19+X, is not in the file.** Re-binning with a forced mm10 allocation returns a byte-identical total of 3,077,034,259.095, which proves the loss is in the deposit and not in the reader [E1]. **Materiality: none.** Rescored over the retained sub-genome (1,193,027 of 1,227,708 universe bins, 97.18%; 1,205 of 1,228 `S_FLAG` bins), `R_H2AK119ub(SGC/DMSO)` at `S_FLAG` moves from **1.4932 to 1.4940** (`PAIRED` convention; the `REP2` companion of the same statistic is **1.3855**, and both exceed 1, so the two conventions agree here) - 0.0008 - and every number involving this file is reported in both forms [E2].
Both directions of the same defect now exist in one series
Q122 found four K562 files in GSE260742 declaring mm10 and carrying hg38 coordinates, and bounded every K562 number to 'the retained 84.46% sub-genome with chr20-22 absent'. This is the same defect in the other direction: a mouse file declaring hg38 and carrying mm10 coordinates.
The two failures are not symmetric in consequence. Q122's direction loses whole chromosomes - chr20, chr21, chr22 have no mm10 counterpart so they vanish entirely, 15.5% of the genome. This direction loses the distal END of every chromosome where the wrong build is shorter - here 8 of 20 chromosomes, 2.76% of the genome, and nothing announces it because every chromosome is still present with data on it. The second is the harder one to notice, and the only reason it was noticed is that the build check offered four candidates instead of two.
What to change in the check, and it is not the threshold
1. Read max_end, not only the declared extent. The declared per-chromosome length is written by igvtools from whatever genome it was handed; max_end is the furthest coordinate that actually carries signal. The declaration is a property of the container and max_end is a property of the data, and when they name different builds, the disagreement is the finding. Q117 has recorded max_end in its output since cycle 68 and no beat has ever used it for the verdict.
2. The has_chr20 clause in this build check - and in Q124's, from which it was copied - is VACUOUS. Q117Bin iterates a hardcoded list of chr1-19+X and never asks any file for chr20, so has_chr20 is false by construction for every file that has ever been checked, in four beats. It has been reported as part of a passing verdict throughout. A check that cannot return the other answer is not a check, and this one has been carried as though it were.
3. Keep four candidate builds, not two. A two-candidate check of mm10 against mm39 would have called this file mm10 by default and never flagged it.
| script | analyses/q125-the-h2ak119ub-arm-of-the-pcgf1-knockout/deviation1_hg38_container.py |
|---|---|
| status | run |
| result summary | GSM8123405: declared hg38 (63 bp distance over 20 chromosomes), data mm10, clipped at the hg38 length on chr13-19 and chrX, 72.8 Mb / 2.76% lost, 97.18% of U_m retained, effect on the headline ratio 0.0008. |
| affected accession | GSM8123405 |
| bp lost | 72812699 |
| PCT lost | 2.76 |
| PCT of universe retained | 97.18 |
| absence class | measured-not-analysed |
| absence class note | The defect is visible in the deposited file's own header and has been public since 2025-11-23. It was not found earlier because no beat had binned GSM8123405 and because the build check in use reads the declared extents only. A sweep of the other 157 samples' declared extents against four candidate builds has NOT been run and would cost one HTTP HEAD plus one header read per file - it is the cheapest outstanding thing in this series. |
| absence class assigned | 2026-09-23 (Q125) |
- A 2.76% silent loss concentrated in the distal arms of eight chromosomes would be invisible to any genome-wide summary statistic and fatal to a locus-specific claim in that region. Any future entry naming a murine locus on the distal arm of chr13-19 or chrX from this file is wrong by construction.
- How many of GSE260742's other 157 samples declare the wrong build? Only the 34 binned by Q117, Q120, Q122, Q124 and this beat have ever had their headers read, and one of those 34 is wrong.
- Does the clip appear in the authors' own published figures? The paper's supplementary figures are image-only with zero extractable characters, so it cannot be checked here.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
GSM8123405 (GSE260742, 'NTvKO Control - SGC0946 5uM H2AK119ub paired', Mus musculus taxid 10090) declares genomeId hg38 and per-chromosome extents matching hg38 to 63 bp total over 20 chromosomes against 382,695,718 bp for mm10, while its coverage stops at the mm10 extent on chr1-chr12 (chr1 max_end 195,371,600) and at exactly the hg38 length on chr13-chr19 and chrX (chr13 114,364,328; chr19 58,617,616; chrX 156,040,895), losing 72,812,699 bp, 2.76% of mm10 chr1-19+X; re-binning with forced mm10 allocation returns a byte-identical total of 3,077,034,259.095.
Q125 re-analysis of GSE260742: criterion I1 and DEVIATION 1: per-chromosome declared extents and max_end for all 34 binned tracks, against live UCSC mm10, mm39, hg38 and hg19 sizes. analyses/q125-the-h2ak119ub-arm-of-the-pcgf1-knockout/run.py, deviation1_hg38_container.py, deviation2_quartiles_and_conventions.py, deviation3_matched_null.py and results.json, run 2026-09-23 · Iqbal Lab oncofusion knowledge graph 2026What this evidence cannot show
The chromosome sizes for all four candidate builds were fetched live from api.genome.ucsc.edu this session, and mm10 was additionally re-fetched and asserted equal to the stored copy on all 20 used chromosomes (0 mismatches). The inference that the file is MOUSE rests on two things - coverage stopping at the mm10 extent on the twelve chromosomes where hg38 is longer, and a Spearman of 0.8698 against a mouse track - and not on any read-level evidence; the FASTQ and BAM were not examined and are not deposited as .tdf.
Evidence that argues against this (1)
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Rescoring every quantity involving GSM8123405 over the retained sub-genome - 1,193,027 of 1,227,708 universe bins (97.18%) and 1,205 of 1,228 S_FLAG bins - moves R_H2AK119ub(SGC/DMSO) at S_FLAG from 1.4932 to 1.4940, a difference of 0.0008, so no verdict in this beat depends on the defect.
Q125 re-analysis of GSE260742: DEVIATION 1 rescoring over the retained sub-genome, both numbers reported for every affected quantity. analyses/q125-the-h2ak119ub-arm-of-the-pcgf1-knockout/run.py, deviation1_hg38_container.py, deviation2_quartiles_and_conventions.py, deviation3_matched_null.py and results.json, run 2026-09-23 · Iqbal Lab oncofusion knowledge graph 2026What this evidence cannot show
Immateriality here follows from the clipped regions being distal, gene-poor and containing 23 of 1,228 S_FLAG bins. It does not generalise: a set concentrated in the distal arms of chr13-19 or chrX would be badly affected, and any future analysis using this file must re-derive the retained mask rather than inherit this conclusion.
Evidence that argues against this (1)
the one genuine fault of three flagged hits - quoted 1.4932 without naming the PAIRED convention; amended to carry the REP2 companion 1.3855
the concrete case the general rule is drawn from
a third processing defect in the same series
the .tdf-only deposition is why the clip cannot be worked around