Oncofusion

start / graph / ana-q125-six-files-titled-h2ak119ub-in-gse260742-name-a-flag-chip-in-their-own-tdf-header

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q125-the-h2ak119ub-arm-of-the-pcgf1-knockout

Six GSE260742 samples titled 'NTvKO ... H2AK119ub rep1' carry ChIP_FLAG filenames AND a ChIP_FLAG string in the TDF header their own depositors' igvtools wrote - two of three independent sources say FLAG, the shape test is near-indecisive, and the channel of half the replicate arm cannot be settled from the deposit

also called the rep1 channel problem · GSM8123407-412 · a title/filename/header three-way disagreement

Six public files are described in one place as measuring a silencing mark and in two other places as measuring the fusion protein itself. Which is right decides whether half of a replicate set is usable, and whether a completely different question becomes answerable. The data itself does not clearly say.

The same thing, for a scientist

**GSM8123407-412 are titled *'NTvKO PCGF1 KO / Control - DMSO / SGC0946 5uM / VTP50469 500nM H2AK119ub rep1'*.** Their `!Sample_supplementary_file_1` basenames are `ChIP_FLAG_CLONE1_DMSO.tdf`, `ChIP_FLAG_NT_DMSO.tdf` and so on - `FLAG`, in all six. The other two conventions say `K119UB` [E1]. **A third and independent source agrees with the filename.** Every TDF carries the originating BAM's name in a header string written by the depositors' own `igvtools` run. For these six it reads `[ChIP_FLAG__CLONE1_DMSO.filt.bam Combined Strands]` - and it is not a copy of the GEO filename, because it differs from it in its underscores and carries a `.filt.bam` suffix the GEO name does not [E1]. **The shape test, pre-registered as the second corroborant, is very nearly indecisive** — and decisive for the other convention, which is what makes its indecision meaningful [E2]: | DMSO track | Spearman vs `paired` H2AK119ub | Spearman vs FLAG | margin | share of own coverage in `S_FLAG` | |---|---:|---:|---:|---:| | `rep1` NT | 0.5785 | 0.5170 | **+0.061** | 0.008392 | | `rep1` C1 | 0.4825 | 0.4575 | **+0.025** | 0.006830 | | `rep2` NT | 0.7833 | 0.5235 | +0.260 | 0.002646 | | `rep2` C1 | 0.8622 | 0.5906 | +0.272 | 0.002404 | and on concentration `rep1` sits **between** the two references rather than at either: 0.0068-0.0084 against **0.0019-0.0026 for H2AK119ub** and **0.0035-0.0059 for the two FLAG tracks that were not used to define the set** [E2]. **The pre-registration said that if the two corroborants disagreed, `rep1` would be dropped and the disagreement would be the finding. It is dropped**, and it is NOT relabelled FLAG either, because the consequence of a FLAG classification is large enough to need better evidence than a contradicted metadata field [E3].

Go deeper
Why it matters more than a labelling nuisance

If the title is right, this arm has three replicate conventions and the spread reported elsewhere in this cycle should have been computed over three rather than two.

If the filename and the header are right, GSE260742 deposits a FLAG ChIP in BOTH genotypes of the PCGF1-knockout experiment under all three drugs - and hyp-q117's prediction 5 clause (a), 'the fusion is evicted normally', which Q124 declared UNREACHABLE for want of a FLAG, MLL1 or MENIN track in the NTvKO arm, becomes reachable. The eviction clause and the silencing clause could then be scored in the same genotypes, in the same experiment, for the first time. That is the largest single thing this deposit could still yield and it turns on which of three metadata fields is believed.

What would settle it, cheaply

The SRA read metadata. Each GSM links to an SRX/SRR whose library name and submitter-supplied attributes are a fourth source, independent of the title, the processed filename and the TDF header. Q124's route note records that elink pubmed->sra is unusable as evidence of absence in this series, but a direct elink gds->sra on these six accessions has not been tried.

A fragment-length or peak-width statistic on re-aligned reads. A FLAG ChIP of an over-expressed transcription factor and a broad repressive-mark ChIP have very different peak widths, and the difference does not depend on which reference set is used. This needs the FASTQs and an aligner.

Asking the depositors. The corresponding author of PMID 41634341 is the cheapest route of all and this project has not used it for anything.

Facts and figures
scriptanalyses/q125-the-h2ak119ub-arm-of-the-pcgf1-knockout/run.py
statusrun
verdictUNRESOLVED - dropped from every H2AK119ub criterion, NOT relabelled FLAG
absence classmeasured-not-searched
absence class noteThe channel of GSM8123407-412 is determinable in principle and was not determinable from the routes searched: the GEO SOFT records (all 419,824 characters, in which `antibod` occurs zero times), the processed filenames, and the TDF headers. THREE routes remain untried and are named in the entry: the SRA library metadata for these six accessions via elink gds->sra, a peak-width statistic on re-aligned reads, and asking the corresponding author of PMID 41634341. The first costs one request.
absence class assigned2026-09-23 (Q125)
accessions
[
  "GSM8123407",
  "GSM8123408",
  "GSM8123409",
  "GSM8123410",
  "GSM8123411",
  "GSM8123412"
]
sources and what each says
{
  "!Sample_title": "H2AK119ub rep1",
  "!Sample_supplementary_file_1": "ChIP_FLAG_{NT,CLONE1}_{DMSO,SGC,VTP}.tdf",
  "TDF-internal trackNames": "[ChIP_FLAG__{NT,CLONE1}_{DMSO,SGC,VTP}.filt.bam  Combined Strands]",
  "shape test": "H2AK119ub, by margins of +0.062 and +0.025"
}
How this could bite us back
  • Six of eighteen files in the best-replicated arm of this series have a contested channel, and the contamination is in the depositors' own processing chain rather than in a GEO web form. Any other beat reusing this series should read the TDF header of every file it downloads and compare it with the title, which costs nothing because the binner already prints it.
Still open
  1. Does elink gds->sra on GSM8123407-412 return a library name that settles the channel? One request, never tried for these accessions.
  2. If they are FLAG, does prediction 5's eviction clause hold in the PCGF1 knockout? That is the measurement this deposit would then contain and nobody has done, and it would complete two of prediction 5's three clauses in one experiment.
Every source behind this entry (3)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    GSM8123407, 8123408, 8123409, 8123410, 8123411 and 8123412 of GSE260742 are titled 'NTvKO ... H2AK119ub rep1' and carry supplementary filenames beginning ChIP_FLAG_ (CLONE1 or NT), and the TDF-internal trackNames string written by the depositors' igvtools run reads '[ChIP_FLAG__CLONE1_DMSO.filt.bam Combined Strands]' and equivalents - a string that differs from the GEO filename in its underscores and its .filt.bam suffix and is therefore not a copy of it. The other twelve H2AK119ub-titled files in the arm carry K119UB in both places.

    What this evidence cannot show

    The titles and filenames are read from the live SOFT records, re-fetched this session and byte-identical to Q124's copy. The TDF-internal string is read from the header of each downloaded file by the binner. 'Not a copy of the GEO filename' is an inference from the two strings differing, not from knowledge of the deposition pipeline. `antibod` occurs zero times in all 419,824 characters of the 158 sample records, so no GEO field names a reagent for any sample in this series and the title is the only place the antibody is asserted.

    Evidence that argues against this (1)
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    The pre-registered shape test classifies rep1 as H2AK119ub by margins of only +0.062 and +0.025 in Spearman over 1,227,708 bins, against margins of +0.260 and +0.272 for the rep2 convention, and rep1's share of its own coverage in S_FLAG (0.0068-0.0084) sits between the H2AK119ub references (0.0019-0.0026) and the two selection-unbiased FLAG references (FLAG_VTP 0.003541, FLAG_SGC 0.005888).

    What this evidence cannot show

    The first version of this test used q117_bin_FLAG_DMSO as the FLAG reference for the concentration comparison, which is SELECTION-BIASED because S_FLAG is defined from that very track (its own share there is 0.019751). The unbiased comparison uses FLAG_SGC and FLAG_VTP and is the one reported. The Spearman comparison is unaffected because it is computed over the whole universe. A correlation margin of 0.025 over 1.2 million bins has no stated error and no null was computed for it, so 'near-indecisive' is a qualitative reading of a small margin, not a test.

    Evidence that argues against this (1)
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    This beat's PREREGISTRATION.md section 2a, written before any file was downloaded, declared both readings, declared the two corroborants, and declared that if they disagreed rep1 would be dropped from every H2AK119ub criterion and the disagreement recorded as the finding, with the consequence that a two-convention range becomes a difference and criterion S3 becomes UNSCOREABLE.

    What this evidence cannot show

    The pre-registration is a file in this repository; its timing rests on a recorded check that C:/Temp/q125 contained no bin file and on git history, not on an external timestamp.

    Evidence that argues against this (1)
Connected to