ANSWERED from the paper's own methods, not from the corresponding author: the `dsgf` token that splits GSE139196's four SUV39H1 tracks into the two high and the two low ones is DSG-plus-formaldehyde double crosslinking - so the 44.5-percentile spread Q114 reported is a protocol variable, not replicate noise
also called what dsgf means · the crosslinker explains the spread · a deposit's undefined filename token is usually in the paper's methods
Two of four supposedly identical experiments were done with an extra chemical that glues proteins together before the DNA is captured, and that is why they gave much higher readings. The clue was a four-letter fragment of a filename that nobody had looked up.
The same thing, for a scientist
**The problem Q114 left open.** Four MCF-7 SUV39H1 ChIP-seq samples in GSE139196 read **90.81, 69.05, 96.29 and 51.78** at the same 146 bp *KMT2A* window - a **44.5-percentile spread**, and this corpus's only replicate estimate of that uncertainty. Q114 noticed that the token `dsgf` in the sample titles splits them **perfectly** into the two high tracks (`GSM4133857`, `GSM4133859`) and the two low ones, and recorded as an unresolved source: *"The meaning of the `dsgf` token in GSM4133857 and GSM4133859 is not given anywhere in the deposit... Not verifiable from GEO; would need the corresponding author."* Its memo listed "ask the corresponding author what `dsgf` means" as next step 10 [E1]. **It did not need the corresponding author.** The methods of the paper the SuperSeries is attached to say, verbatim: *"All of the ChIP experiments (except G9a ChIP samples prepared by double crosslinking with formaldehyde and **disuccinimidyl glutarate (DSG)**) were performed with formaldehyde crosslinking."* **`dsg` + `f` = DSG + formaldehyde.** It is the only use of DSG anywhere in the paper and the only expansion of those three letters that fits [E2]. **Why it matters beyond a filename.** DSG is a protein-protein crosslinker with an 7.7 A spacer arm; formaldehyde is a short zero-length protein-DNA crosslinker. Double crosslinking is the standard remedy for a factor that contacts chromatin **indirectly**, through another protein, and it routinely raises apparent occupancy for such factors by a large factor. So the correct reading of Q114's spread is **not** "four replicates of one antibody disagree by 44.5 percentile points"; it is **"two DSG-F samples read 90.81 and 96.29, two formaldehyde-only samples read 69.05 and 51.78, and the difference is the crosslinker."** Within protocol, the spreads are 5.5 and 17.3 points. **And it changes what the SUV39H1 numbers mean.** A protein whose signal at a site depends on a protein-protein crosslinker is a protein that is plausibly **not in direct contact with the DNA there**. This is consistent with, and not proof of, `hyp-q114`'s own alternative that SUV39H1's presence at the *KMT2A* locus is indirect or scaffolding rather than catalytic-at-the-site. **A discrepancy this entry does not resolve.** The paper's sentence names **G9a** as the DSG exception and does not mention SUV39H1. The deposit's sample titles carry `dsgf` on **two SUV39H1** samples and on **none** of the four G9a samples. Metadata and methods disagree about which antigen got the treatment, and neither source is obviously right.
The general rule, which is the reusable part
An undefined token in a deposited filename is usually defined in the paper's methods, and the paper is cheaper to read than the corresponding author is to email.
The tokens in GSE139196's file names are the submitter's own internal shorthand - Fan67, S45, L004, dsgf, srm, dir, ucsc. Most are pipeline noise. dsgf was the one that partitioned the data, and Q114 correctly spotted the partition and correctly declined to guess. The step it missed was: grep the linked paper's methods for the token's expansion before declaring it unverifiable.
The operational version:
- When a filename token splits a set of samples along the axis you are measuring, it is an experimental variable, not a serial number. Q114's partition was perfect - 2 high, 2 low, no exceptions - which is what a variable looks like and not what a label looks like.
- Resolve it against the linked publication's methods, via the SuperSeries'
Series_pubmed_id, before recording it as unverifiable. Europe PMC'sfullTextXMLreturned HTTP 500 for this paper; the PMC HTML route worked, which is the third route this project's brief already documents. - Re-read every number computed across that partition. A spread that is really a protocol contrast is not an uncertainty estimate, and quoting it as one overstates the noise in every other single-track measurement in the same deposit.
What Q114's four SUV39H1 numbers become
| sample | title token | crosslinking, by the dsgf reading | exon-11 percentile (Q114) |
|---|---|---|---|
| GSM4133857 | SUV39h1_minusE2_dsgf_Fan67 | DSG + formaldehyde | 90.81 |
| GSM4133859 | SUV39h1_plusE2_dsgf_Fan68 | DSG + formaldehyde | 96.29 |
| GSM4133858 | SUV39h1_minusE2_Fan66 | formaldehyde only | 69.05 |
| GSM4133860 | SUV39h1_plusE2_Fan65 | formaldehyde only | 51.78 |
Within-protocol spread: 5.48 points (DSG-F) and 17.27 points (formaldehyde-only). Across protocol: ~33 points between the two pair means. Q114's headline 44.5 is the range across a mixed set.
This does not rescue the anchor result. Q114's verdict was NEITHER, driven by the six-non-anchor-interval clause - the whole 90 kb gene body sits at the 81st-87th percentile in all four tracks, DSG-F or not - and nothing here changes that. What changes is the interpretation of the spread, and the standing caution this corpus should attach to any single ChIP track: check whether the deposit's filenames encode a protocol variable before treating replicate disagreement as noise.
| token | dsgf |
|---|---|
| expansion | disuccinimidyl glutarate (DSG) plus formaldehyde - double crosslinking |
| source of the expansion | PMID 34194047 ChIP-seq methods, read via PMC8439151 on 2026-09-23 |
| across protocol spread of pair means percentile points | 33.14 |
| q114 headline spread now partly explained | 44.51 |
| question this closes | docs/memos/2026-09-22-q114-suv39h1-at-the-locus.md, Missing/next item 10: 'Ask the corresponding author what dsgf means.' ANSWERED 2026-09-23 from the paper's own methods; no author contact needed. |
| residual discrepancy | The methods name G9a as the DSG exception; the deposit puts the token on two SUV39H1 samples and none of the G9a samples. Unresolved. |
| route note | Europe PMC fullTextXML returned HTTP 500 for PMC8439151; the PMC HTML route returned the full methods. This is the third route the standing brief documents, and it was the one that worked. |
| absence class | measured-not-searched |
| absence class note | The answer was in a public open-access methods section the whole time. The search space that excluded it was 'the GEO deposit', which is where the question was asked and where the answer is not. |
samples carrying the token
[ "GSM4133857", "GSM4133859" ]
samples not carrying it
[ "GSM4133858", "GSM4133860" ]
q114 exon11 percentiles with token
[ 90.81, 96.29 ]
q114 exon11 percentiles without token
[ 69.05, 51.78 ]
within protocol spread percentile points
{
"dsgf": 5.48,
"formaldehyde_only": 17.27
}- Which samples in GSE139196 were actually double-crosslinked? The methods say the G9a ChIPs were and the filenames say two SUV39H1 ChIPs were, and those two statements cannot both be complete. This is the one part that would need the corresponding author.
- If SUV39H1's apparent occupancy at the KMT2A locus depends on a protein-protein crosslinker, is the enzyme there at all, or is it crosslinked to something that is? A DSG-F versus formaldehyde-only comparison genome-wide in the same deposit would measure how much of SUV39H1's signal is DSG-dependent, and this beat did not run it.
- How many other single-track percentile values in this corpus come from a deposit whose filenames encode an unread protocol variable? Nothing has swept for that, and the cost of the sweep is one grep per deposit.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Q114 recorded the dsgf token as splitting GSE139196's four SUV39H1 tracks perfectly into the two high-percentile and two low-percentile tracks (90.81/96.29 versus 69.05/51.78 at the KMT2A exon-11 window), and recorded its meaning as not given anywhere in the deposit and requiring the corresponding author.
docs/memos/2026-09-22-q114-suv39h1-at-the-locus.md and ana-q114-the-four-mcf7-suv39h1-tracks-barely-agree-with-each-other-and-estradiol-changes-nothing · Q114-suv39h1-at-the-locus · oncofusion-knowledge-graph 2026What this evidence cannot show
The four percentile values are quoted from Q114's entries and memo and were not recomputed by this beat; this beat did not open the four SUV39H1 bedGraphs. The partition itself (which two samples carry the token) was re-verified against the live GEO sample titles on 2026-09-23.
Evidence that argues against this (1)
- With n = 2 per protocol arm, a perfect partition of four numbers into two pairs has a 1-in-3 chance of arising from any random split, so the partition alone is not evidence that the crosslinker caused the difference. The evidence that dsgf MEANS DSG-formaldehyde is the paper's methods sentence; the evidence that it CAUSED the spread is only the partition, and that is weak. source
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The paper attached to GSE139199, the SuperSeries of GSE139196, states in its ChIP-seq methods: 'All of the ChIP experiments (except G9a ChIP samples prepared by double crosslinking with formaldehyde and disuccinimidyl glutarate (DSG)) were performed with formaldehyde crosslinking.' This is the only occurrence of DSG in the paper.
Shape of promoter antisense RNAs regulates ligand-induced transcription activation · Yang F, Tanasa B, Micheletti R, Ohgi KA, Aggarwal AK, Rosenfeld MG · Nature 2021What this evidence cannot show
Full text read via the PMC HTML route (PMC8439151) on 2026-09-23 after Europe PMC's fullTextXML endpoint returned HTTP 500 for this article. The sentence is quoted verbatim from the rendered methods section. GSE139199's Series_pubmed_id field was read live from GEO on 2026-09-23 and gives 34194047.
Evidence that argues against this (1)
- THE METHODS AND THE METADATA DISAGREE ABOUT WHICH ANTIGEN. The sentence names G9a as the DSG exception; the deposit puts the dsgf token on two SUV39H1 samples (GSM4133857, GSM4133859) and on NEITHER of the four G9a samples (GSM4133792-95, titles Fan83/Fan91/Fan84/Fan92). So either the methods sentence is incomplete, or the filename token does not track the protocol the way this entry assumes. The token's EXPANSION is well supported; its per-sample APPLICATION is not, and that is why this entry's confidence is moderate rather than strong. source
the 44.5-percentile spread that entry reports as replicate disagreement is substantially a crosslinker contrast: within protocol the spreads are 5.48 and 17.27 points
resolves the meaning of the dsgf token in two of that experiment's four samples, from the linked paper's methods
a protocol variable encoded in that deposit's filenames, unread until now, that partitions one antigen's tracks
the paper's methods name G9a as the DSG-crosslinked antigen, so the four G9a tracks read here may carry the same protocol difference the deposit's filenames do not record