GSE260742's PCGF1-knockout samples carry no deposited measurement of PCGF1 - no protein, no transcript, no sequencing of the edited locus - and the knockout rests on one guide and one clone with no rescue, so every null result from that arm is a null in an unverified genotype
also called the knockout is unverified in the deposit · one guide, one clone, no rescue
The dataset labels six samples as coming from cells with PCGF1 deleted, but nothing in the dataset shows that PCGF1 was actually lost in those cells. There is one guide RNA, one clone, and no rescue experiment. So a result saying 'deleting PCGF1 changes nothing' cannot be separated from 'PCGF1 was not deleted'.
The same thing, for a scientist
**The PCGF1-knockout arm of GSE260742 is the deposit `hyp-q117`'s prediction 5 names, and this beat analysed it. Its genotype is not verified anywhere in the deposit** [E1]. | what would verify the knockout | is it in the deposit? | is it in the key-resources workbook? | |---|---|---| | PCGF1 western blot or protein measurement | **no** | no | | PCGF1 transcript level (RNA-seq, qPCR) | **no** - the NTvKO arm deposits only ChIP-seq | no | | sequencing of the edited locus / indel spectrum | **no** | no | | a **second independent guide** | **no** - one mouse guide only | **`Mouse PCGF1 sgRNA1` is the only mouse guide listed; the human arm has two** | | a **second clone** | **no** - `C1`/`CLONE1` throughout | no | | a **rescue construct** | **no** | no | **The reagent, read verbatim from the workbook this session:** `Mouse PCGF1 sgRNA1 F = CACCGCTGTGGATTTAGCATTAGT`, `R = AAACACTAATGCTAAATCCACAGC`, supplier IDT. **The human arm of the same paper uses two guides** (`Human PCGF1 sgRNA1` `CACCGGGGCAGTACTTGCTAGTT` and `sgRNA2` `CACCGAGGTACGCCATCTCCGG`), which is the standard precaution against an off-target or partial edit, and the mouse arm does not have it. **Why this is filed as its own entry rather than as a caveat.** This beat's ARM 2 produced a clean null - the menin inhibitor's H3K79me2 effect changes by 0.0166 between the genotypes and the DOT1L inhibitor's by 0.0101 - and **a null in an unverified knockout is the single most likely way for that result to be wrong.** The corpus rule that a caveat naming a gap is a task applies directly: the gap is named here, priced, and left open rather than absorbed into a caveat where it would stop being visible. **What the deposit DOES show is that the two genotypes are not identical material**: their H3K79me2 DMSO top-0.1% sets overlap at Jaccard **0.6915** and their DMSO share vectors correlate at Spearman **0.5040**, and pre-registered refutation **RP3** - *the knockout is not a knockout*, which fires only on near-identity - **does not fire.** So the clone differs from its control; what is unverified is *why*.
| reagent as deposited | Mouse PCGF1 sgRNA1 F CACCGCTGTGGATTTAGCATTAGT / R AAACACTAATGCTAAATCCACAGC (IDT), one guide; the human arm of the same paper uses two |
|---|---|
| clone | C1 (H3K79me2 and input filenames) / CLONE1 (H2AK119ub filenames), a single clone throughout |
| what the deposit does show | the two genotypes are not identical material: Jaccard 0.6915 between their H3K79me2 DMSO top-0.1% sets, Spearman 0.5040 between their DMSO share vectors, and pre-registered refutation RP3 does not fire |
| absence class | measured-not-reported |
| absence class note | The knockout was almost certainly validated at the bench - a Nature Cell Biology paper's PCGF1-KO clone will have a blot - and the number was not written into anything this project can read: the deposit has no such sample and the key-resources workbook lists reagents rather than results. Cost to close: open the paper's figure panels, which are image-only with zero extractable characters (Q120), so it is a human reading a PDF rather than an afternoon of scripting. Filed measured-not-reported and NOT never-measured, because the distinction decides whether the next step is an email or an experiment. |
| absence class assigned | 2026-09-23 (Q124) |
what is missing
[ "PCGF1 protein measurement", "PCGF1 transcript measurement", "sequencing of the edited locus", "a second independent guide", "a second clone", "a rescue construct" ]
- A null result in an unverified knockout is the weakest form of a null, and this beat used one to score a standing prediction. If the C1 clone retains PCGF1 function, criteria P1 and P3 measured nothing and prediction 5's DOT1L-axis clause is still untested - and nothing in the deposit can distinguish those cases.
- The internal validation is available and this beat did not run it. The same clone's H2AK119ub tracks should show a loss of PRC1.1 deposition; if they do not, the knockout is in doubt and so is every number in ARM 2. That check was outside this beat's declared scope and it is the first thing the next beat should do.
- Does the paper's figure set contain a PCGF1 validation panel for the C1 clone? MOESM1 is image-only with zero extractable characters, so this needs a human to look. It is the cheapest thing anyone could do to strengthen or kill this beat's ARM 2 null.
- Is the single mouse guide's protospacer (CTGTGGATTTAGCATTAGT) unique in mm10? An off-target check against the reference is pure computation and was not run; a guide with a second perfect match would make the clone's phenotype ambiguous in a different way.
- Do the H2AK119ub tracks from the same C1 clone show the loss of PRC1.1 activity that a PCGF1 knockout should produce? That is the deposit's own internal validation of the genotype and it is twelve downloads away - GSM8123399/401/402, 407/408/409 and 417/418/419 - and it would verify the knockout from its expected phenotype rather than from its label.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Of GSE260742's 158 sample records, the six NTvKO H3K79me2 samples (GSM8123435/436/437 PCGF1 KO and GSM8123438/439/440 control) are ChIP-seq only and no sample in the series measures PCGF1 protein, PCGF1 transcript or the edited PCGF1 locus in those cells; the depositing paper's key-resources workbook sheet Oligos lists exactly one mouse PCGF1 guide (Mouse PCGF1 sgRNA1 F CACCGCTGTGGATTTAGCATTAGT / R AAACACTAATGCTAAATCCACAGC, IDT) against two for the human arm, and lists no rescue construct; all knockout H3K79me2 and H2AK119ub files are named C1 or CLONE1, i.e. a single clone. The two genotypes' H3K79me2 DMSO top-0.1% sets overlap at Jaccard 0.6915 and their DMSO share vectors correlate at Spearman 0.5040 over the 1,227,708-bin universe.
GEO series GSE260742 (Gilan Lab, Monash), per-sample SOFT records: 158 samples, 419,824 characters, fetched 2026-09-23 · NCBI Gene Expression Omnibus 2026What this evidence cannot show
THIS IS AN ABSENCE IN THE DEPOSIT AND IN THE KEY-RESOURCES WORKBOOK, NOT IN THE PAPER. The paper very likely contains a validation panel; Q120 established that PMID 41634341's MOESM1 is image-only with zero extractable characters, so its figures were not examined by this beat and no attempt was made. The claim is therefore scoped to what is downloadable. The channel inventory rests on a grep over !Sample_title and the characteristics blocks of the 158 records for PCGF1, RNA and expression vocabulary; a validation sample titled in a way that grep did not anticipate would be missed. The Jaccard and Spearman figures show the two genotypes are different material but say nothing about whether the difference is PCGF1 loss, clonal drift, or both.
Evidence that argues against this (1)
the single most likely way that null is wrong: no PCGF1 protein, transcript or locus measurement is deposited for GSM8123435/436/437, one guide, one clone, no rescue - so 'PCGF1 loss changes nothing' cannot be separated from 'PCGF1 was not lost'
the only PCGF1-knockout chromatin deposit this project has analysed carries no measurement of PCGF1 itself
the deposit's own internal validation of its genotype: H2AK119ub in the C1 clone validates the knockout from its expected phenotype rather than from its label, which no deposited PCGF1 measurement can do because none exists
the absence this entry works around with three phenotype validations