start / graph / exp-q124-the-h2ak119ub-arm-of-the-pcgf1-knockout-is-twelve-downloads-and-scores-two-falsifiers-at-once

ExperimentstrongQ124-the-timecourse-and-the-pcgf1-knockout

Bin the deposited H2AK119ub NTvKO 2 x 3 design: twelve files, three replicate conventions, public since 2025-11-23 and never analysed - it is the only route to the silencing clause of hyp-q117's prediction 5, it is the internal validation of the PCGF1 knockout this project's null depends on, and it carries a spread the H3K79me2 arm cannot

also called the twelve-download H2AK119ub arm · the internal validation of the C1 clone · prediction 5's silencing clause

A dozen public files measure the repressive chromatin mark in exactly the control-versus-knockout, three-drug design this cycle analysed for a different mark. Nobody has looked at them. They would test whether the vacated sites really are being switched off, and they would also check, from its expected effect, that the knockout worked at all.

The same thing, for a scientist

**Analysis, not wet work.** Twelve deposited `.tdf` files, binned with the identical 2 kb instrument and the identical `U_m` universe this project has used since Q117, answer **three separate standing questions at once** [E1]. **1. The silencing clause of `hyp-q117`'s prediction 5.** That prediction reads *"in a PCGF1 or BCOR knockout the fusion is evicted normally but the vacated sites are not silenced"*, and names the NTvKO arm as the relevant deposit. **H2AK119ub is the silencing readout and PCGF1 is the subunit that carries PRC1.1's ubiquitylation activity**, so the prediction's sharpest form is: the VTP-induced H2AK119ub gain at the fusion's most-evicted bins should be **abolished or strongly reduced** in the knockout. **CORRECTION (cycle 90): this entry originally said Q117 MEASURED that H2AK119ub gain. Q117 measured no histone mark at all** - it binned FLAG, KDM2B, BCOR and RING1B occupancy only (`analyses/q117-gained-fusion-occupancy/results.json`, `gsm` block, re-read 2026-09-23). **So criteria H1 (1.25/1.10) and H3 (rho >= 0.80) below are calibrated on RING1B occupancy while scoring an H2AK119ub mark - a different channel.** `hyp-q124`'s F3 states the channel correctly, so this was a slip in one field rather than a design error. Caught by Q125. Q117's control-genotype numbers are already in this graph - RING1B rising **1.38x** in the most-evicted quartile against 0.81-0.88x elsewhere, 16.4x at Meis1, 7.5x at Jmjd1c, 3.6x at Six1 - so **only the knockout half is new.** **2. The internal validation this cycle's null depends on.** `ana-q124-the-menin-inhibitors-h3k79me2-effect-is-unchanged-in-a-pcgf1-knockout` is a null in a clone whose genotype is unverified in the deposit - one guide, one clone, no rescue, no PCGF1 measurement. **H2AK119ub in the same clone validates the knockout from its expected phenotype rather than from its label**, and it is falsifier F2 of `hyp-q124`: if the knockout's H2AK119ub responds to VTP-50469 exactly as its control's does, the clone did not lose PRC1.1 function and this cycle's P1 and P3 measured nothing. **3. A spread.** The H3K79me2 arm has n = 1 per condition per genotype. **The H2AK119ub arm has three replicate conventions** - `paired`, `rep1` and `rep2` - so every figure can carry a between-convention range that no H3K79me2 figure in this corpus can. **The twelve files, transcribed from the live records this session:** ``` paired: GSM8123403 / 405 / 406 (Control DMSO / SGC0946 5uM / VTP50469 500nM) GSM8123399 / 401 / 402 (PCGF1 KO DMSO / SGC0946 5uM / VTP50469 500nM) rep1: GSM8123410 / 411 / 412 (Control) GSM8123407 / 408 / 409 (PCGF1 KO) rep2: GSM8123420 / 421 / 422 (Control) GSM8123417 / 418 / 419 (PCGF1 KO) inputs: GSM8123416 (NT, .tdf) and GSM8123415 (C1, .tdf) - USABLE GSM8123404 (F3_NT, .filt.bam.tdf) and GSM8123400 (F3_C1, .filt.bam.tdf) - usable GSM8123413 and GSM8123414 - DEPOSITED AS .filt.bam AND NOT .tdf, so NOT readable by this instrument ```

Go deeper

A pre-registerable criterion set, written here so the next beat does not have to invent one

Offered, not pre-registered - the next beat writes its own protocol before it computes. But the thresholds below are measured quantities from this graph rather than invented ones, which is the part that is hard to reconstruct.

id	criterion	population	fires if	source of the threshold
H1	the silencing clause holds	murine MLL-AF9, FLAG-DMSO top-0.1% of `U_m`, VTP-50469 500 nM 48 h vs DMSO 72 h, control genotype against PCGF1-KO	`R_K119ub(VTP/DMSO)` >= 1.25 in the control AND <= 1.10 in the knockout	1.25 and 1.10 bracket Q117's measured 1.38 in the most-evicted quartile; band +/-20% on the gap
H2	the knockout is a knockout (= `hyp-q124` F2)	as H1	`abs(R_K119ub,KO - R_K119ub,control) >= 0.05`; if it is below 0.05 the clone shows the same in-fill as its control and every ARM 2 null from cycle 83 is void	the +/-0.05 absolute share-ratio band this corpus uses, Q120 section 5
H3	the quartile grading reproduces (= `hyp-q124` F3)	as Q117: the FLAG-defined set stratified into eviction quartiles	Spearman of H2AK119ub gain against eviction depth >= 0.80 across quartiles in the control	Q117 measured 1.38 / 0.88 / 0.81 across its quartiles
H4	the three replicate conventions agree	all three, control genotype, `S_FLAG`	the range of `R_K119ub(VTP/DMSO)` across `paired`, `rep1` and `rep2` is < 0.0584	Q122's measured between-culture spread for H3K79me2 at the same set
H0	instrument	every file	largest single bin < 5% of the non-blacklisted total; build calls mm10 from each file's own extents against mm10/mm39/hg38/hg19	this project's standing checks

Two hazards to carry into it. Q117 measured that this series' trimmed_* files carry 14-46% of their coverage in blacklist intervals against 1.6-1.9% for ChIP_* files, and the paired convention's files are .filt.bam.tdf while rep1 and rep2 are bare .tdf - three processing states across the three conventions, so H4's agreement test is partly a test of processing. And the three conventions use two filename families (F3_NT/F3_C1 for paired, bare NT/CLONE1 for rep1 and rep2), so whether 'replicate' means a technical repeat is not settled by the records - the same ambiguity Q122's section 2a found for H3K79me2.

Why this is the highest-value unanalysed thing in the series

Three cycles have now named this arm and none has run it. hyp-q117's prediction 5 named it in cycle 68 and recorded that the beat did not touch it. Q122's section 7 declined it with its accessions. This beat declined the H2AK119ub half by declared scope and priced it instead.

What makes it better value than anything else left in GSE260742 is that it is the only unanalysed arm which simultaneously scores a standing prediction, validates a reagent this project's newest null depends on, and supplies a spread for a corpus whose H3K79me2 numbers are all n = 1. The H3K27me3 (GSM8123426/427/428 and 423/424/425) and H3K4me3 (GSM8123432/433/434 and 429/430/431) arms of the same design are a further twelve files and would extend the same design to the bivalency axis that PMID 36635503 places menin and PCGF1-PRC1.1 on together - but they do not validate the knockout and they do not score prediction 5, so they come second.

Facts and figures

type	re-analysis of deposited data; no wet work, no new sequencing
cost	twelve downloads totalling an estimated 1.0-1.6 GB by analogy with the H3K79me2 arm's file sizes, one at a time with deletion after binning, plus about two minutes of compute; six inputs of which four are readable
antibody	H2AK119ub is Ubiquityl-Histone H2A (Lys119) (D27C4) XP Rabbit mAb, New England Biolabs 8240S, read from the paper's key-resources sheet this session; it CANNOT distinguish PRC1.1 from canonical PRC1 deposition, nor deposition from loss of a deubiquitinase
status	proposed
absence class	measured-not-analysed
absence class note	Public since 2025-11-23, named by hyp-q117's prediction 5 in cycle 68 as 'the relevant deposit', named with accessions by Q122's section 7 in cycle 82, declined by scope in cycle 83, and never binned by anyone. Cost: twelve downloads and two minutes of compute. It is the cheapest unrun measurement this project currently has that could change a committed conclusion - specifically, it could void cycle 83's PCGF1 null outright via criterion H2.
absence class assigned	2026-09-23 (Q124)

accessions h2ak119ub

{
  "paired_control": [
    "GSM8123403",
    "GSM8123405",
    "GSM8123406"
  ],
  "paired_pcgf1_ko": [
    "GSM8123399",
    "GSM8123401",
    "GSM8123402"
  ],
  "rep1_control": [
    "GSM8123410",
    "GSM8123411",
    "GSM8123412"
  ],
  "rep1_pcgf1_ko": [
    "GSM8123407",
    "GSM8123408",
    "GSM8123409"
  ],
  "rep2_control": [
    "GSM8123420",
    "GSM8123421",
    "GSM8123422"
  ],
  "rep2_pcgf1_ko": [
    "GSM8123417",
    "GSM8123418",
    "GSM8123419"
  ]
}

accessions inputs

{
  "usable_tdf": [
    "GSM8123416 (NT)",
    "GSM8123415 (C1)",
    "GSM8123404 (F3_NT)",
    "GSM8123400 (F3_C1)"
  ],
  "not_readable_deposited_as_filt_bam": [
    "GSM8123413",
    "GSM8123414"
  ]
}

accessions second priority

{
  "h3k27me3_control": [
    "GSM8123426",
    "GSM8123427",
    "GSM8123428"
  ],
  "h3k27me3_pcgf1_ko": [
    "GSM8123423",
    "GSM8123424",
    "GSM8123425"
  ],
  "h3k4me3_control": [
    "GSM8123432",
    "GSM8123433",
    "GSM8123434"
  ],
  "h3k4me3_pcgf1_ko": [
    "GSM8123429",
    "GSM8123430",
    "GSM8123431"
  ]
}

offered criteria

[
  "H0 instrument: largest single bin < 5% of non-blacklisted total; build calls mm10 from each file's own extents against mm10/mm39/hg38/hg19",
  "H1 silencing clause: R_K119ub(VTP/DMSO) >= 1.25 in the control AND <= 1.10 in the knockout, at the FLAG-DMSO top-0.1% of U_m; thresholds bracket Q117's measured 1.38",
  "H2 the knockout is a knockout: abs(R_KO - R_control) >= 0.05; below 0.05 voids every ARM 2 null from cycle 83",
  "H3 quartile grading reproduces: Spearman of H2AK119ub gain against eviction depth >= 0.80 across quartiles in the control, against Q117's 1.38/0.88/0.81",
  "H4 the three replicate conventions agree: range of R_K119ub(VTP/DMSO) across paired, rep1 and rep2 < 0.0584, Q122's measured between-culture spread"
]

hazards

[
  "three processing states across the three replicate conventions (paired is .filt.bam.tdf, rep1 and rep2 are bare .tdf), and Q117 measured 14-46% blacklist load in this series' trimmed_* family against 1.6-1.9% for ChIP_*",
  "the three conventions use two filename families, so 'replicate' is not established to mean a technical repeat",
  "two of the six inputs are .filt.bam and not readable by the TDFReader instrument",
  "no spike-in anywhere in the series, so every number is a within-sample share, blind to uniform change",
  "the same 48 h VTP against 72 h vehicle asymmetry as every other arm of this series"
]

How this could bite us back

The most likely outcome of criterion H2 is the uncomfortable one for this cycle: if the PCGF1-knockout clone's H2AK119ub responds to VTP-50469 the same way its control's does, then cycle 83's P1 and P3 nulls measured an ineffective edit and the entries built on them are void. That is the correct thing for this experiment to be able to do, and it is why H2 is listed above H1.
If H1 fires - in-fill in the control, none in the knockout - the result is a confirmation of a prediction made in cycle 68 from a deposit named in cycle 68, which is a good outcome for the hypothesis and a poor one for the argument that this corpus's predictions are hard to test. Three cycles passed before anyone priced twelve downloads.

Still open

Do 'paired', 'rep1' and 'rep2' denote technical replicates, biological replicates, or related cultures? The three conventions span two filename families and two processing states, so criterion H4's agreement test is partly a processing test and its result will need reading with that in mind.
Should the H3K27me3 and H3K4me3 arms be run in the same pass? They are twelve more files in the same design and they reach the bivalency axis that PMID 36635503 places menin and PCGF1-PRC1.1 on together, which is the strongest published version of the competing hypothesis to hyp-q124. They do not validate the knockout, so they are second priority rather than co-equal.
Is there an H2AK119ub arm in a human KMT2A-rearranged line with a PCGF1 knockout anywhere? PMID 39158067 worked in human and mouse KMT2A-rearranged AML models and may have deposited one; no cross-series search was run.

Every source behind this entry (1)

E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
GSE260742 deposits an H2AK119ub NTvKO design of twelve ChIP samples in three replicate conventions - paired GSM8123403/405/406 (control DMSO/SGC0946 5 uM/VTP50469 500 nM) and GSM8123399/401/402 (PCGF1 KO), rep1 GSM8123410/411/412 and GSM8123407/408/409, rep2 GSM8123420/421/422 and GSM8123417/418/419 - plus inputs GSM8123404 and GSM8123400 (.filt.bam.tdf), GSM8123416 and GSM8123415 (.tdf), and GSM8123413 and GSM8123414 which are deposited as .filt.bam and not .tdf; and an H3K27me3 design (GSM8123426/427/428 control, GSM8123423/424/425 knockout) and an H3K4me3 design (GSM8123432/433/434 control, GSM8123429/430/431 knockout) in the same 2 x 3 structure. No beat of this project has binned any of them.
GEO series GSE260742 (Gilan Lab, Monash), per-sample SOFT records: 158 samples, 419,824 characters, fetched 2026-09-23 · NCBI Gene Expression Omnibus 2026
What this evidence cannot show
The accession-to-condition mapping is read from !Sample_title and the supplementary filename of each record; the assignment of 'paired', 'rep1' and 'rep2' to biological or technical replication is NOT settled by the records - the three conventions use two filename families (F3_NT/F3_C1 against bare NT/CLONE1) and Q122's section 2a established that this series' titles and filenames disagree about grouping. Two of the six inputs are .filt.bam and not readable by the TDFReader instrument this project uses. Q117 measured that this series' trimmed_* files carry 14-46% of coverage in blacklist intervals against 1.6-1.9% for ChIP_* files, and the three conventions here are in two different processing states. 'No beat has binned any of them' is a claim about this repository, checked against the q117, q118, q120 and q122 scratch directories and their committed results.json.
Evidence that argues against this (1)
- Nature Cell Biology 2026

Connected to

tests →H3K79me2 reports only the DOT1L arm of a menin inhibitor's action and is blind to its PRC1.1 arm - so the mark's response is unchanged in a PCGF1 knockout even though PCGF1 loss is a published resistance mechanism, and a cell resistant through PRC1.1 loss would look fully drug-engaged by this readout
scores falsifiers F2 and F3 of that hypothesis, and F2 can void it outright: if the knockout's H2AK119ub responds to VTP-50469 as its control's does, the clone did not lose PRC1.1 function and cycle 83's nulls measured nothing
tests →Menin inhibition removes the fusion from its strongest sites without relocating it to a new programme, and PRC1.1 fills in only where the eviction is complete - so the drug's chromatin action is subtraction plus local substitution, not redistribution
the only route to the silencing clause of prediction 5: H2AK119ub is the silencing readout and PCGF1 carries PRC1.1's ubiquitylation activity, so the VTP-induced in-fill Q117 measured at 1.38x in the most-evicted quartile should be abolished in the knockout
supports →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
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
tests →What fills the space: RING1B and BCOR arrive at the fusion's sites under BOTH drugs, but the DOT1L inhibitor puts them everywhere on the fusion's territory while the menin inhibitor puts them only in the quartile where the fusion is most completely evicted - and at Meis1 RING1B rises 16-fold
supplies the knockout half of a measurement whose control half is already here: Q117's RING1B 1.38x in the most-evicted quartile against 0.81-0.88x elsewhere is the comparator, and criteria H1 and H3 are written against it
related to →PRC1.1 — the complex this graph has been leaning on, now enumerated subunit by subunit, and the enumeration changes the conclusion
the H2AK119ub antibody (NEB 8240S clone D27C4) cannot distinguish PRC1.1 from canonical PRC1 deposition, so the experiment's PRC1.1 attribution rests on the PCGF1 genotype and not on the reagent
← refutesQ124's criterion H2 was written to be able to void Q124's own conclusion, and it cannot - because it is scored on a contrast in which the control genotype shows no effect to lose. Its raw arithmetic FIRES at 0.0813 against a 0.05 threshold, which reads as 'the knockout is a knockout' and is not evidence of anything
its criterion H2 cannot void Q124's ARM 2 in either direction
← testsThe first replicated share ratio this corpus has ever had: two title-declared replicates of ONE murine culture span 0.2623 on the same quantity - 4.5x the 0.0584 between-culture spread this corpus uses as its indifference threshold - and they disagree in SIGN about whether the PCGF1 knockout blunts the DOT1L-inhibitor response
scores its offered criterion H4: the range is 0.2623 against a 0.0584 threshold
← refutesThe H3K27me3 and H3K4me3 NTvKO arms - twelve files nobody had ever binned, listed as second priority - carry the largest effect in this cycle (H3K27me3 2.0007 at the most-evicted quartile, collapsing to 1.0134 in the PCGF1 knockout) and the free internal control that rules out a library artefact (H3K4me3 runs the opposite way, Spearman -1.000)
its 'second priority' ranking of the H3K27me3 and H3K4me3 arms: both premises were wrong