Oncofusion

start / graph / ana-q125-the-two-second-priority-marks-carry-the-largest-effect-and-the-control-that-rules-out-an-artefact

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q125-the-h2ak119ub-arm-of-the-pcgf1-knockout

The 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)

also called the second-priority arms were the best value · H3K27me3 at the vacated sites · the opposite-sign control

Two sets of files that everyone had ranked lower turned out to matter most. One shows the strongest switching-off signal anywhere in this work; the other, by moving in the opposite direction in exactly the same places, is the reason we can be confident the first is real and not a quirk of the measurement.

The same thing, for a scientist

**Twelve deposited `.tdf` files, public since 2025-11-23, never binned by anyone**: H3K27me3 GSM8123426/427/428 (control) and GSM8123423/424/425 (PCGF1-KO), H3K4me3 GSM8123432/433/434 and GSM8123429/430/431, in the same 2 x 3 DMSO / SGC0946 5 uM / VTP-50469 500 nM design as the H2AK119ub arm. Q124's specification ranked them second priority *'because they do not validate the knockout and they do not score prediction 5'*. **Both judgements turned out to be wrong** [E1]. **VTP-50469 by eviction quartile of the fusion's own top-0.1% set (Q1 = most fusion lost):** | | Q1 | Q2 | Q3 | Q4 | Spearman | |---|---:|---:|---:|---:|---:| | **H3K27me3, control** | **2.0007** | 1.5120 | 1.1304 | 0.9780 | **+1.000** | | **H3K27me3, PCGF1-KO** | 1.0134 | 1.0581 | 0.9014 | 0.8058 | +0.800 | | **H3K4me3, control** | **0.9491** | 1.1298 | 1.1761 | 1.2025 | **-1.000** | | **H3K4me3, PCGF1-KO** | 0.7852 | 0.8601 | 0.8780 | 0.9259 | -1.000 | **H3K27me3 doubles at the most-evicted bins and collapses to 1.0134 in the knockout - a genotype difference of -0.9873, the largest number in this cycle** [E1]. It scores prediction 5's silencing clause in a second channel and it grades the knockout, so it does both of the things it was said not to do. **H3K4me3 is the control that makes all of it interpretable.** Across the SAME four sets of bins, with the SAME baseline coverage structure and the same libraries, it runs the other way - lost where the fusion is most evicted, gained where the fusion stays. **A shared library, coverage or normalisation artefact cannot give two marks Spearman +1.000 and a third -1.000**, and no matched null was needed to establish that (though one was run and is flat to 0.003) [E2]. **The DOT1L-inhibitor arm separates the two drugs sharply.** SGC0946 gives H3K27me3 0.8868 -> 0.6685 across Q1..Q4, i.e. a LOSS graded the other way from VTP's gain, and H3K4me3 0.8958 -> 1.0598. **The two drugs do not act on one axis in this channel either**, which is the third independent statistic to say so after Q122's R4 and Q124's DEVIATION 5 floor result [E1].

Go deeper
Why the priority ranking was wrong, and it is a generalisable reason

Q124 ranked these arms second because they 'do not validate the knockout' and 'do not score prediction 5'. Both premises were about what the CHANNEL is for rather than about what the CONTRAST would show:

  • H3K27me3 does score prediction 5's silencing clause - it is a repressive mark and the clause says 'not silenced', which does not name an antibody.
  • H3K27me3 does grade the knockout, and more sharply than H2AK119ub does: -0.9873 at Q1 against -0.3125 and -0.1309 in the two H2AK119ub conventions. PRC2 recruitment is downstream of H2AK119ub via JARID2/AEBP2, so a PRC1.1 deficit can show up amplified in the PRC2 readout.
  • H3K4me3 was ranked as a specificity control and is one, and it is the single most load-bearing control in this cycle. Without it the +1.000 gradients would need an external argument; with it they need none.

The generalisable form: rank an arm by the CONTRAST it supports, not by what its channel is usually used for. A mark that is not the mechanism's own readout can still be the only thing that makes the mechanism's readout believable, and it will always look like a low priority on a mechanism-first ranking.

The reagents are not resolved and that is recorded rather than guessed

Q124 read the H2AK119ub (NEB 8240S, clone D27C4), H3K79me2 (ABCAM ab3594) and FLAG (NEB 14793S) rows from the paper's key-resources workbook. The H3K27me3 and H3K4me3 rows were not read and are recorded UNRESOLVED here rather than guessed at. antibod occurs zero times in all 419,824 characters of the 158 GEO sample records, so the workbook is the only route and it is one spreadsheet read away. That matters because H3K27me3 antibodies differ substantially in me2/me3 cross-reactivity, and because a polyclonal would carry the same me2/me3 ambiguity ab3594 carries for H3K79.

Facts and figures
scriptanalyses/q125-the-h2ak119ub-arm-of-the-pcgf1-knockout/deviation2_quartiles_and_conventions.py
statusrun
result summaryH3K27me3 VTP/DMSO at the most-evicted quartile: 2.0007 in the control, 1.0134 in the PCGF1 knockout. H3K4me3: 0.9491 and 0.7852, with Spearman -1.000 against eviction depth in both genotypes.
reagentsH3K27me3 and H3K4me3 antibodies NOT RESOLVED; the paper's key-resources workbook 41556_2025_1859_MOESM6_ESM.xlsx (on disk at C:/Temp/q120) has the rows and they were not read
absence classmeasured-not-analysed
absence class noteTwelve .tdf files in GSE260742, public since 2025-11-23, never binned by any beat of this project. Named with accessions by Q122 section 7 in cycle 82 and by Q124's specification in cycle 83, both times as a lower priority. Cost: 1.45 GB of download and about three minutes. What remains MEASURED-NOT-REPORTED is the two antibodies' identities, which are rows in a spreadsheet already on disk.
absence class assigned2026-09-23 (Q125)
accessions h3k27me3
{
  "control": [
    "GSM8123426",
    "GSM8123427",
    "GSM8123428"
  ],
  "pcgf1_ko": [
    "GSM8123423",
    "GSM8123424",
    "GSM8123425"
  ]
}
accessions h3k4me3
{
  "control": [
    "GSM8123432",
    "GSM8123433",
    "GSM8123434"
  ],
  "pcgf1_ko": [
    "GSM8123429",
    "GSM8123430",
    "GSM8123431"
  ]
}
cgi ratio knockout over control dmso
{
  "K27": 0.9198,
  "K4": 1.2638
}
How this could bite us back
  • If H3K27me3 doubling at the vacated sites is part of a menin inhibitor's mechanism, then combining a menin inhibitor with an EZH2 inhibitor - a combination that looks attractive on a bivalency rationale - could antagonise the very silencing the menin inhibitor depends on. This entry does not test it and the combination is being explored clinically in other contexts.
  • The knockout's CpG islands carry less H3K27me3 and more H3K4me3 than its control's before any drug. A cell that has drifted that way is primed to resist the silencing arm, and nothing in a fusion-occupancy assay would show it.
Still open
  1. Which antibodies? Two rows of 41556_2025_1859_MOESM6_ESM.xlsx, already downloaded to C:/Temp/q120 by Q120, would close it and would tell us whether the H3K27me3 reagent can separate me2 from me3.
  2. Do the eviction quartiles overlap the bivalent promoters Sparbier et al. describe? That is the one computation that would let this beat's result and theirs be compared instead of merely coexisting, and it needs only an H3K4me3/H3K27me3 co-occupancy call on the DMSO tracks now on disk.
  3. Is the H3K27me3 gain PRC2-dependent, or is it JARID2/AEBP2 reading the H2AK119ub gain? An EZH2 inhibitor arm would separate them and the series has none.
Every source behind this entry (2)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    In murine MLL-AF9 (GSE260742), across Q117's four eviction quartiles of the FLAG-defined top-0.1% set, 500 nM VTP-50469 48 h against DMSO 72 h gives H3K27me3 share ratios of [2.0007, 1.512, 1.1304, 0.978] in the non-targeting control (GSM8123426/428) and [1.0134, 1.0581, 0.9014, 0.8058] in the PCGF1-knockout (GSM8123423/425), and H3K4me3 [0.9491, 1.1298, 1.1761, 1.2025] (GSM8123432/434) and [0.7852, 0.8601, 0.878, 0.9259] (GSM8123429/431), with Spearman against eviction depth of +1.000, +0.800, -1.000 and -1.000; under SGC0946 5 uM 72 h the control's H3K27me3 runs [0.8868, 0.839, 0.8193, 0.6685] and H3K4me3 [0.8958, 1.006, 1.0541, 1.0598]. No beat of this project had previously binned any H3K27me3 or H3K4me3 file from this series.

    What this evidence cannot show

    NO SPIKE-IN NORMALISATION IS POSSIBLE. Q117 verified zero hits for the entire spike-in vocabulary across all 419,824 characters of GSE260742's 158 sample records and no scale factor is deposited, so every cross-condition number here is a WITHIN-SAMPLE SHARE, which cancels the unknown per-sample scale factor exactly and IS BLIND TO A UNIFORM CHANGE BY CONSTRUCTION. Nothing here measures any mark's abundance; everything measures its distribution. EXPOSURE ASYMMETRY: DMSO 72 h and SGC0946 5 uM 72 h are matched, VTP-50469 500 nM is 48 h, so every VTP ratio is a LOWER BOUND on the 72-hour effect - identical in both genotypes, so the genotype DIFFERENCE is unaffected. REAGENTS: H2AK119ub is Ubiquityl-Histone H2A (Lys119) (D27C4) XP Rabbit mAb, NEB 8240S, and it CANNOT distinguish PRC1.1 from canonical PRC1 - RING1A/B is the ligase in both and PCGF1 is one of six PCGF paralogs - nor deposition from loss of a deubiquitinase. The H3K27me3 and H3K4me3 reagents are NOT RESOLVED and are recorded unresolved rather than guessed; `antibod` occurs zero times in the 158 GEO records. S_FLAG is defined by DYKDDDDK Tag (D6W5B) Rabbit mAb, NEB 14793S, on a RETROVIRALLY OVER-EXPRESSED tagged MLLAF9 construct, so it is the over-expressed molecule's sites and not an endogenous fusion's. THE KNOCKOUT is one CRISPR guide, one clone C1, no second clone, no rescue and NO deposited PCGF1 protein or transcript measurement. SYSTEM: murine MLL-AF9 (KMT2A::MLLT3) myeloid blood cells, NOT the infant B-lineage KMT2A::AFF1 disease this graph is about. THESE TWO ARMS HAVE ONE CONVENTION EACH AND THEREFORE NO SPREAD AT ALL: n = 1 per condition per genotype, and the H2AK119ub arm's own two conventions differ by up to 0.2623 on the equivalent aggregate quantity, so a single-track H3K27me3 or H3K4me3 number should be read with at least that resolution. THE H3K27me3 AND H3K4me3 ANTIBODIES ARE NOT RESOLVED - the key-resources workbook rows were not read - so neither reagent's me2/me3 cross-reactivity is known. The quartile stratification is Q117's own DEVIATION 3 and applying a threshold to the genotype difference at a quartile is POST HOC in this beat.

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

    Against a null in which each quartile is matched to bins outside the FLAG-defined set on the decile of the same track's own DMSO share (20 draws, seed 20260923), the H3K27me3 matched null runs 0.9247 to 0.9180 across the four quartiles and the H3K4me3 null 1.0333 to 1.0303 - flat to within 0.007 and 0.003 respectively - so neither observed gradient is a baseline-coverage effect, and the two marks' opposite signs across identical bins exclude a shared library artefact without needing the null at all.

    What this evidence cannot show

    The null is matched on one covariate only and is POST HOC. Its medians sit at 0.88-1.03 rather than exactly 1.0, the expected consequence of matching on a decile of a skewed distribution, and that offset is not interpreted. The opposite-sign argument is qualitative: it establishes that no single multiplicative library factor can produce both gradients, not that no confounder can.

    Evidence that argues against this (1)
Connected to