Oncofusion

start / graph / ana-q125-the-menin-inhibitor-does-silence-the-sites-it-evicts-the-fusion-from-and-the-effect-is-graded-by-how-much-fusion-left

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

At the 2 kb bins a menin inhibitor strips the most fusion from, both repressive marks GAIN and the active mark is LOST, monotonically in eviction depth (Spearman +1.000, +1.000 and -1.000) against a baseline-matched null that is flat to within 0.007 - the silencing clause of hyp-q117's prediction 5, measured for the first time

also called the eviction gradient · prediction 5's silencing clause, scored · H2AK119ub and H3K27me3 gain where the fusion leaves

When a drug pulls the cancer fusion protein off the DNA, the spots it leaves behind get chemically switched off - and the more of the fusion that leaves a spot, the more strongly that spot is switched off. Two different 'off' marks go up together and the 'on' mark goes down, in exactly the same places, which is why this is not a measurement artefact.

The same thing, for a scientist

**Deposited data, re-analysed; no wet work.** GSE260742's NTvKO arm was binned at 2 kb on the mm10 grid with the instrument Q117, Q118, Q120, Q122 and Q124 used, over the identical universe `U_m` (1,227,708 bins, asserted in code against Q122's committed JSON) and the identical `S_FLAG` set (FLAG-DMSO GSM8123359 top 0.1%, 1,228 bins) [E1]. **The aggregate over all 1,228 bins says nothing.** `R_H2AK119ub(VTP/DMSO)` in the non-targeting control is **1.1694** in the `paired` convention and **0.9071** in `rep2` - straddling 1.0, mean 1.0383. Q122 measured **0.8473** for the same quantity in a different culture of the same line. On that number alone the silencing clause looks dead, and this beat's pre-registered refutation **RA fired** [E1]. **Stratified by how much fusion the drug actually removed, it is not dead at all.** Splitting `S_FLAG` into Q117's four eviction quartiles (Q1 = most fusion lost, cuts 0.0992 / 0.1748 / 0.2717, 306-308 bins each) [E2]: | VTP-50469 500 nM vs DMSO, control genotype | Q1 | Q2 | Q3 | Q4 | Spearman vs eviction depth | |---|---:|---:|---:|---:|---:| | **H2AK119ub** `paired` | **1.4450** | 1.1483 | 1.1238 | 1.0630 | **+1.000** | | **H2AK119ub** `rep2` | **1.0151** | 0.9063 | 0.8878 | 0.8584 | **+1.000** | | **H3K27me3** | **2.0007** | 1.5120 | 1.1304 | 0.9780 | **+1.000** | | **H3K4me3** | **0.9491** | 1.1298 | 1.1761 | 1.2025 | **-1.000** | **The two repressive marks rise together where the fusion leaves; the active mark falls there and rises where the fusion stays.** [E1] **It is not regression to the mean.** Each quartile was matched to bins OUTSIDE `S_FLAG` on the decile of the same track's own DMSO share, 20 draws, seed 20260923. **Every matched null is flat to within 0.007 across the four quartiles** - H2AK119ub `paired` 0.9933 to 0.9881, H3K27me3 0.9247 to 0.9180, H3K4me3 1.0333 to 1.0303 [E3]. **And the decisive internal control is free: H3K4me3 runs the OPPOSITE way across the SAME bins with the SAME baseline structure. A shared library or coverage artefact cannot give two marks +1.000 and a third -1.000.** [E3]

Go deeper
Why the aggregate and the gradient disagree, and which one to believe

They do not disagree. S_FLAG is the fusion's top 1,228 bins in DMSO, and the drug does not remove the fusion equally from all of them - that is the whole content of Q117's quartile split, whose cuts run from a VTP/DMSO FLAG share ratio of 0.0992 to 0.2717. Averaging a quantity that is monotone in eviction depth over a set that spans a 3.5-fold range of eviction depth returns its mean, which here is near 1.0. The aggregate is arithmetically correct and biologically uninformative.

The general form is worth keeping, because this corpus has committed several aggregate share ratios on this exact set: a set defined by where a protein IS is not stratified by where a drug takes it away, and a within-sample share over the unstratified set will average a real gradient to nothing. Q122's 0.8473, Q122's criterion D1 verdict DOES NOT FIRE, and this beat's own S1, S2 and V1 are all of that shape.

What this does and does not say about prediction 5

hyp-q117's prediction 5 has three clauses and a disjunction. This entry scores clause (b), 'the vacated sites are not silenced', in the control genotype: the vacated sites ARE silenced, so the clause as written is REFUTED for a normal cell - which is the good outcome for the hypothesis's mechanism and the bad outcome for its wording, because the clause is a prediction about the KNOCKOUT and its premise about the control had never been measured.

  • (a) 'the fusion is evicted normally' - not scored here. There is no FLAG, MLL1 or MENIN track in the NTvKO arm by title. Six files carry a FLAG ChIP filename in their own TDF header and their channel could not be settled; see the companion entry.
  • (c) 'the programme returns on drug withdrawal' - unreachable from the whole series; no washout arm exists in any of the 158 records.
  • 'or BCOR' - unreachable; the series' only genetic perturbation is the PCGF1 knockout.

So prediction 5 is PARTIAL and the proposed cashed-outcome text says which clause was scored.

Scale, and what the numbers are NOT

Every figure is a within-sample share ratio and is blind to uniform change by construction, so nothing here says H2AK119ub or H3K27me3 increased in amount - only that their distribution moved towards the bins the fusion vacated. The exposure is also asymmetric (VTP 48 h against a 72 h vehicle), so each VTP ratio is a lower bound on the 72-hour effect. And the largest effect measured is H3K27me3's 2.0007 at Q1 in a channel nobody in this project had ever binned, from an arm Q124's specification listed as second priority.

Facts and figures
scriptanalyses/q125-the-h2ak119ub-arm-of-the-pcgf1-knockout/run.py
statusrun
result summaryAt the most-evicted quartile of the fusion's own top-0.1% set, VTP-50469 gives H2AK119ub 1.4450 (paired) and 1.0151 (rep2), H3K27me3 2.0007 and H3K4me3 0.9491, with Spearman against eviction depth of +1.000, +1.000, +1.000 and -1.000 and matched nulls flat to 0.007.
absence classmeasured-not-analysed
absence class noteGSE260742's H2AK119ub, H3K27me3 and H3K4me3 NTvKO arms were public from 2025-11-23 and had never been binned by anyone in this project. hyp-q117's prediction 5 named the NTvKO arm as 'the relevant deposit' in cycle 68; Q122 section 7 declined it with accessions in cycle 82; Q124 declined it by scope and priced it in cycle 83. This entry is the compute. It cost 5.33 GB of download and about ten minutes.
absence class assigned2026-09-23 (Q125)
inputs
[
  "GSE260742 GSM8123403/405/406 (H2AK119ub NT paired)",
  "GSM8123420/421/422 (H2AK119ub NT rep2)",
  "GSM8123426/427/428 (H3K27me3 NT)",
  "GSM8123432/433/434 (H3K4me3 NT)",
  "GSM8123359 FLAG-DMSO and GSM8123361 FLAG-VTP (Q117, reused)",
  "GSM8549823 input (Q117, reused, defines U_m)"
]
universe
{
  "U_m_bins": 1227708,
  "S_FLAG_bins": 1228,
  "n_per_quartile": [
    308,
    306,
    307,
    307
  ]
}
per quartile VTP control genotype
{
  "H2AK119ub_paired": [
    1.445,
    1.1483,
    1.1238,
    1.063
  ],
  "H2AK119ub_rep2": [
    1.0151,
    0.9063,
    0.8878,
    0.8584
  ],
  "H3K27me3": [
    2.0007,
    1.512,
    1.1304,
    0.978
  ],
  "H3K4me3": [
    0.9491,
    1.1298,
    1.1761,
    1.2025
  ]
}
matched null medians
{
  "K119_PAIRED_NT_VTP": [
    0.9933,
    0.9835,
    0.9858,
    0.9881
  ],
  "K119_REP2_NT_VTP": [
    0.9687,
    0.9609,
    0.9637,
    0.963
  ],
  "K27_NT_VTP": [
    0.9247,
    0.9054,
    0.9061,
    0.918
  ],
  "K4_NT_VTP": [
    1.0333,
    1.0212,
    1.034,
    1.0303
  ]
}
How this could bite us back
  • If the menin inhibitor's therapeutic effect runs through polycomb deposition at vacated sites rather than through eviction alone, then a leukaemia that cannot deposit H2AK119ub - a BCOR, BCORL1, KDM2B or PCGF1 loss, all of which occur - should be evicted normally and respond poorly. That is a resistance mechanism that a fusion-occupancy pharmacodynamic marker would score as full engagement.
  • The effect is strongest exactly where the drug works hardest, so a partial responder - lower exposure, poorer penetration - would sit in the Q3/Q4 regime where the marks barely move. The gradient means the biomarker is dose-shaped, and a single-threshold assay on unstratified sites would read near 1.0 in a responder and in a non-responder alike.
Still open
  1. Does the gradient hold at the human index fusion? Everything here is murine MLL-AF9 (KMT2A::MLLT3) and this graph exists for infant KMT2A::AFF1. No human H2AK119ub arm with a matched FLAG or MLL1 eviction measurement has been identified in any series.
  2. Is the H3K4me3 sign reversal a loss of the active mark or a redistribution towards the bins where the fusion stays? A within-sample share cannot separate those, and an H3K4me3 spike-in arm would.
  3. Does the gradient track transcription? No RNA-seq from these exact samples has been located, so 'silenced' here means 'gains two repressive marks and loses an active one', not 'stops being transcribed'.
Every source behind this entry (3)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    In murine MLL-AF9 (GSE260742, cell line MLLAF9, myeloid, blood, taxid 10090), at the FLAG-DMSO top-0.1% set of U_m stratified into Q117's four eviction quartiles, 500 nM VTP-50469 48 h against DMSO 72 h in the non-targeting control gives H2AK119ub share ratios of [1.445, 1.1483, 1.1238, 1.063] (paired, GSM8123403/406) and [1.0151, 0.9063, 0.8878, 0.8584] (rep2, GSM8123420/422), H3K27me3 [2.0007, 1.512, 1.1304, 0.978] (GSM8123426/428) and H3K4me3 [0.9491, 1.1298, 1.1761, 1.2025] (GSM8123432/434), with Spearman against eviction depth of +1.000, +1.000, +1.000 and -1.000. The aggregate over all 1,228 bins is 1.1694, 0.9071, 1.3612 and 1.1276 respectively.

    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. n = 1 per condition per convention; H3K27me3 and H3K4me3 have ONE convention and therefore NO spread at all. The two H2AK119ub conventions differ at the aggregate by 0.2623, which is larger than Q122's 0.0584 between-culture spread that this corpus uses as its indifference threshold - the QUARTILE ORDERING agrees between them (both Spearman +1.000) while the LEVEL does not.

    Evidence that argues against this (1)
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    The eviction quartiles are Q117's, reused unchanged: the 1,228 FLAG-held bins split by their own FLAG VTP/DMSO within-sample share ratio at cuts 0.0992 / 0.1748 / 0.2717, giving 308/306/307/307 bins against Q117's 307/307/307/307. Q117 measured RING1B - the PRC1 catalytic subunit, not the H2AK119ub mark - across those quartiles at 1.3808 / 0.8708 / 0.8070 / 0.8792, a Spearman against eviction depth of 0.400.

    What this evidence cannot show

    The quartile cuts and the 1,228-bin set are read from Q117's committed results.json in code, not typed. The 308/306 split against Q117's 307/307 is a boundary-tie difference of one bin and is reported rather than adjusted. Q117's RING1B numbers are an OCCUPANCY ChIP of the enzyme, not a measurement of the mark: the mark had never been measured on this set in this series until now, and the H2AK119ub gradient (Spearman +1.000) is CLEANER than the RING1B gradient (0.400) the prediction was built from.

    Evidence that argues against this (1)
  3. E3moderate 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 S_FLAG on the decile of the same track's own DMSO share (20 draws, seed 20260923), every matched null is flat across the four quartiles to within 0.007 - H2AK119ub paired [0.9933, 0.9835, 0.9858, 0.9881], H3K27me3 [0.9247, 0.9054, 0.9061, 0.918], H3K4me3 [1.0333, 1.0212, 1.034, 1.0303] - so the observed gradients are not a baseline-coverage effect; and H3K4me3's gradient runs in the opposite direction across the same bins, which a shared library artefact cannot produce.

    What this evidence cannot show

    The null is matched on ONE covariate - the same track's own DMSO share decile - and not on GC, mappability, CpG-island status or distance to a TSS. It therefore excludes regression to the mean and does not exclude every confounder. It is POST HOC: it was written after the gradient was seen, and it is reported as a control rather than as a pre-registered test. The matched null medians sit at 0.88-1.03 rather than exactly 1.0, which is the expected consequence of matching on a decile of a skewed distribution and is not interpreted.

    Evidence that argues against this (1)
Connected to