Oncofusion

start / graph / ana-q130-the-pcgf1-knockouts-h3k79me2-difference-sits-at-the-least-evicted-quartile-not-the-most-so-q124s-threat-stands

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q130-does-the-eviction-gradient-exist-in-h3k79me2

Stratifying by eviction depth does NOT rescue H3K79me2 as a resistance-detecting readout: at the most-evicted quartile the PCGF1-knockout and the control are 0.6123 against 0.5860, a difference of 0.0263, while the whole genotype effect sits at the LEAST-evicted quartile (-0.1401) - the opposite of eviction-graded

also called G3 does not fire · Q124's threat survives stratification · the resistant cell still looks drug-engaged where it matters

A cell engineered to carry a known resistance change loses the activating tag just as completely as a normal cell does at the places the drug clears the fusion from. Splitting the measurement up did not reveal a hidden difference there - the only difference is at the places the drug barely touches.

The same thing, for a scientist

**This is the criterion the beat was dispatched to test and it DOES NOT FIRE.** Q124's cycle-83 headline was that the menin inhibitor's effect on H3K79me2 is PCGF1-independent - `V_NT = 0.8156` against `V_KO = 0.7990`, a genotype difference of 0.0166 - so a cell carrying a published resistance genotype scores as fully drug-engaged. The pre-registration asked whether stratification would reveal a difference the aggregate hid [E1]. | VTP/DMSO, H3K79me2 | Q1 most evicted | Q2 | Q3 | Q4 least | aggregate | |---|---:|---:|---:|---:|---:| | non-targeting control | 0.5860 | 0.8188 | 0.9280 | 1.1865 | 0.8156 | | PCGF1-KO clone C1 | 0.6123 | 0.7955 | 0.8928 | 1.0464 | 0.7990 | | **difference KO - control** | **+0.0263** | -0.0233 | -0.0352 | **-0.1401** | -0.0166 | **The pre-registered rule had two clauses and the second one is the one that failed.** G3 required `|Q1 difference| >= 0.0498` (three times Q124's aggregate difference) AND `|Q1 difference| > |Q4 difference|`. The observed Q1 difference is 0.0263 - below the threshold - and it is **5.3 times SMALLER** than the Q4 difference, so the genotype effect is concentrated at the quartile the drug barely touches [E2]. **The pre-registration stated in advance why that matters:** a genotype difference of the same size at Q4 as at Q1 would be a uniform genotype offset and would not rescue anything. What was found is worse than that for the rescue hypothesis - the difference is *inverted*, largest where eviction is smallest [E2]. **The baseline check passes, so the comparison is fair.** The two genotypes' H3K79me2 DMSO shares at `S_FLAG` are 0.03473 and 0.032475, a relative difference of **6.5%**, well inside the pre-registered 20% limit above which the criterion would have been declared uninformative [E2]. **So Q124's threat stands undiminished at the sites that matter.** At the bins the menin inhibitor actually clears the fusion from, a PCGF1-null cell's H3K79me2 response is indistinguishable from a control cell's at the resolution this deposit supports.

Go deeper
What DID move, and why it is the less useful half

The knockout's gradient is real and shallower: span 0.4341 against the control's 0.6005, both at Spearman -1.000. The whole of that 0.166 difference in span comes from Q4 - the knockout gains 1.0464 where the control gains 1.1865.

That is a difference in where the mark goes, not in whether the drug works. Q125 found that the PCGF1 knockout blunts the repressive-mark gain at the vacated sites, and the natural reading of the Q4 result is the mirror image: in a cell that cannot fully silence the vacated sites, less H3K79me2 accumulates at the retained ones. That is a hypothesis this beat cannot test, because it would need the accounting of where the displaced share goes, which nothing here provides.

What would still rescue a resistance-detecting readout, and what would not

Would not: any statistic computed on H3K79me2 at eviction-stratified fusion sites. The Q1 difference is 0.0263 on a scale where the drug effect itself is 0.41, i.e. 6% of the signal, in a single unreplicated library.

Might: the Q4 difference, 0.1401, is five times larger and is in a quartile that is easier to sample because the fusion is still there. A readout built on where the mark accumulates rather than where it leaves would be reading the 0.1401. That inverts the intuition behind every occupancy-based pharmacodynamic proposal in this corpus and is worth an entry of its own; it is written as a hypothesis rather than asserted here.

The knockout's own limits apply to all of this: one CRISPR guide, one clone, no second clone, no rescue, and no deposited PCGF1 protein or transcript measurement anywhere in the series. Every genotype statement is conditional on the deposit's label.

Facts and figures
scriptanalyses/q130-does-the-eviction-gradient-exist-in-h3k79me2/run.py
statusrun
result summaryG3 DOES NOT FIRE. Genotype difference by quartile +0.0263 / -0.0233 / -0.0352 / -0.1401, aggregate -0.0166; Q1 difference below threshold and 5.3x smaller than Q4's; baseline check passes at 6.5% relative.
aggregate difference-0.0166
q124 aggregate difference0.0166
dmso share S FLAG NT0.03473
dmso share S FLAG C10.032475
relative baseline difference0.0649
control span-0.6005
knockout span-0.4341
absence classmeasured-not-analysed
absence class noteThe stratified genotype comparison was computable from tracks Q124 binned in cycle 83 and quartiles Q117 committed in cycle 68. It had never been run, which is why the threat has stood on an aggregate for ten cycles.
absence class assigned2026-09-23 (Q130)
epigenetic readoutH3K79me2 VTP/DMSO share ratio by eviction quartile, control genotype against PCGF1-knockout clone
inputs
[
  "GSM8123438/440 NT DMSO/VTP",
  "GSM8123435/437 PCGF1-KO C1 DMSO/VTP",
  "GSM8123359/GSM8123361 FLAG (define the quartiles)"
]
genotype difference by quartile
[
  0.0263,
  -0.0233,
  -0.0352,
  -0.1401
]
How this could bite us back
  • This is the bite-back. A PCGF1-null leukaemia treated with a menin inhibitor loses H3K79me2 at the evicted sites exactly as a wild-type one does, so an H3K79me2 pharmacodynamic assay - stratified or not - reports full target engagement in a cell the model predicts will not respond. Stratification does not fix this and this beat's headline result must not be read as if it did.
  • The only genotype signal that did appear sits at the least-evicted quartile, where the drug is doing least. A biomarker built there is a biomarker read out of the compartment with the weakest drug effect and the highest residual fusion, which is an uncomfortable place to put a clinical decision.
  • If polycomb-deficient leukaemias are evicted normally and silenced less, then combining a menin inhibitor with anything that restores silencing is the mechanistic move - and the assay that would select patients for it is a repressive-mark assay, not the H3K79me2 one this project has most data on.
Still open
  1. Is the Q4 genotype difference of -0.1401 real? It is the largest genotype signal on this mark anywhere in the deposit and it is in the quartile nobody was watching, in a single unreplicated pair of libraries.
  2. Would a second PCGF1-KO clone reproduce any of this? GSE260742 deposits one clone and one guide, so clonal drift is entirely unexcluded and Q125's matched-input screen is the only check that exists.
  3. Does the mirror-image reading hold - that less silencing of the vacated sites means less H3K79me2 accumulation at the retained ones? Testing it needs a per-bin correlation between the repressive-mark gain and the H3K79me2 change in the same bins, which is one script and was not in this beat's scope.
Every source behind this entry (3)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    Q124 committed in cycle 83 that the menin inhibitor's effect on H3K79me2 at S_FLAG is PCGF1-independent - V_NT = 0.8156 against V_KO = 0.7990, its criterion P1 firing - which means a cell carrying a published resistance genotype scores as fully drug-engaged on that readout.

    What this evidence cannot show

    Read from the committed JSON in code. This beat reproduces Q124's aggregate genotype difference exactly at -0.0166 from independently recomputed shares, which is a check on the pipeline rather than on the biology.

    Evidence that argues against this (1)
    • PCGF1 loss is a resistance genotype by inference from the source paper's model, not by demonstration in a patient; treating the C1 clone as 'a resistant cell' imports that inference. Neville D et al. 2026
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    Stratified by eviction depth, the PCGF1-KO minus control difference in the H3K79me2 VTP/DMSO share ratio is +0.0263 / -0.0233 / -0.0352 / -0.1401 across Q1..Q4 - below the pre-registered 0.0498 threshold at Q1 and 5.3 times smaller at Q1 than at Q4 - with the two genotypes' DMSO shares at S_FLAG differing by only 6.5% relatively, so the pre-registered criterion G3 DOES NOT FIRE and the comparison's baseline check passes.

    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. It cancels the unknown per-sample scale factor exactly and IS BLIND TO A UNIFORM CHANGE BY CONSTRUCTION: nothing here measures H3K79me2'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. REPLICATION, CORRECTED MID-BEAT: the NTvKO arm deposits ONE H3K79me2 ChIP per genotype per drug, so ITS spans carry no technical error bar and its bin-level bootstrap intervals measure sampling within one library, NOT library-to-library variation. A REPLICATE PAIR DOES EXIST for a different culture - GSM8123372/374 and GSM8123386/388, both titled MLLAF9 Control, the second carrying `rep2` - already binned by Q122 and scored in Q130 DEVIATION 4: the SIGN and the monotonicity reproduce (Spearman -1.000 in all four murine VTP contrasts in this deposit) while the SPAN reproduces only to within a factor of 1.99 (-0.2438 against -0.4847). EVERY SPAN IN THIS BEAT SHOULD BE READ WITH THAT 2-FOLD REPRODUCIBILITY. THE KNOCKOUT is one CRISPR guide, one clone C1, no second clone, no rescue, and NO deposited PCGF1 protein or transcript measurement. S_FLAG is defined by a DYKDDDDK-tag ChIP on a RETROVIRALLY OVER-EXPRESSED tagged MLLAF9 construct, so it is the over-expressed molecule's sites and not an endogenous fusion's. THE H3K79me2 ANTIBODY IS NOT RESOLVED - `antibod` occurs zero times in the 158 GEO records - and is recorded unresolved rather than guessed. SYSTEM: murine MLL-AF9 (KMT2A::MLLT3) myeloid blood cells, NOT the infant B-lineage KMT2A::AFF1 disease this graph is about. CRITERION-SPECIFIC: the 0.0498 threshold is three times Q124's own aggregate difference and has no external justification, which the pre-registration stated; the raw differences at all four quartiles are given so any threshold can be applied. The observed 0.0263 is below the 0.03 lower edge of the pre-registered UNDECIDED band, so the verdict does not depend on the multiplier.

    Evidence that argues against this (1)
    • A null result on a single unreplicated pair of libraries is weak evidence of no effect, and this design's power against a genotype difference of 0.03 at Q1 is unquantified. Iqbal Lab oncofusion knowledge graph 2026
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    Q125 found in the same cultures that the PCGF1 knockout reduces but does not abolish the repressive-mark gain at the vacated sites, so the genotype IS detectable on H2AK119ub and H3K27me3 where it is not detectable on H3K79me2 at the most-evicted quartile.

    What this evidence cannot show

    Read from Q125's committed JSON. Q125's own conclusion is that losing PCGF1 reduces the silencing of the vacated sites without abolishing it; the H2AK119ub REP2 convention disagrees with PAIRED on the genotype contrast, so the repressive-side genotype signal is itself convention-dependent.

    Evidence that argues against this (1)
    • If the genotype is visible on the repressive marks and not on H3K79me2, the resistance blind spot is a property of the H3K79me2 readout and not of the cell - which is a better outcome for the therapeutic hypothesis than this entry's title suggests. Iqbal Lab oncofusion knowledge graph 2026
Connected to