Oncofusion

start / graph / hyp-q130-the-signature-is-a-gradient-in-eviction-depth-so-every-aggregate-readout-is-underpowered

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

A menin inhibitor's chromatin signature is organised by how much fusion it removed from each site, so every mark's response is a GRADIENT in eviction depth - and any readout that averages over a fusion-occupancy set is underpowered by construction, sometimes to exactly 1.000

also called the eviction-gradient hypothesis, generalised · the aggregate is the wrong statistic · Q130's hypothesis

The drug's effect on chromatin is not one number per mark - it is a slope across sites, steepest where the drug removed the most fusion protein. Any measurement that averages across all the sites at once will understate the effect, and can report exactly nothing when a great deal has happened.

The same thing, for a scientist

**Statement.** For a mark whose deposition depends on the fusion complex, the menin inhibitor's effect at a given site is a monotone function of how much fusion the drug removed from that site. The aggregate over a fusion-occupancy set therefore reports the mean of a monotone function over the range it varies across, which is a number belonging to no site and which can sit arbitrarily close to 1.000. **The evidence assembled so far, all from deposited data and all in this corpus** [E1][E2]: | mark | Spearman vs eviction depth | span | aggregate | |---|---:|---:|---:| | H3K27me3 (Q125) | +1.000 | 1.0227 | 1.3612 | | H2AK119ub (Q125, two conventions) | +1.000, +1.000 | 0.3820, 0.1567 | 1.1694, 0.9071 | | H3K4me3 (Q125) | -1.000 | 0.2534 | 1.1276 | | H3K79me2 (Q130, murine, 4 cultures) | -1.000 x4 | 0.2438-0.6005 | 0.7736-0.8324 | | H3K79me2 (Q130, human MOLM13) | -1.000 | 0.3675 | **0.998** | | DOT1L occupancy (Q130, human) | -1.000 | 0.2549 | 0.2773 | **Six channels, two species, two deposits, two drug-target classes. Every one is monotone in eviction depth. Two of the six have aggregates within 0.09 of 1.000.** **Prior probability 0.6.** The pattern is strong and the mechanism is plausible, but every measurement supporting it comes from KMT2A::MLLT3 systems, three of the six channels come from one beat, and the corpus has repeatedly found a clean pattern to be a property of an instrument. The DOT1L-inhibitor control is what lifts it above 0.5: a drug that moves the same mark five times as hard produces no gradient at all, in four of four cultures. **The strongest single argument against it** is that the strata themselves may encode chromatin state rather than drug action: all three murine inputs are monotone across the quartiles at Spearman -1.000, and no accessibility measurement exists in either deposit to separate the two [E3].

Go deeper
What follows if it is true
  1. Every committed aggregate share ratio on a fusion-occupancy set in this corpus should be re-scored stratified. Q125 wrote an experiment entry proposing exactly that; this beat is the first execution of it and it changed the reading of Q124's headline.
  2. Pharmacodynamic proposals built on unstratified occupancy readouts are underpowered by construction, not by assay quality, and the human MOLM13 aggregate of 0.998 is the worked demonstration.
  3. Cross-study disagreements about whether a menin inhibitor moves a mark may be disagreements about which sites were sampled, not about the biology.
What follows if it is false

The most likely way for it to be false is that the eviction quartiles are a proxy for something else - chromatin accessibility, replication timing, or distance to a strong promoter - and that the drug's effect is organised by that rather than by eviction. That would not make the gradient an artefact; it would relabel its axis, and the practical consequence for a pharmacodynamic readout would be the opposite of reassuring, because the correct stratifier would then be something nobody is measuring.

Facts and figures
statementA menin inhibitor's effect on any fusion-dependent chromatin mark at a site is a monotone function of how much fusion the drug removed from that site, so an aggregate over a fusion-occupancy set is the mean of a monotone function over its own range and is underpowered by construction.
origindata-pattern
prior probability0.6
absence classmeasured-not-analysed
absence class noteFalsifier 2 - the DMSO-versus-DMSO selection control - is computable today from tracks in C:/Temp/q122 at zero download cost, and was not run in this beat. Cycle 83 named the same missing control for the MLL1 channel and it has not been run there either.
absence class assigned2026-09-23 (Q130)
epigenetic readoutall marks, eviction-depth stratified, under a menin inhibitor
predictions
[
  "Any additional mark measured over S_FLAG's eviction quartiles under a menin inhibitor will be monotone at |Spearman| = 1.000, with active marks negative and repressive marks positive.",
  "Re-scoring any committed aggregate share ratio on a fusion-occupancy set will reveal a gradient whose Q1-to-Q4 span exceeds the aggregate's distance from 1.000.",
  "A drug that does NOT act on the fusion's chromatin binding - a DOT1L inhibitor, an EZH2 inhibitor - will produce a flat profile across the same quartiles however large its aggregate effect.",
  "In RS4;11 (KMT2A::AFF1), H3K79me2 stratified by MENIN eviction depth under 330 nM VTP-50469 for 3 days will run monotonically with Spearman -1.000."
]
analyses
[
  "ana-q130-the-eviction-gradient-exists-in-h3k79me2-and-is-the-steepest-of-any-active-mark-in-these-bins",
  "ana-q130-in-a-human-line-and-an-independent-deposit-an-aggregate-of-0-998-hides-a-monotone-eviction-gradient",
  "ana-q130-the-dot1l-inhibitor-removes-86-percent-of-h3k79me2-at-these-bins-with-no-eviction-depth-structure-at-all"
]
experiments
[
  "exp-q130-bin-the-rs4-11-arm-of-gse127507-and-score-the-index-fusions-own-eviction-gradient"
]
falsifiers
[
  "In RS4;11 (human KMT2A::AFF1, GSE127507, 330 nM VTP-50469 3 days against DMSO), H3K79me2's within-sample share ratio across four MENIN-eviction-depth quartiles of MENIN's DMSO top 0.1% gives |Spearman| < 1.000 OR a Q1-minus-Q4 span below 0.10 - i.e. below the smallest span the bin-level bootstrap can resolve at this bin count. That would show the gradient is a property of KMT2A::MLLT3 systems and not of menin inhibition.",
  "A DMSO-versus-DMSO contrast at S_FLAG, stratified the same way, produces a Q1-minus-Q4 span above 0.15 in murine MLL-AF9 - i.e. more than a quarter of the observed VTP span - showing that selection alone reproduces most of the gradient. GSE260742 deposits no second DMSO H3K79me2 for the NTvKO arm, but the MLLAF9 Control replicate pair supplies one: GSM8123372 against GSM8123386, DMSO against DMSO.",
  "An ATAC-seq or MNase measurement in murine MLL-AF9 under 500 nM VTP-50469 48 h shows accessibility at S_FLAG changing monotonically across the eviction quartiles by more than 1.5-fold Q1 to Q4, which would relabel the gradient's axis as accessibility rather than eviction.",
  "Adding a fifth and sixth mark with no expected relationship to the fusion - H3K9me3, H3K36me3 - over the same 1,228 bins under the same drug returns |Spearman| = 1.000 for both, which would show that any mark measured over these strata is monotone and the panel's two-and-two split is not informative."
]
How this could bite us back
  • If every aggregate readout of menin inhibition is underpowered by construction, then the pharmacodynamic data supporting current dosing in this drug class may be systematically understating engagement - which cuts both ways, because it also means a dose that looks insufficient on an aggregate readout may be sufficient.
  • The hypothesis makes the assay harder, not easier: it says the useful measurement requires a paired fusion-occupancy ChIP per sample. If that is undeliverable in a patient, the practical conclusion is that this drug class has no usable chromatin pharmacodynamic marker at all, which is cycle 83's threat restated with a mechanism.
Still open
  1. Does the hypothesis extend to transcription? No RNA-seq from any of these cultures has been located, so 'the chromatin signature is a gradient' cannot yet be connected to whether output is a gradient.
  2. Is there a mark that is NOT monotone in eviction depth? Every channel tested so far is, which is either the hypothesis being right or the strata being a general axis of variation. A mark with no fusion relationship is the discriminating test and none has been measured.
  3. What is the right functional form? Four quartiles cannot distinguish linear from threshold behaviour, and a per-bin regression against the continuous eviction ratio would - n = 1,228 instead of n = 4.
Every source behind this entry (3)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    Six channels across two species and two deposits are monotone in eviction depth at |Spearman| = 1.000: H3K27me3, H2AK119ub (two conventions), H3K4me3, H3K79me2 (four murine cultures and one human line) and DOT1L occupancy - with the two repressive marks positive and the active marks and the writer negative.

    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. HYPOTHESIS-SPECIFIC: the six channels are NOT six independent tests. Four of them are measured over the SAME 1,228 murine bins in overlapping cultures, and the two human channels are measured over the same 1,389 human bins in one culture. A joint probability is not computed anywhere and would be wrong if it were.

    Evidence that argues against this (2)
    • A published methods study establishes that STANDARD ChIP-seq normalisation FAILS TO DETECT a genome-wide decrease in a histone mark under an inhibitor of its writer: standard methods see no H3K27me3 reduction under an EZH2 inhibitor, and a substantial reduction appears only with a Drosophila chromatin spike-in. Every number in this entry is exactly the kind of normalisation that paper shows to be blind. Egan B et al. 2016
    • The system is murine MLL-AF9 (KMT2A::MLLT3) myeloid cells carrying a retrovirally over-expressed FLAG-tagged construct. This graph exists for human infant B-lineage KMT2A::AFF1, and this corpus's commonest documented synthesis failure is generalising a murine MLL-AF9 myeloid result to that disease. Neville D et al. 2026
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    Two of the six channels have aggregates within 0.09 of 1.000 - H2AK119ub rep2 at 0.9071 and human H3K79me2 at 0.998 - while their eviction spans are 0.1567 and 0.3675, so an aggregate near unity is compatible with a large organised effect.

    What this evidence cannot show

    A within-sample share ratio near 1.000 is also what no effect looks like, and nothing distinguishes the two without the stratification. That is the hypothesis's content and also its circularity risk: the claim that the aggregate is uninformative is supported by the same data that the stratified statistic is computed from.

    Evidence that argues against this (1)
    • If the aggregate is near 1.000 because the share that leaves Q1 arrives at Q4 within the same set, the gradient is partly an arithmetic consequence of the normalisation rather than two independent facts. Iqbal Lab oncofusion knowledge graph 2026
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    The strongest argument against the hypothesis is that the eviction quartiles may encode chromatin state rather than drug action: all three murine input tracks are monotone across the same quartiles at Spearman -1.000 with max/min 1.13-1.17, and no accessibility or copy-number measurement exists in either deposit to separate accessibility from eviction.

    What this evidence cannot show

    The human series' input is NOT graded (Spearman +0.400, empirical p = 0.90), which weakens this objection for the human arm and not for the murine one. An input track conflates copy number, sonication accessibility and mappability and cannot separate them.

    Evidence that argues against this (1)
Connected to