Oncofusion

start / graph / ana-q130-the-eviction-span-is-sensitive-where-the-aggregate-is-blind-and-specific-where-the-aggregate-is-not

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

Across eight murine contrasts and two human stratifications, the H3K79me2 eviction span separates a menin inhibitor from a DOT1L inhibitor 22-fold with no overlap AND detects menin engagement in a human cell whose aggregate reads 0.998 - two properties the aggregate has one of and one of

also called the candidate pharmacodynamic statistic · shape, not size · sensitive where the aggregate is blind

Two different measurements of the same chemical tag behave differently. The simple average can tell the two drugs apart but in human cells it cannot tell the menin drug from no drug at all. The pattern across sites does both.

The same thing, for a scientist

**Cycle 83 concluded that this corpus has no validated occupancy-based pharmacodynamic marker for menin inhibition**, because Q122's C7 found a DOT1L inhibitor moving the MLL1-N occupancy channel as much as a menin inhibitor did - 0.3012 against 0.3298 at MLL1's own top sites. **That was measured on the MLL1-N channel and nothing here refutes it** [E1]. **This entry asks a narrower question about a different channel:** does any statistic on H3K79me2 have BOTH the properties a pharmacodynamic marker needs - sensitivity to menin engagement, and specificity against the other drug that moves the same mark? | | detects menin engagement? | distinguishes it from DOT1L inhibition? | |---|---|---| | **H3K79me2 aggregate** | **NO in human** - MOLM13 reads **0.998** at MLL1-N's top sites | YES - murine VTP 0.77-0.83 against SGC 0.09-0.15 | | **H3K79me2 eviction span** | **YES** - human span 0.3675, 0 of 20 matched-null draws reach it | **YES** - see below | **The specificity, on eight murine contrasts with no overlap** [E2]: | | four VTP-50469 spans | four SGC0946 spans | |---|---|---| | values | -0.2438, -0.4847, **-0.6005**, -0.4341 | +0.0031, +0.0036, +0.0088, **+0.0109** | | Spearman vs eviction depth | -1.000 in all four | +0.200, +0.400, +0.400, +0.800 | **The smallest menin-inhibitor span is 0.2438 and the largest DOT1L-inhibitor span is 0.0109 - a 22.4-fold separation with no overlap across eight contrasts in four cultures.** And the direction differs: every VTP span is negative and every SGC span is positive [E2]. **The sensitivity, in the human independent series** [E3]: the aggregate is 0.998 - the number a pharmacodynamic assay would call 'no engagement' - while the same 1,389 bins split by MLL1-N eviction depth give a span of 0.3675 that **no matched-null draw of 20 reached** (largest null span 0.053). **So the statistic that has both properties is the SHAPE of the response across eviction depth, and it is not the size of the response.** On magnitude alone the DOT1L inhibitor's effect on this mark is five times the menin inhibitor's [E2].

Go deeper
What this is NOT, because the gap between a statistic and an assay is the whole problem

It is a candidate pharmacodynamic STATISTIC, not a candidate ASSAY. Computing the span requires knowing each bin's eviction depth, which requires a paired fusion-occupancy measurement in the same sample. In the murine system that is a FLAG ChIP on a tagged, over-expressed construct. In a human cell with an endogenous fusion it has to be MLL1-N or MENIN - and cycle 83's C7 is precisely the finding that the MLL1-N channel is not drug-specific. So the human version of this assay needs two or three ChIPs per sample and inherits C7's problem in its stratifier.

Practical objections, all unsolved here: a marrow aspirate has limited cells; three ChIPs at two timepoints is a real cost; and the strata are defined by a within-sample ranking that needs enough depth to be stable. None of this is addressed by anything in this beat, and the entry does not claim otherwise.

Where the aggregate and the span disagree, and which to believe

They do not disagree; they measure different things. The aggregate measures how much of the mark's share moved out of the set. The span measures whether the movement was organised by how much fusion the drug removed. A DOT1L inhibitor moves a great deal of share and organises none of it. A menin inhibitor moves less and organises all of it.

The murine aggregate (0.8156) and the human aggregate (0.998) differ by 18 points while their spans are both large and both negative. Four things differ between them at once - species, set definition (FLAG top sites against MLL1-N top sites), dose and schedule (500 nM 48 h against 330 nM 3 d), and deposit - so nothing here decomposes that difference. What it demonstrates is that the aggregate is the less stable of the two statistics across systems, which is an argument for the span and not a measurement of why.

Facts and figures
scriptanalyses/q130-does-the-eviction-gradient-exist-in-h3k79me2/run.py
statusrun
result summaryVTP spans -0.2438 to -0.6005 (Spearman -1.000 x4) against SGC spans +0.0031 to +0.0109 (Spearman +0.2 to +0.8): 22.4-fold separation, no overlap, opposite signs. Human aggregate 0.998 with span 0.3675.
separation ratio smallest vtp over largest sgc22.4
human aggregate0.998
human span-0.3675
human null span max0.053
absence classmeasured-not-analysed
absence class noteCycle 83 wrote that this corpus has no validated occupancy-based pharmacodynamic marker for menin inhibition and named it as the sharpest threat of that cycle. Every number needed to test a candidate was already binned on disk in C:/Temp/q120, q122 and q124.
absence class assigned2026-09-23 (Q130)
epigenetic readoutH3K79me2 eviction-depth span as a candidate pharmacodynamic statistic for menin inhibition, scored against a DOT1L inhibitor
four vtp spans
NTvKO Control (GSM8123440)-0.601NTvKO Control (GSM8123440): -0.6005MLLAF9 Control rep2 (GSM8…-0.485MLLAF9 Control rep2 (GSM8123388): -0.4847NTvKO PCGF1-KO (GSM812343…-0.434NTvKO PCGF1-KO (GSM8123437): -0.4341MLLAF9 Control rep1 (GSM8…-0.244MLLAF9 Control rep1 (GSM8123374): -0.2438
four sgc spans
MLLAF9 Control rep10.003MLLAF9 Control rep1: 0.0031NTvKO Control0.004NTvKO Control: 0.0036MLLAF9 Control rep20.009MLLAF9 Control rep2: 0.0088NTvKO PCGF1-KO0.011NTvKO PCGF1-KO: 0.0109
four vtp spearmans
MLLAF9 Control rep1-1.000MLLAF9 Control rep1: -1MLLAF9 Control rep2-1.000MLLAF9 Control rep2: -1NTvKO Control-1.000NTvKO Control: -1NTvKO PCGF1-KO-1.000NTvKO PCGF1-KO: -1
four vtp spans — the numbers behind the chart
{
  "MLLAF9 Control rep1 (GSM8123374)": -0.2438,
  "MLLAF9 Control rep2 (GSM8123388)": -0.4847,
  "NTvKO Control (GSM8123440)": -0.6005,
  "NTvKO PCGF1-KO (GSM8123437)": -0.4341
}
four sgc spans — the numbers behind the chart
{
  "MLLAF9 Control rep1": 0.0031,
  "MLLAF9 Control rep2": 0.0088,
  "NTvKO Control": 0.0036,
  "NTvKO PCGF1-KO": 0.0109
}
four vtp spearmans — the numbers behind the chart
{
  "MLLAF9 Control rep1": -1,
  "MLLAF9 Control rep2": -1,
  "NTvKO Control": -1,
  "NTvKO PCGF1-KO": -1
}
How this could bite us back
  • Proposing a pharmacodynamic statistic that needs a paired fusion-occupancy ChIP in the same patient sample is proposing something that may simply not be deliverable in a marrow aspirate. If it is not, the honest conclusion is that cycle 83's threat stands in practice even though a statistic exists in principle.
  • The stratifier available in a human cell is MLL1-N or MENIN, and cycle 83 showed the MLL1-N channel moves under a DOT1L inhibitor too. So an assay built this way could be fooled by a DOT1L inhibitor acting on its stratifier even while the mark's shape stays flat - and nobody has run that combination.
  • A statistic whose magnitude reproduces only to within a factor of two cannot grade depth of response. It can say engaged or not engaged, and it can say which drug class. Any proposal to titrate a dose on it is unsupported by anything measured here.
Still open
  1. Does the span separate the two drugs in a HUMAN cell? GSE127507 has no SGC0946 arm and GSE260742's human MOLM13 arm has SGC but no deposited MLL1-N or MENIN track to stratify by, so the human specificity test needs a series that has all three and none has been identified.
  2. Can the strata be defined without a paired occupancy ChIP - for instance from a fixed reference set of high-eviction loci derived once in a model system? That would turn a statistic into an assay and is the single highest-value follow-up.
  3. Does the span respond to dose? Everything here is one dose per series, and a dose-response would be the strongest argument that the statistic tracks engagement rather than exposure.
Every source behind this entry (3)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    Cycle 83 recorded, from Q122's criterion C7, that a DOT1L inhibitor moves the MLL1-N occupancy channel as much as a menin inhibitor does - 0.3012 under SGC0946 against 0.3298 under VTP-50469 at MLL1's own top sites in murine MLL-AF9, both far below a pre-registered floor of 0.85 - and concluded that this corpus has no validated occupancy-based pharmacodynamic marker for menin inhibition.

    What this evidence cannot show

    C7 is a measurement on the MLL1-N OCCUPANCY channel at MLL1's own top sites. This entry is about the H3K79me2 HISTONE MARK at fusion-occupancy sites. They are different channels in different bin sets and nothing here refutes C7 or its arithmetic. Cycle 83 also noted that a raw share ratio at a channel's own top sites is biased downward by selection, and that a DMSO-versus-DMSO contrast at S_MLL1 was never run.

    Evidence that argues against this (1)
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    Across four murine cultures, the four VTP-50469 H3K79me2 eviction spans are -0.2438, -0.4847, -0.6005 and -0.4341 (Spearman -1.000 in all four) and the four SGC0946 spans are +0.0031, +0.0036, +0.0088 and +0.0109 (Spearman +0.200 to +0.800) - a 22.4-fold separation between the smallest VTP span and the largest SGC span, with no overlap and with opposite signs, while on aggregate magnitude the DOT1L inhibitor's effect on the same mark is about five times the menin inhibitor's.

    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. COMPARISON-SPECIFIC: SGC0946 is exposure-matched to its vehicle at 72 h and VTP-50469 is 48 h, so the VTP spans are lower bounds and the separation is conservative. The four cultures are not four independent biological replicates - two of them may be a technical replicate pair, which the companion dispute entry reports.

    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
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    In human MOLM13 from the independent series GSE127507, the H3K79me2 aggregate at MLL1-N's DMSO top 0.1% is 0.998 under 330 nM VTP-50469 for 3 days while the eviction span over the same bins is 0.3675 with Spearman -1.000 and 0 of 20 matched-null draws reaching it (largest null span 0.053), so the span detects an engagement the aggregate does not.

    What this evidence cannot show

    The human arm has no SGC0946 sample, so its SPECIFICITY is imported from the murine contrasts and is not demonstrated in human cells. Shared MLL1-N epitope, same-culture stratifier, rank quartiles, n = 1. THE ONE DOWNLOAD: the hg19 blacklist was re-fetched because C:/Temp/q118 has been deleted.

    Evidence that argues against this (1)
Connected to