Oncofusion

start / graph / ana-q118-q117s-myb-prediction-fires-in-rs4-11-and-fails-in-the-matched-fusion-line

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q118-menin-at-the-human-index-fusion

Q117 predicted an MLL1 ChIP would diverge from FLAG at MYB and agree at MEIS1 - both halves fire in RS4;11 (MYB 1.21 against mouse 0.71, MEIS1 0.25 against 0.23), but the MYB divergence does NOT reproduce in MOLM13, which carries the same fusion as Q117's mouse, so the MYB difference is a partner difference and not an antibody artefact

also called Q117 P5 cashed · MYB diverges, MEIS1 agrees · the prediction that reversed its own explanation

Last cycle's mouse work predicted that a different antibody would disagree about one gene, MYB, and agree about another, MEIS1. Both predictions came true in the human cells this lab cares about - but the MYB disagreement vanishes in a second human line carrying the mouse's own fusion, so the cause is the fusion partner, not the antibody.

The same thing, for a scientist

**Q117 wrote, as its own next step**: *"If a shared-epitope antibody masks fusion loss, MLL1 and FLAG should diverge at Myb and agree at Meis1."* It could not run the test. This is that test, run in human orthologues, normalised on both sides to a floor so the two datasets are comparable [E1]. All figures are the within-sample share ratio (VTP/DMSO) divided by the track's own floor. **The human floor is the aggregate share ratio over 8,579-8,598 windows in DMSO percentile 50-90; the mouse floor is Q117's Gapdh at 0.508, which is two genes and is the best available mouse-side normaliser.** | locus | mouse FLAG / mouse floor | **RS4;11 MLL1-N / floor** | RS4;11 MENIN / floor | **MOLM13 MLL1-N / floor** | ML-2 MLL1-N / floor | |---|---:|---:|---:|---:|---:| | **MEIS1** | 0.230 | **0.245** | 0.010 | 0.440 | (baseline too low) | | **MYB** | 0.713 | **1.205** | 0.282 | **0.625** | 1.318 | | JMJD1C | 0.329 | 0.387 | 0.005 | 0.387 | 0.094 | | HOXA9 | 1.299 | 0.831 | 0.145 | 0.741 | 1.023 | | MEF2C | 1.713 | **0.181** | 0.130 | **0.248** | 0.040 | | GAPDH | 1.000 | 1.763 | 1.960 | 0.770 | 1.548 | | ACTB | 1.124 | 2.649 | 2.404 | 0.743 | 1.767 | **MEIS1: prediction CONFIRMED.** 0.245 in RS4;11 against 0.230 in mouse - agreement to 6%. MEIS1 is the locus where both instruments and both species say the same thing, and menin's own figure there is 0.010, the most complete loss in the whole table. **MYB: prediction CONFIRMED in RS4;11 and REFUTED as an explanation.** In RS4;11 the MLL1-N ratio is **1.205** - above its own floor, i.e. no loss whatever - against the mouse FLAG's 0.713, a 1.69-fold divergence, exactly as predicted. **But in MOLM13, which carries KMT2A::MLLT3 (the same fusion as Q117's murine MLL-AF9), MLL1-N gives 0.625 against the mouse FLAG's 0.713 - agreement to 12%.** The antibody is the same in both human lines. **So the divergence at MYB is a property of the line and its fusion partner, not of the shared epitope.** Q117's prediction fired and its stated mechanism for the prediction did not. That inference would have been unavailable from RS4;11 alone, and it is the reason this beat measured all three lines rather than only the index fusion. **MEF2C: the third locus, and it resolves a disagreement in Q117's favour's opposite direction.** Q117 measured murine Mef2c at 0.870 raw / 1.713 over floor - i.e. **retained** - and recorded that as a break with the published account, which names MEF2C among the loci that lose. **Human MLL1-N loses MEF2C in all three lines: 0.181, 0.248, 0.040 over floor.** See `ana-q118-mef2c-is-lost-in-all-three-human-lines-and-q117s-mouse-retention-does-not-generalise`.

Go deeper
Scoring the prediction honestly

Q117's prediction as written had no numeric threshold, so this beat declared one in its own pre-registration (§5, P5): diverge at MYB means human MLL1-N >= 0.8 where mouse FLAG is 0.362 raw; agree at MEIS1 means human MLL1-N < 0.5 where mouse FLAG is 0.117 raw.

thresholdRS4;11 rawverdict
MYB divergesMLL1-N >= 0.81.183FIRES
MEIS1 agreesMLL1-N < 0.50.240FIRES

On the floor-normalised scale, which is the better one, MYB is 1.205 against 0.713 and MEIS1 is 0.245 against 0.230. Both halves fire on both scalings.

This is the second falsifier-style prediction from the Q117 hypothesis entry to be cashed, and the outcome belongs inside the falsifier text in hyp-q117-menin-inhibition-evicts-the-fusion-without-redistributing-it-to-a-recognisable-programme. That entry is not mine to edit and the HUMAN ARM falsifier stored there is phrased in terms of the gained set's median enrichment, not the locus table, so it needs both: the locus prediction is cashed here and the HUMAN ARM falsifier is scored in ana-q118-fifty-three-non-island-bins-gain-mll1-n-signal-in-two-lines-with-a-flat-input.

The floor had to be rebuilt before any of this could be read, and GAPDH is why

Q117's memo listed this as an open gap in its own work: "A proper floor. Gapdh and Actb are two genes." It matters here more than it did there.

GAPDH and ACTB GAIN share in RS4;11 (1.76x and 2.65x the floor) and in ML-2 (1.55x, 1.77x), and Q117 measured them LOSING share in mouse (0.508, 0.571). A locus table read against a housekeeping control would therefore be normalised by a quantity that points in opposite directions in the two datasets, and the MYB comparison - the whole point of the exercise - would move by a factor of 3.5 depending on which side's housekeeping genes you trusted.

The floor used here is the aggregate share ratio over every window in DMSO percentile 50-90, n = 8,579-8,598 per track: RS4;11 MLL1n 0.981, MOLM13 MLL1n 1.048, ML-2 MLL1n 1.007. That it comes out within 5% of 1.0 for all three MLL1-N tracks is the reassuring part - the MLL1-N distribution does not globally drift - and it is why the locus figures can be read as local effects.

And the floor had to be the AGGREGATE and not the per-window median: MOLM13's MENIN VTP track is 58.7% zeros and its per-window median floor ratio is exactly 0.000, which would have made every normalised ratio in that line undefined. A ratio whose denominator can be zero is a gate that passes or fails by construction; the aggregate's denominator is a sum over 8,600 windows and is asserted non-zero.

Facts and figures
scriptanalyses/q118-menin-at-the-human-index-fusion/roseB.py
statusrun
result summaryQ117's P5 prediction fires on both halves in RS4;11 (MYB 1.205 vs 0.713; MEIS1 0.245 vs 0.230) and the MYB half does not reproduce in MOLM13 (0.625 vs 0.713), so the divergence is a line/partner property rather than an antibody artefact. Floors: RS4;11 MLL1n 0.981, MOLM13 1.048, ML-2 1.007, each over ~8,600 windows.
floor definitionaggregate within-sample share ratio over all TSS windows in DMSO percentile 50-90, n = 8,579-8,598 per track
cell lineRS4;11, MOLM13, ML-2
genome buildhg19
shown for fusionKMT2A::AFF1 (RS4;11), KMT2A::MLLT3 (MOLM13), KMT2A::AFDN (ML-2), all via an MLL1 N-terminus antibody; the murine comparator is FLAG-tagged KMT2A::MLLT3 in GSE260742
inputs
[
  "GSE127507 ROSE TSS signal table",
  "Q117 locus-table.json / hypothesis entry"
]
CORRECTED CYCLE 79 a tagged measurement overturns the not an artefact claim
{
  "what_this_entry_said": "that the MYB divergence is a line/partner property and 'not an antibody artefact'",
  "what_q120_measured": "in the SAME murine nuclei, with a FLAG-tagged fusion beside the MLL1 antibody: MYB FLAG 0.3623 vs MLL1 0.6645, a 1.83x divergence; and genome-wide at the tagged fusion's own top 0.1%, FLAG 0.1793 vs MLL1 0.3995 - 2.2281x. MEIS1 still agrees (0.1171 vs 0.1120, 4%).",
  "why_it_is_decisive": "the murine MLL1 ChIP uses Bethyl A300-086A - THE IDENTICAL ANTIBODY to the human MLL1n track - established from the paper's supplementary sheet (PMID 41634341, MOESM6, sheet `Antibody`), not from GEO, where `antibod` occurs zero times across 158 sample records. So the comparison is tag-versus-antibody in one system, which is exactly the measurement this entry lacked.",
  "the_corrected_claim": "the MYB divergence is NOT cleanly separable from an antibody artefact. A 1.83x tag-versus-antibody divergence at MYB itself, and 2.23x genome-wide, is the same order as the effect this entry attributed to line/partner. BOTH readings remain live and this entry's title overstates.",
  "what_survives": "that MEIS1 agrees while MYB diverges is unchanged and is the interesting asymmetry. The claim that had to go is the negative one - 'not an antibody artefact' - which was never measurable without a tag.",
  "corrected_by": "q120, cycle 79; numbers re-grepped from its results-arm2.json by the king"
}
CORRECTED CYCLE 79 the mouse side floor is 1 83x too low
{
  "what_this_entry_used": "Gapdh at 0.508 as the murine floor, described in the entry as a stopgap",
  "what_it_should_be": "Q118's own aggregate-floor definition applied to the murine data gives FLAG 0.9295 over 505,306 bins",
  "two_readings_change_sign": "mouse Hoxa9 1.299 -> 0.710 and Mef2c 1.713 -> 0.936 - both cross 1.0, i.e. from 'retained above floor' to 'below floor'",
  "and_a_headline_number_moves": "'MEIS1 agreement to 6%' was 0.245 vs 0.230; on the correct floor it is 0.245 vs 0.126",
  "what_survives": "the DIRECTION of every comparison. The precision does not.",
  "corrected_by": "q120, cycle 79"
}
How this could bite us back
  • MYB is the locus where the two human lines disagree, and MYB is a dependency in both. If the drug leaves MYB occupancy intact in the index fusion and reduces it in MLL-AF9, then the same drug is doing different things to the same dependency in two diseases it is prescribed for.
  • A prediction that fires while its stated mechanism fails is the most seductive kind of confirmation. The right record here is that Q117 called the observation correctly and the reason it gave for expecting it is not supported, and anyone quoting the confirmation without the MOLM13 row will propagate the wrong mechanism.
Still open
  1. Is the MYB difference a partner effect or a lineage effect? RS4;11 is B-lineage KMT2A::AFF1 and MOLM13 is myeloid KMT2A::MLLT3, and the two are confounded. A B-lineage KMT2A::MLLT3 line, or a myeloid KMT2A::AFF1 line such as MV4;11, would separate them - and MV4;11 has an entry in this graph as line-mv4-11-kmt2a-aff1-myeloid.
  2. MYB's occupancy in KMT2A-r leukaemia is largely an enhancer story, and instrument A sees only a 6 kb TSS window. What happens at the MYB -84 kb enhancer under VTP-50469 in RS4;11 is not measured here and the binned .tdf data on disk could answer it.
  3. HOXA10 is absent from the deposited ROSE table entirely, so the murine Hoxa10 figure (0.561) has no human counterpart in this beat. Whether that is a RefGene annotation gap or a ROSE filtering decision was not established.
Every source behind this entry (1)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    Floor-normalised within-sample share ratios under VTP-50469: at MYB, human RS4;11 MLL1-N is 1.205 against murine FLAG-MLLAF9's 0.713 (divergence 1.69-fold) while human MOLM13 MLL1-N is 0.625 (agreement to 12%); at MEIS1, RS4;11 MLL1-N is 0.245 against the mouse's 0.230 (agreement to 6%); at MEF2C the human lines give 0.181, 0.248 and 0.040 against the mouse's 1.713.

    What this evidence cannot show

    NEITHER ANTIBODY IN THIS SERIES IS FUSION-SPECIFIC AND THE ANTIBODY NAME SAYS SO. MLL1n is anti-MLL1 N-terminus (Bethyl A300-086A, lots 5 or 6); the N-terminus is retained in KMT2A::AFF1, so that track is the sum of the fusion and wild-type MLL1 in unknown proportion. MENIN is Bethyl A300-105A Lot 8; menin binds the fusion, wild-type MLL1 and non-KMT2A partners. Q117 could attribute its numbers because its fusion was FLAG-tagged; nothing here can. Every occupancy statement in this beat is about MLL1-N-reactive material or menin-occupied sites, never about the fusion. No scale factor exists for any sample. The depositors state that a Drosophila spike-in was run on one batch, that fly/human ratios were 'found to be small', and that NO spike-in normalisation was applied; no number for that measurement is reported and none is deposited. Every cross-condition figure here is therefore a WITHIN-SAMPLE share or a within-sample rank, which cancel an unknown per-sample scale factor exactly, and never a fold change. n = 1 per antibody per condition per line - there is no replicate anywhere in GSE127507 - so no number here carries a spread and none is precise. The mouse-side figures are NOT re-derived here: they are taken from Q117's stored locus table and hypothesis entry in this repository, and the mouse floor is Q117's Gapdh at 0.508, a TWO-GENE floor against this beat's ~8,600-window one. Instrument A is a 6 kb TSS window, so 'MYB' means the MYB TSS window and not the MYB gene body or its known upstream enhancers. Gene symbols are the deposited table's own RefGene annotation, not resolved through HGNC. MOLM13's MENIN samples are day 4 while everything else is day 3, so MOLM13's MENIN column is not exposure-matched. HOXA10 is absent from the deposited table so the mouse Hoxa10 figure has no human counterpart here.

    Evidence that argues against this (2)
    • The MYB result has a second reading this entry cannot exclude: RS4;11 and MOLM13 differ in lineage (B versus myeloid) as well as in fusion partner, and MYB is a lineage-dependent locus. 'Partner difference' and 'lineage difference' are confounded in a two-line comparison, and ML-2 does not resolve it (1.318, closer to RS4;11 than to MOLM13, and ML-2 is myeloid). Iqbal Lab oncofusion knowledge graph 2026
    • Q117's own murine locus figures are n = 1 within-sample shares from an unnormalised .tdf, exactly as these are, and Q117 recorded that its locus table is a comparison against what this graph records PMID 31821784 as saying rather than against the paper's figures. So the mouse side of every row above inherits that limitation. Iqbal Lab oncofusion knowledge graph 2026
Connected to