Oncofusion

start / graph / exp-q130-bin-the-rs4-11-arm-of-gse127507-and-score-the-index-fusions-own-eviction-gradient

Experimentstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q130-does-the-eviction-gradient-exist-in-h3k79me2

Download and bin GSE127507's nine RS4;11 tracks (1.99 GB), rebuild a human binner, and score the eviction gradient in the index fusion KMT2A::AFF1 - the one measurement that would move this finding from KMT2A::MLLT3 to the disease this graph is about

also called the RS4;11 run · the index-fusion experiment · 1.99 GB and about an hour

Download nine public files from a 2019 study of the exact fusion this project studies, count the signal in fixed-size windows, and repeat the analysis that was just done in two other cell types. No laboratory work is needed.

The same thing, for a scientist

**Design.** Exactly this beat's analysis, in RS4;11 (human KMT2A::AFF1), on GSE127507's deposited tracks [E1]: 1. **Rebuild a human binner.** `Q117Bin.java` survives in `C:/Temp/q117` and hardcodes mm10 chr1-19+X; the change is the chromosome list. `Q118Bin` is gone with `C:/Temp/q118`. 2. **Download, bin, delete, one at a time** - Q124's pattern, so peak disk is one TDF. Nine files, 1,989,180,464 bytes total, sizes measured by HEAD request and recorded per accession. 3. **Build the universe** from `GSM3635699` (RS4;11 input DMSO) > 0 and not in the hg19 ENCODE blacklist v2, already re-fetched to `C:/Temp/q130/q130_hg19_blacklist.bed`; hg19 chromosome sizes from `C:/Temp/q125/q125_hg19_chroms.json`. 4. **Run the per-file build check** - `abs_bp_distance_per_candidate` against hg19, hg38, mm10 and mm39 on each file's own declared per-chromosome extents. **Do not use `has_chr20`**; it is vacuous. 5. **Define two stratifiers**, both at matched d3 exposure, which MOLM13 could not offer: `S_MEN` = MENIN-DMSO top 0.1%, eviction ratio = MENIN VTP/DMSO share ratio; and `S_MLL1` = MLL1n-DMSO top 0.1%, same construction. Quartiles by **rank**. 6. **Score** H3K79me2 and DOT1L VTP/DMSO per quartile, with the **input as the negative control** stratified identically, the DMSO-decile-matched null (20 draws, seed 20260923) and the bin-level bootstrap (2,000 resamples). **Readout.** The four-quartile profile, its Spearman against eviction depth, its Q1-minus-Q4 span, the empirical p against the matched null, and the aggregate over the same set for comparison. **Controls, and what each is for** [E2]: - **Input stratified identically** - the negative control. It was flat in MOLM13 (Spearman +0.400, p = 0.90) and graded in all three murine inputs, so it must be reported whichever way it falls. - **DOT1L occupancy** - the writer. If it is graded and the mark follows, the MOLM13 mechanism transfers to the index fusion. - **Two stratifiers** - MENIN and MLL1-N must agree, as they did in MOLM13. - **NO drug-shape control is possible**: GSE127507 deposits no DOT1L-inhibitor arm in any line. **Cost.** 1.99 GB of download, a chromosome-list edit to an existing Java class, and roughly an hour of compute on the machine this beat ran on. **No wet work, no reagent, no cohort.**

Go deeper
What each outcome would mean, written before it is run
  • Spearman -1.000 with a span above 0.10 (the smallest span the bin-level bootstrap resolves at this bin count): the gradient is a property of menin inhibition and not of KMT2A::MLLT3, and every aggregate H3K79me2 readout in the index fusion is underpowered by construction. That is the result this graph would act on.
  • |Spearman| < 1.000 or a span below 0.10: the first pre-registered falsifier of hyp-q130 fires. The gradient is a property of KMT2A::MLLT3 systems, the two published systems it was found in share a fusion partner, and the finding does not transfer to the index disease. That is a real and publishable negative.
  • A graded INPUT: the result is uninterpretable in RS4;11 and the honest report is that the index-fusion line's bins are not comparable, not that the gradient is absent.
  • The two stratifiers disagreeing in sign: report both and score nothing, as Q125 did when its two H2AK119ub conventions disagreed.
Two cheaper things to run first, both at zero download cost

1. The DMSO-versus-DMSO selection control, which is hyp-q130's second falsifier and which cycle 83 also named as missing for the MLL1 channel. GSM8123372 against GSM8123386 are two DMSO H3K79me2 samples of the same murine culture, both already binned in C:/Temp/q122. Scoring one against the other over the same eviction quartiles gives the span that selection alone produces. If it exceeds 0.15 - a quarter of the observed VTP span - a large part of every gradient in this beat is selection. This is one script and no download and it should be run before the 1.99 GB.

2. The per-bin battery on Q125's three marks. Q125's gradients were reported as ratios of sums with a matched null and no per-bin statistic; deviation1_per_bin_robustness.py runs unchanged against C:/Temp/q125's tracks with four track-name substitutions.

Facts and figures
designre-run Q130's eviction-gradient analysis in RS4;11 (human KMT2A::AFF1) on GSE127507's nine deposited tracks, with two stratifiers, an input negative control, a DMSO-decile-matched null and a bin-level bootstrap
readoutfour-quartile H3K79me2 and DOT1L profiles, Spearman against eviction depth, Q1-minus-Q4 span, empirical p against the matched null, and the aggregate over the same set
controlsinput stratified identically (must be flat); DOT1L occupancy (positive control that the assay resolves a gradient); two independent stratifiers that must agree in sign; per-file genome-build check on declared extents; NO DOT1L-inhibitor arm is available
cost tierlow
timelineone working session: 1.99 GB download, a chromosome-list edit, about an hour of compute
download bytes1989180464
absence classmeasured-not-analysed
absence class noteThe data has been public since 2019 and the stratification has existed in this corpus since cycle 68. The obstacle is 1.99 GB and a deleted binner.
absence class assigned2026-09-23 (Q130)
epigenetic readoutH3K79me2 and DOT1L eviction-depth profiles in the index fusion, proposed
accessions
[
  "GSM3635697",
  "GSM3635698",
  "GSM3635693",
  "GSM3635694",
  "GSM3635695",
  "GSM3635696",
  "GSM3635691",
  "GSM3635692",
  "GSM3635699"
]
prerequisites
[
  "a human binner (edit Q117Bin.java's chromosome list to chr1-22+X)",
  "C:/Temp/q130/q130_hg19_blacklist.bed (already re-fetched, 834 intervals)",
  "C:/Temp/q125/q125_hg19_chroms.json (hg19 sizes, chr1 249250621)"
]
run these first at zero cost
[
  "the DMSO-versus-DMSO selection control: GSM8123372 against GSM8123386, both already binned in C:/Temp/q122, scored over the same eviction quartiles - hyp-q130's second falsifier",
  "the per-bin robustness battery on Q125's H2AK119ub, H3K27me3 and H3K4me3 tracks in C:/Temp/q125"
]
How this could bite us back
  • If the experiment returns a flat profile in the index fusion, the correct conclusion is that this beat's finding is a KMT2A::MLLT3 result and the pharmacodynamic proposal built on it does not apply to the disease this graph exists for. That outcome should be written up as prominently as a positive one, and the hypothesis entry's falsifier is worded so that it can be.
  • Running a 1.99 GB download to extend a finding is the kind of decision that is easy to justify after a positive result and hard before one. The two zero-cost checks named above should be run first precisely because one of them could make the download unnecessary.
Still open
  1. Should the DMSO-versus-DMSO selection control be run before the download? Yes - it costs nothing, it is a pre-registered falsifier of the hypothesis this experiment tests, and a large selection span would change what the RS4;11 result could mean.
  2. Is RS4;11's KMT2A::AFF1 breakpoint and lineage close enough to infant t(4;11) disease for the result to transfer? RS4;11 is an adult-derived B-lymphoblastic line and this graph's disease is infant; that gap remains after this experiment.
  3. Could the same design be run on the ML2 line, also in GSE127507? ML2 has DOT1L, MENIN, MLL1n and input pairs but NO H3K79me2 sample, so it can supply a writer gradient and not a mark gradient.
Every source behind this entry (2)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    Every track this experiment needs is deposited in GSE127507 for RS4;11, the human KMT2A::AFF1 index-fusion line - H3K79me2, MENIN, MLL1-N and DOT1L DMSO/VTP pairs plus an input, all at d3 exposure - totalling 1.99 GB, so the experiment is compute against public data with no wet work.

    What this evidence cannot show

    The RS4;11 arm has no SGC0946 sample, no second input and one replicate per antibody per condition, so the drug-shape control and any reproducibility estimate are unavailable in it. The stratifiers inherit the shared-epitope problem (MLL1-N) and the direct-target problem (MENIN). WITHIN-SAMPLE SHARE ONLY: the series ran a Drosophila spike-in, measured it and deliberately did not use it.

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

    The control set is transferable from this beat's human arm, where the input stratified identically was flat (Spearman +0.400, span 0.0120, empirical p = 0.90), DOT1L occupancy was graded at Spearman -1.000 and the two stratifiers MENIN and MLL1-N agreed in sign.

    What this evidence cannot show

    Transferring a control set from MOLM13 to RS4;11 assumes the two lines behave comparably in ways nothing establishes; RS4;11 is B-lymphoblastic and MOLM13 myeloid, and Q118 already measured different retained-occupancy values in the two.

    Evidence that argues against this (1)
    • 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
Connected to