An occupancy-matched floor built from ~8,600 mid-occupancy promoter windows per track closes a gap Q117 named in its own memo - and it was necessary, because GAPDH and ACTB GAIN share in two of three human lines while Q117 measured them LOSING share in mouse
also called the proper floor · GAPDH is not a control, measured twice in opposite directions
Comparing treated and untreated samples needs a reference point that does not move, and everyone uses one or two housekeeping genes. Those genes move - upward here, downward in last cycle's mouse data - so this beat built a reference from about eight thousand ordinary promoters instead.
The same thing, for a scientist
Q117's memo listed this as an open gap in its own work: *"A proper floor. Gapdh and Actb are two genes. The floor should be the distribution of the share ratio over several thousand expression-matched, fusion-unbound promoters."* **This beat built it, and it turned out to be necessary rather than tidy** [E1]. **The floor.** For each of the eighteen tracks in the deposited TSS table, the aggregate within-sample share ratio over every window in DMSO percentile 50-90 - occupancy-matched, n = **8,579 to 8,598** windows per track: | track | aggregate floor | per-window median | zero fraction among floor windows | |---|---:|---:|---:| | RS4;11 MLL1-N | **0.981** | 0.867 | 0.014 | | MOLM13 MLL1-N | **1.048** | 1.026 | 0.001 | | ML-2 MLL1-N | **1.007** | 0.947 | 0.026 | | RS4;11 MENIN | 0.945 | 0.772 | 0.168 | | MOLM13 MENIN | 0.934 | **0.000** | **0.548** | | ML-2 MENIN | 1.007 | 0.886 | 0.162 | | RS4;11 DOT1L | 1.086 | 0.743 | 0.311 | | MOLM13 DOT1L | 0.920 | 0.832 | 0.120 | | ML-2 DOT1L | 0.982 | 0.566 | 0.362 | **All three MLL1-N floors are within 5% of 1.0**, which is the reassuring result: the MLL1-N share distribution does not globally drift, so a locus-level figure can be read as a local effect. **Why the two-gene floor had to go.** GAPDH and ACTB, floor-normalised [E1]: | | GAPDH | ACTB | |---|---:|---:| | RS4;11 MLL1-N | **1.763** | **2.649** | | ML-2 MLL1-N | 1.548 | 1.767 | | MOLM13 MLL1-N | 0.770 | 0.743 | | **murine FLAG-MLLAF9 (Q117, raw)** | **0.508** | **0.571** | **The housekeeping floor points in opposite directions in the two datasets** - down two-fold in mouse, up 1.8 to 2.6-fold in human RS4;11 - so a locus table normalised to it would move by a factor of 3.5 depending on which side you trusted. Q117's own surprise list recorded *"GAPDH is not a valid internal control here. It loses half its share under the drug"*; **this beat finds the same conclusion from the opposite observation**, which is the strongest form the conclusion can take. **And the floor had to be the AGGREGATE, 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** - so a median-based normaliser would have made every MOLM13 menin locus figure undefined, and a ratio gate built on it would have divided by zero. The aggregate's denominator is a sum of shares over ~8,600 windows and is asserted non-zero before use. Recorded as this beat's DEVIATION 2/4.
| script | analyses/q118-menin-at-the-human-index-fusion/roseB.py (FLOOR block) |
|---|---|
| status | run |
| result summary | Occupancy-matched floors over ~8,600 windows per track are 0.92-1.09 for all nine tracks; GAPDH/ACTB are 1.76/2.65 in RS4;11 against 0.508/0.571 in Q117's mouse data, so the housekeeping floor moves in opposite directions in the two datasets. |
| floor definition | aggregate within-sample share ratio over all deposited TSS windows in DMSO percentile 50-90, per track |
| n floor windows | 8,579 to 8,598 per track |
| genome build | hg19 |
| cell line | RS4;11, MOLM13, ML-2 |
inputs
[ "GSE127507 ROSE TSS signal table", "Q117 locus-table.json" ]
- The floor is occupancy-matched and not expression-matched, which is what Q117 asked for. The GSE127508 SuperSeries contains RNA-seq for these same lines and conditions; joining it would give a genuinely expression-matched floor and was not done.
- Why do GAPDH and ACTB gain share in RS4;11 and ML-2 and lose it in MOLM13 and in mouse? If the answer is that the housekeeping promoters absorb signal released from evicted sites, then their behaviour is a readout of how much was evicted, which would make them informative in exactly the way they are not usable as controls.
- A mid-occupancy floor near 1.0 may be too weak a normaliser to detect a true global drift. A second, independent floor - for instance the share over intergenic bins with input above median - would test that, and the binned genome-wide data on disk supports it.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Occupancy-matched aggregate floors over 8,579-8,598 TSS windows in DMSO percentile 50-90 are 0.981 / 1.048 / 1.007 for MLL1-N, 0.945 / 0.934 / 1.007 for MENIN and 1.086 / 0.920 / 0.982 for DOT1L in RS4;11 / MOLM13 / ML-2; against those floors GAPDH is 1.763 and ACTB 2.649 in RS4;11 MLL1-N while Q117 measured murine Gapdh at 0.508 and Actb at 0.571; and MOLM13 MENIN's per-window median floor ratio is exactly 0.000 because that track is 58.7% zeros.
Q118 re-analysis of GSE127507, instrument A: the deposited series-level ROSE TSS-associated signal table, 21,461 windows x 18 samples; analyses/q118-menin-at-the-human-index-fusion/results-instrumentA.json and results-instrumentA-addendum.json, run 2026-09-23 · Iqbal Lab oncofusion knowledge graph 2026What this evidence cannot show
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 floor is OCCUPANCY-matched (DMSO percentile 50-90) and not EXPRESSION-matched, which is what Q117's memo asked for - no RNA-seq was joined in this beat, and the GSE127508 SuperSeries does contain RNA-seq subseries that were not touched. 'Fusion-unbound' is also not established for the floor windows: they are mid-occupancy, which is not the same as unbound. The murine GAPDH/ACTB figures are Q117's stored raw values and were not re-derived.
Evidence that argues against this (1)
- A floor of ~1.0 could mean the distribution does not drift, or it could be a construction artefact: the share over a large subset of a universe is bound to be near 1 if the universe's shape is broadly preserved, so a mid-occupancy floor is close to tautologically near 1 and may therefore be too weak a normaliser to detect a genuine global drift. A spike-in would settle it and there is none. Iqbal Lab oncofusion knowledge graph 2026
the comparison is only readable against an occupancy-matched floor, because GAPDH and ACTB move in opposite directions in the two datasets
closes a gap Q117's own memo named - a floor from several thousand promoters rather than two housekeeping genes
the same argument applied to the mouse side: an aggregate floor over 505,306 bins replaces Gapdh at 0.508 and moves every mouse-side normalised figure by 1.83-fold