The KMT2A C-terminal site class gives VTP-50469/DMSO = 1.13 - an apparent GAIN of H3K4me3 - and its IgG negative control sits at 2.4-2.6, so the class is background and the number must not be reported as a result
One of our three site definitions produced an eye-catching answer in the opposite direction. The no-antibody control showed the same pattern, which means that definition was picking out places where any experiment accumulates data rather than places the protein actually is. We are recording the number together with the reason it cannot be used.
The same thing, for a scientist
Three site classes were built identically - signal over a matched IgG control, classes matched within DMSO H3K4me3 deciles. **Menin-over-IgG: IgG bound/control 0.076-0.078 (clean), VTP/DMSO 0.766. KMT2A N-over-IgG: IgG 0.135-0.167 (clean), VTP/DMSO 0.871. KMT2A C-over-IgG: IgG 2.36-2.56 (FAILED), VTP/DMSO 1.132.** An IgG library enriched 2.4-fold in a class that is supposed to mean 'KMT2A C-terminus is here' means the class is largely accessibility, and the 1.132 is therefore not evidence of an H3K4me3 gain at C-terminal sites under a menin inhibitor.
Why the failing class is the interesting one to publish
The 1.132 is the only number in this beat that points against the menin-interface reading, and it is the one we are discarding. That asymmetry deserves stating plainly: we discarded the inconvenient number, and the reason is a control value printed beside it. The defence is that the same control was applied to all three classes before any of them were read, that it is the same control that killed a convenient null in ana-q140-a-menin-high-bin-class-is-an-accessibility-class-until-you-divide-by-igg, and that the threshold is not a judgement call - a negative control cannot be 2.4-fold enriched in a class defined by a real antibody's specificity.
Why this class in particular fails
The C-terminal KMT2A antibody's libraries are the weakest of the four KMT2A libraries in this line, and the AutoCUT&RUN IgG control from the same batch (GSM4842303) is unusually deep at 9.7M fragments - deeper than three of the four KMT2A libraries it controls. A deep background library and a weak specific library make an enrichment ratio that is dominated by the background's own structure. The N-terminal classes, built against the same IgG, pass - so this is a property of the C-terminal libraries and not of the control.
| script | analyses/sem-arm-identity-from-signal/run3.py |
|---|---|
| status | run |
| pass criterion | the IgG bound/control ratio must be below 1; a negative control cannot be enriched in a specificity-defined class |
classes — the numbers behind the chart
{
"menin_over_IgG": {
"IgG_control": [
0.0763,
0.0781
],
"VTP_over_DMSO": 0.7657,
"verdict": "ACCEPTED"
},
"KMT2A_Nterm_over_IgG": {
"IgG_control": [
0.1669,
0.1352
],
"VTP_over_DMSO": 0.8707,
"verdict": "ACCEPTED"
},
"KMT2A_Cterm_over_IgG": {
"IgG_control": [
2.5567,
2.3569
],
"VTP_over_DMSO": 1.1321,
"verdict": "REJECTED - negative control enriched"
}
}- A discarded result that points against the house hypothesis is exactly the kind of decision that should be re-examined by someone else. The raw values are in the data payload precisely so that a sceptic does not have to re-run anything to challenge the call.
- Would the C-terminal class pass with a protocol-matched AutoCUT&RUN IgG used for both definition and testing? That is one re-run and it would turn an uninterpretable number into either a result or a clean rejection.
- How many site classes elsewhere in this graph would fail this test if it were applied retrospectively?
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Built identically to the two accepted contrasts, the KMT2A C-terminal site class gives an IgG bound/control ratio of 2.557 and 2.357 and a VTP/DMSO value of 1.132; the menin and KMT2A N-terminal classes give IgG ratios of 0.076-0.078 and 0.135-0.167.
sem-arm-identity-from-signal: run3.py and results3.json (this repository) · Q140-sem-menin-arm · Oncofusion knowledge graph 2026What this evidence cannot show
The IgG libraries used as the pass/fail test are the AutoCUT&Tag ones (GSM5139717/18), while the class was defined against the AutoCUT&RUN IgG (GSM4842303). A protocol mismatch between the defining control and the testing control is a real limitation, and it is why this is reported as 'uninterpretable' rather than 'refuted'.
Evidence that argues against this (1)
- It is possible that H3K4me3 genuinely rises at KMT2A C-terminal sites under a menin inhibitor - the C-terminus carries the SET domain and a redistribution is not absurd. We are not claiming otherwise; we are claiming this measurement cannot support it. Distinguishing them needs a C-terminal library with a protocol-matched control. Krivtsov AV et al. 2019
The same IgG-in-the-class test that validated the menin class rejected the KMT2A C-terminal one
The third site class fails its own negative control and is excluded from the result