Every WRAD subunit, DOT1L, ENL, AF9 and LEDGF carries a cited histone-mark finding and none carries an epigenetics field - 12 of the 32 index-fusion core proteins are pure notation gaps, 9 are covered, and the 11 with nothing are all on the negative-regulatory side of elongation
also called q181 core axis audit · WRAD epigenetics field gap
For the 32 proteins at the heart of this project's disease, nine record their chromatin biology in a field a reader can filter on, twelve describe it only in the text, and eleven say nothing at all. Those eleven are not scattered - they are the brake that holds transcription in reserve, not the accelerator the fusion hijacks.
The same thing, for a scientist
The 39.2% headline invites the reading that the project's most important proteins are among the uncovered. Auditing the dispatch's own priority list - the retained KMT2A N-terminus partners, the elongation machinery, the WRAD components, menin and LEDGF, extended with the AFF and cyclin paralogues and the 7SK snRNP - gives 32 entries, all of which exist on disk. Nine carry a dedicated epigenetics field: KMT2A, AFF1, ELL, ELL2, EAF1, MEN1, CDK9, CTR9, HEXIM1. Twelve carry a cited mark finding in prose with no field: AFF4, DOT1L, MLLT1, MLLT3, MLLT10, WDR5, RBBP5, ASH2L, DPY30, PSIP1, PAF1, CDC73 - which is the whole of WRAD, the H3K79 writer, both YEATS readers and the chromatin tether. Eleven have neither: LEO1, AFF2, AFF3, ELL3, CCNT1, CCNT2, MED26, HEXIM2, LARP7, MEPCE, CDK12. So 21 of 32 (65.6%) fail the metric while only 11 of 32 (34.4%) lack the content, and the missing content is structured, not random.
The audit
All 32 ids were resolved against the live id set before classification; none is absent from disk 1.
| stratum | n | members |
|---|---|---|
| A -- field present | 9 | KMT2A, AFF1, ELL, ELL2, EAF1, MEN1, CDK9, CTR9, HEXIM1 |
| B -- cited mark in prose, no field | 12 | AFF4, DOT1L, MLLT1/ENL, MLLT3/AF9, MLLT10, WDR5, RBBP5, ASH2L, DPY30, PSIP1/LEDGF, PAF1, CDC73 |
| C -- nothing | 11 | LEO1, AFF2, AFF3, ELL3, CCNT1, CCNT2, MED26, HEXIM2, LARP7, MEPCE, CDK12 |
Stratum B here is the cleanest demonstration in this beat that the metric is measuring notation. WRAD is the H3K4 methyltransferase activation module and all four of its subunits -- WDR5, RBBP5, ASH2L, DPY30 -- sit in it. DOT1L, the H3K79 writer this project's therapeutic argument turns on, sits in it. Both YEATS-domain acyl-lysine readers, ENL and AF9, sit in it. LEDGF/PSIP1, whose PWWP domain reads H3K36me2/3 and tethers menin-KMT2A to chromatin, sits in it. Nobody would claim this project does not know the chromatin biology of WRAD.
The eleven with nothing share a shape
The C stratum is not a random eleven. Grouped:
- The 7SK snRNP brake -- HEXIM2, LARP7, MEPCE. HEXIM1, the fourth member, is the one that HAS a field.
- The P-TEFb cyclins -- CCNT1, CCNT2. CDK9, their catalytic partner, HAS a field.
- The AFF paralogues -- AFF2 (SEC-L2), AFF3 (SEC-L3). AFF1 and AFF4 are covered or prose-covered.
- The remainder -- LEO1 (PAF1C), MED26, ELL3, CDK12.
Within three separate complexes, the subunit with the field and the subunits without sit side by side. CDK9 covered / CCNT1 empty. HEXIM1 covered / HEXIM2, LARP7, MEPCE empty. AFF1 covered / AFF2, AFF3 empty. PAF1 and CDC73 prose-covered / LEO1 empty.
That pattern is diagnostic and it is not about biology. Coverage is following which protein a beat happened to be dispatched on, not which protein has published chromatin biology. A beat written about CDK9 records CDK9's epigenetic readout and leaves its obligate cyclin partner blank, because the cyclin was never the subject of a beat.
What this does and does not say about the menin-interface hypothesis
Nothing here threatens it. The menin arm is the best-covered part of the audit: gene-men1, gene-kmt2a, gene-aff1, gene-cdk9 and gene-ctr9 all carry fields, and gene-psip1 carries the H3K36me2/3-reading content in prose.
The one observation worth passing on is the reverse of a threat. The metric's worst-covered block is the negative-regulatory arm of elongation -- the 7SK reservoir and the cyclins that P-TEFb is released from. This project already holds protein-7sk-snrnp, noncoding-rn7sk, mechanism-ptefb-extraction-from-the-7sk-reservoir and mechanism-kmt2a-aff1-cannot-be-sequestered-in-the-7sk-snrnp, so the mechanism is written; what is missing is the per-protein chromatin annotation on the subunits. A coverage metric reading the gene entries alone cannot see the complex-level entries that already answer it, which is a further, separate reason the 39.2% understates.
| script | analyses/q181-is-the-epigenetics-gap-real/core.py |
|---|---|
| status | run |
| generated | 2026-09-24 |
| IDS resolved against disk | true |
strata
{
"field_present": [
"gene-kmt2a",
"gene-aff1",
"gene-ell",
"gene-ell2",
"gene-eaf1",
"gene-men1",
"gene-cdk9",
"gene-ctr9",
"gene-hexim1"
],
"prose_only_no_field": [
"gene-aff4",
"gene-dot1l",
"gene-mllt1",
"gene-mllt3",
"gene-mllt10",
"gene-wdr5",
"gene-rbbp5",
"gene-ash2l",
"gene-dpy30",
"gene-psip1",
"gene-paf1",
"gene-cdc73"
],
"nothing": [
"gene-leo1",
"gene-aff2",
"gene-aff3",
"gene-ell3",
"gene-ccnt1",
"gene-ccnt2",
"gene-med26",
"gene-hexim2",
"gene-larp7",
"gene-mepce",
"gene-cdk12"
]
}measurements — the numbers behind the chart
[
{
"quantity": "index-fusion core proteins failing the epigenetics field test",
"value": 21,
"units": "of 32",
"n": 32,
"subject": "the dispatch priority list plus the AFF/cyclin paralogues and the 7SK snRNP",
"universe": "32 ids, all confirmed present on disk before classification, commit ac2aeab",
"null": 329,
"note": "65.6% against the corpus-wide 60.8%; the core is NOT better covered than the corpus"
},
{
"quantity": "index-fusion core proteins lacking the CONTENT, not merely the field",
"value": 11,
"units": "of 32",
"n": 32,
"subject": "same 32-protein list, entries with neither an epigenetics key nor any mark token",
"universe": "same",
"null": null,
"note": "34.4% - and 8 of the 11 are the 7SK brake, the P-TEFb cyclins and the AFF paralogues"
},
{
"quantity": "WRAD subunits with a cited mark finding and no epigenetics field",
"value": 4,
"units": "of 4",
"n": 4,
"subject": "WDR5, RBBP5, ASH2L, DPY30 - the H3K4 methyltransferase activation module",
"universe": "the four WRAD components as entries in this corpus",
"null": null,
"note": "4 of 4 - the single sharpest case that the metric is counting field names"
}
]- Do the twelve stratum-B core entries state their own subject's chromatin biology, or the complex's? Only two were hand-read. WDR5 is the case most likely to be imported boilerplate about KMT2A and is worth reading first.
- Is the subunit-beside-subunit pattern (CDK9 covered / CCNT1 empty; HEXIM1 covered / HEXIM2, LARP7, MEPCE empty) an artefact of which proteins beats were dispatched on, or of which proteins have published chromatin biology? The dispatch history is in the king logs and was not consulted.
- The complex-level entries protein-7sk-snrnp and mechanism-ptefb-extraction-from-the-7sk-reservoir already carry chromatin content for proteins whose gene entries score zero. Should a per-protein coverage metric credit a linked complex entry, and if so by which edge types? Rule 89 says an edge may only count when the edge type can assert the claim, and part_of arguably can here.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
Of 32 index-fusion core proteins - KMT2A, AFF1, AFF4, ELL, ELL2, EAF1, DOT1L, MLLT1, MLLT3, MLLT10, WDR5, RBBP5, ASH2L, DPY30, MEN1, PSIP1, CDK9, PAF1, CTR9, CDC73, LEO1, AFF2, AFF3, ELL3, CCNT1, CCNT2, MED26, HEXIM1, HEXIM2, LARP7, MEPCE, CDK12 - all 32 exist as entries; 9 carry a data key check-pi-dimensions.py's _EPI_KEY matches, 12 carry a specific mark token with no such key (AFF4, DOT1L, MLLT1, MLLT3, MLLT10, WDR5, RBBP5, ASH2L, DPY30, PSIP1, PAF1, CDC73), and 11 carry neither (LEO1, AFF2, AFF3, ELL3, CCNT1, CCNT2, MED26, HEXIM2, LARP7, MEPCE, CDK12).
What this evidence cannot show
The 32-member list is the dispatch brief's priority list extended by me with the AFF paralogues, the cyclins, the 7SK components and CDK12; a different extension gives a different denominator, and the 9/12/11 split should be read as a description of this list rather than as a rate. Classification uses the detector's own two regexes, so stratum B membership inherits epi_text's known over-generosity - a mark token in an AFF4 entry may be a statement about KMT2A.
Evidence that argues against this (2)
- The claim that stratum B proves the metric measures notation rests on the assumption that these twelve entries state their subject's own chromatin biology, which was verified by hand for only two of them (gene-mllt1's YEATS-H3K27ac structure 5J9S, and gene-dot1l) as part of the parent sample. The other ten were classified by regex alone. WDR5's entry in particular could plausibly discuss H3K4 methylation entirely as a property of the KMT2A complex rather than of WDR5, which would make it the same imported-boilerplate failure the parent analysis found in gene-lmo2.
- The subunit-beside-subunit pattern (CDK9 covered, CCNT1 empty) is offered as evidence that coverage follows dispatch history rather than biology, but no dispatch history was consulted. An equally good explanation is that the covered subunit is in each case the one with more published chromatin biology - CDK9, HEXIM1 and AFF1 are all substantially better studied than their listed partners - in which case the pattern is about the literature and not about this project's process.
the core-axis audit reproduces the schema-artefact dominance on a pre-specified 32-protein list
the 8-member floor is a subset of the 11 core-axis entries with nothing; both isolate the P-TEFb reserve arm
- related to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
audits the epigenetics record of the 32 proteins on the index fusion's own elongation and WRAD axis