This graph recorded 'nobody has looked at AKAP95 in any KMT2A fusion' one cycle before reading the 2026 paper that did - KMT2A returns 0, MLL returns 2, and every cell-line token returns 0 because cell lines live in figures
also called AKAP8 KMT2A nomenclature panel · q166 absence audit
We wrote down that no one had studied AKAP95 in MLL-rearranged leukaemia. A paper published in 2026 does exactly that. Searching for the modern gene name found nothing; searching for the old name found it immediately.
The same thing, for a scientist
**The corpus entry `interface-kmt2a-akap8-a-prey-only-grade-on-a-direct-mll-coactivator` records `shown_for_fusion: 'NO - nobody has looked in any KMT2A fusion'` and `absence_class: 'never-measured'`, written 2026-09-24 in cycle 146. PMID 41501053 (*Nat Commun*, 2026) is titled 'AKAP95 condensates regulate transcription and can be targeted in MLL-fusion driven oncogenesis' and reports an Akap95-null mouse MLL-AF9 transplant, shRNA growth curves in THP-1, MOLM-13 and MV4-11, ChIP-seq, and a therapeutic peptide** [E1][E2]. **Three independent instrument failures made it invisible, and all three are already rules in this project's brief.** (1) **Nomenclature**: `AKAP8/AKAP95 AND KMT2A` = **0**; the same query with `MLL` = **2**, `MLL1` = **2**, `MLLr` = **1**, `MLL-rearranged` = **1**, `MLL-AF9` = **1**. (2) **Abstract-level bound**: **every** agent-and-model token - MV4-11, MOLM-13, RS4;11, SEM, THP-1, NOMO-1, revumenib, ziftomenib, pinometostat, MI-503 - returns **0** against the abstract index, because the paper names its cell lines only in the body. (3) **Spelling inside the full text**: `CITES:23995757_MED AND BODY:"MV4-11"` = **0**; `BODY:"MV411"` = **1**, the same paper. **The instrument that worked was neither: the whole AKAP8/AKAP95 literature is 92 records, and reading all 92 titles found it in one pass** [E3].
The translated list, as the rules require
Every count below is from scripts/pubmed_safe.py, so no zero is a silent instrument failure, and every querytranslation was read back. Operand: (AKAP8[tiab] OR AKAP95[tiab]). Bare-operand control: AKAP8[tiab] = 30, AKAP95[tiab] = 66, union = 92, KMT2A[tiab] = 1654 - so both operands are alive and the zeros below are about the conjunction, not about a dead term.
| second term | count | translation (abbreviated) |
|---|---|---|
| KMT2A | 0 | "KMT2A"[Title/Abstract] |
| MLL | 2 | "MLL"[Title/Abstract] |
| MLL1 | 2 | "MLL1"[Title/Abstract] |
| MLL-rearranged | 1 | "MLL-rearranged"[Title/Abstract] |
| MLLr | 1 | "MLLr"[Title/Abstract] |
| MLL-AF9 | 1 | "MLL-AF9"[Title/Abstract] |
| MLL-r | 0 | "MLL-r"[Title/Abstract] |
| MLL-AF4 | 0 | |
| "KMT2A::AFF1" | 0 | "kmt2a aff1"[Title/Abstract] |
| AFF1 | 0 | |
| "t(4;11)" | 0 | "t 4 11"[Title/Abstract] - quoted per rule 95, so this zero is genuine |
| 11q23 | 0 | |
| leukemia OR leukaemia | 3 | |
| menin | 0 | |
| MV4-11 / MV4;11 / MV411 | 0 | |
| MOLM-13 / MOLM13 | 0 | |
| "RS4;11" / RS4-11 | 0 | both translate to "rs4 11" |
| SEM AND leukemia | 0 | |
| THP-1 / THP1 | 0 | |
| NOMO-1 / NOMO1 | 0 | |
| revumenib / SNDX-5613 / VTP-50469 | 0 | |
| ziftomenib / KO-539 | 0 | |
| pinometostat / EPZ-5676 / EPZ5676 | 0 | |
| MI-503 / MI503 / MI-2-2 | 0 | |
| ZNF384 | 1 | the AKAP8::ZNF384 fusion review |
| DPY30 | 1 | |
| WDR5 | 0 | |
| ASH2L | 0 |
Read the cell-line block. Six cell-line tokens, all zero, in a literature that contains a paper reporting knockdown growth curves in three of them. The agents-and-models heuristic - query drug names and cell-line names rather than disease terms - is correct in intent and gives a FALSE ZERO when run against the abstract index, because a cell line is a Methods and figure object. This is rule 80 in its purest form.
And KMT2A vs MLL is 0 vs 2 in a 92-record literature. The project's own memory rule says hit sets can be disjoint. Here they are: the single most important paper carries MLL in its title and KMT2A nowhere in its abstract.
The full-text route, and where it also failed
Europe PMC over the citation network of the 2013 paper (CITES:23995757_MED, 52 citing records) is a stronger instrument and it does find the paper:
| query | hits | finds PMID 41501053? |
|---|---|---|
CITES:23995757_MED | 52 | yes, 4th |
... AND BODY:"leukemia" | 12 | yes, 1st |
... AND BODY:"MLL-AF9" | 6 | yes |
... AND BODY:"MLL-AF4" | 6 | yes |
... AND BODY:"AFF1" | 2 | yes |
... AND BODY:"KMT2A" | 9 | NO |
... AND BODY:"THP-1" | 2 | yes |
... AND BODY:"MOLM-13" | 1 | NO |
... AND BODY:"MV4-11" | 0 | NO |
... AND BODY:"MV411" | 1 | yes |
Two of these are worth keeping.
BODY:"KMT2A"over the full text still misses it. The paper uses MLL nomenclature throughout - MLL1T, MLL-AF9, MLLr - and evidently never writes KMT2A anywhere. Moving from abstract to full text does not rescue a nomenclature miss. The two failure modes are orthogonal and you need both fixes.MV4-11= 0 andMV411= 1 on the same nine-paper set. The hyphen-vs-no-hyphen hazard is already recorded forVTP-50469/VTP50469in rule 72 clause 3 as a drug-code hazard. It applies identically to cell lines, anddata/symbol-collisions.jsondoes not say so - the RS4;11 row covers the semicolon, nothing covers MV4-11 / MV411 / MV4;11. This beat recommends adding that row.
What actually worked was enumeration. AKAP8[tiab] OR AKAP95[tiab] returns 92 records. Fetching all 92 titles and abstracts and reading them took one efetch call. When a literature is under a few hundred records, enumerate it; do not bound it. A bounded absence is an inference from a query, and this cycle produced four separate queries whose zeros were all false.
What survives as a genuine absence, stated at the right level
After enumerating all 92 records and reading the two relevant full texts, these absences stand:
- No AKAP95 experiment in a KMT2A::AFF1 in vivo model, primary patient sample, or any KMT2A::AFF1 chromatin assay. The index fusion appears once, as an shRNA growth curve in MV4-11.
- No AKAP95 experiment combined with any menin inhibitor or DOT1L inhibitor, in any system. The drug-name panel returns 0 at abstract level and nothing in the 92-record enumeration or in the 52-paper citation network combines them.
- No AKAP95 HMT assay with an MLL1/KMT2A complex, thirteen years after the authors named it as the obvious next experiment.
- No structure of AKAP95 with any MLL-complex subunit, and no Kd for any of these contacts.
These are bounded by an enumerated literature plus two full texts, not by a keyword zero.
| script | analyses/q166-the-akap95-dpy30-axis/ (pubmed_nomen.json, epmc_citenet.json, epmc2.json, akap-corpus-92-titles.json) |
|---|---|
| status | run |
| result summary | A bounded absence in this corpus was false. Three orthogonal instrument failures hid the refuting paper: gene-symbol nomenclature (KMT2A 0 vs MLL 2), abstract-level bounding of cell-line tokens (all 6 zero, in a paper that uses three of them), and a hyphenation difference inside the full text (MV4-11 0 vs MV411 1). Enumerating the 92-record literature found it in one pass. |
inputs
[ "PubMed esearch via scripts/pubmed_safe.py", "Europe PMC REST search", "PubMed efetch" ]
measurements — the numbers behind the chart
[
{
"quantity": "count",
"subject": "PubMed records for (AKAP8 OR AKAP95)[tiab] AND KMT2A[tiab]",
"value": 0,
"units": "records",
"n": null,
"universe": "PubMed, 2026-09-24, translation '(\"AKAP8\"[Title/Abstract] OR \"AKAP95\"[Title/Abstract]) AND \"KMT2A\"[Title/Abstract]'",
"null": null,
"note": "genuine zero by the pubmed_safe signature; and FALSE as a statement about the literature"
},
{
"quantity": "count",
"subject": "PubMed records for the same operand with MLL[tiab] instead of KMT2A[tiab]",
"value": 2,
"units": "records",
"n": null,
"universe": "PubMed, 2026-09-24, same operand",
"null": 0,
"note": "the null here is the KMT2A count; same question, disjoint hit sets"
},
{
"quantity": "count",
"subject": "cell-line and drug tokens in the agents-and-models panel returning zero at abstract level",
"value": 10,
"units": "tokens",
"n": 10,
"universe": "MV4-11/MV4;11/MV411, MOLM-13/MOLM13, RS4;11/RS4-11, SEM+leukemia, THP-1/THP1, NOMO-1/NOMO1, revumenib group, ziftomenib group, pinometostat group, MI-503 group",
"null": null,
"note": "10 of 10. Three of those cell lines carry published AKAP95 knockdown growth curves. The heuristic is right and the index is wrong."
},
{
"quantity": "count",
"subject": "records in the complete AKAP8/AKAP95 title-abstract literature",
"value": 92,
"units": "records",
"n": null,
"universe": "PubMed, AKAP8[tiab] OR AKAP95[tiab], 2026-09-24",
"null": null,
"note": "small enough to enumerate; enumeration is what worked"
},
{
"quantity": "count",
"subject": "papers citing PMID 23995757 in Europe PMC",
"value": 52,
"units": "records",
"n": null,
"universe": "CITES:23995757_MED, Europe PMC, 2026-09-24",
"null": null,
"note": "12 with BODY:leukemia, 6 with BODY:MLL-AF9, 9 with BODY:KMT2A - and the key paper is in the leukemia set but NOT in the KMT2A set"
}
]- This analysis licences a correction to a published corpus entry. If the correction is applied mechanically - flipping shown_for_fusion from NO to YES - it will overstate what the 2026 paper shows for the INDEX fusion, which is one growth curve in one cell line. The right correction narrows the absence, it does not delete it.
- How many other 'never-measured' absence_class fields in this corpus were written from a KMT2A-only query? The MLL/KMT2A disjointness is a property of the field's writing habits, not of AKAP8, so the exposure is corpus-wide and measurable with one script.
- Should `data/symbol-collisions.json` gain a cell-line hyphenation row? MV4-11 / MV4;11 / MV411 are three tokens, the registry covers only the RS4;11 semicolon, and the hyphen hazard cost a full-text query its only hit in this beat.
- Is there a size threshold below which a beat should enumerate rather than bound? 92 records took one efetch. At what n does that stop being the better instrument - and is the current default (bound, then caveat) ever correct for a literature under 500?
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
A 29-row nomenclature panel run through scripts/pubmed_safe.py on the operand (AKAP8[tiab] OR AKAP95[tiab]) returns 0 for KMT2A, 2 for MLL, 2 for MLL1, 1 for MLL-rearranged, 1 for MLLr, 1 for MLL-AF9, 3 for leukemia/leukaemia, and 0 for every one of MV4-11/MV4;11/MV411, MOLM-13/MOLM13, RS4;11/RS4-11, SEM AND leukemia, THP-1/THP1, NOMO-1/NOMO1, revumenib/SNDX-5613/VTP-50469, ziftomenib/KO-539, pinometostat/EPZ-5676/EPZ5676 and MI-503/MI503/MI-2-2. Bare-operand controls: AKAP8 30, AKAP95 66, union 92, KMT2A 1654. Every querytranslation was read back and is recorded; no query returned an empty translation.
What this evidence cannot show
These are this project's own search results, not the contents of any paper - support_location is 'this-project' for exactly that reason. The counts are a snapshot of PubMed on 2026-09-24 and will drift. The [tiab] tags are applied per term, never as (A OR B)[tiab], per rule 94, and every cytogenetic term is quoted, per rule 95 - the t(4;11) row translates to 't 4 11'[Title/Abstract] and its zero is therefore genuine rather than a UID-filter artefact.
Evidence that argues against this (1)
- The panel's headline contrast - KMT2A 0 versus MLL 2 - rests on a two-record difference in a 92-record literature, which is not a general measurement of nomenclature drift. A reader should not generalise a rate from it. What it establishes is existence: at least one load-bearing paper is invisible to the modern symbol. The corpus-wide exposure is unmeasured and is listed as an open question. this project 2026
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
PMID 41501053 (Yan et al., Nat Commun 2026, PMC12886973) reports that AKAP95 is required for MLL-rearranged leukaemogenesis: shRNA knockdown reduces growth of THP-1, MOLM-13 and MV4-11 (written MV411) but not K562 or HL-60; Akap95-/- marrow transformed with MLL-AF9 gives longer survival and fewer serial-replating colonies; Akap95-/- marrow reconstitutes normally in competitive transplant; AKAP95 IDR1 (aa101-390) co-condenses with the MLL1 translocated fragment; and a designed peptide JD-PI95 bridging AKAP95 to HSP70 reduces THP-1 growth in vitro and tumour burden in NCG mice.
AKAP95 condensates regulate transcription and can be targeted in MLL-fusion driven oncogenesis. · Yan X, Luan Y, Zhu Y, Cai Y, Ye F, Yang L, Zhu K, Hu J, Luo S, Lin J, Wang Y, Huang Y, Sun P, Yang Z, Shi B, Jiang H, et al. · Nat Commun 2026What this evidence cannot show
Full text read via PMC12886973; supplementary figures not available. The expected_absent_terms entry is the finding: a Europe PMC full-text query for BODY:"KMT2A" over this paper's citation set does not return it, consistent with the paper never using the modern symbol.
Evidence that argues against this (1)
- This paper refutes the absence but does not fully close it for the index fusion. KMT2A::AFF1 appears once, as an shRNA growth curve in MV4-11; there is no KMT2A::AFF1 in vivo model, imaging, co-IP, ChIP or patient sample. A mechanical flip of shown_for_fusion from NO to YES would overstate what was measured. Yan X et al. 2026
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
The complete AKAP8/AKAP95 title-and-abstract literature is 92 PubMed records. Fetching all 92 and keyword-scanning titles and abstracts for leukaemia, KMT2A, MLL, 11q23, myeloid, lymphoblast, AML, blast, haematopoiesis and lymphoma returns 14 records, of which PMID 41501053 is the first by relevance and PMID 37306722 (the ZNF384 fusion review naming AKAP8 as a partner) the second.
analyses/q166-the-akap95-dpy30-axis/akap-corpus-92-titles.json - complete efetch of all 92 records matching AKAP8[tiab] OR AKAP95[tiab], with titles and years, 2026-09-24 · this project 2026What this evidence cannot show
Enumeration bounds the TITLE/ABSTRACT literature, not the literature. A paper that studies AKAP95 without naming it in the title or abstract - for instance a proteomics screen reporting it in a supplementary table - is outside this set, and rule 80 says such papers exist. The 92 is a complete enumeration of a defined index, not of the field.
Evidence that argues against this (1)
- This enumeration demonstrably misses at least one relevant source of its own kind: the MLL-fusion-protein interactome that first flagged AKAP95 against MLL-AF9, MLL-EEN, MLL-ENL and MLL-GAS7 is cited as reference 20 of PMID 41501053 and is certainly not an AKAP8/AKAP95 title-abstract paper. So enumerating a symbol's own literature recovers the papers ABOUT the protein and not the screens that FOUND it, which is the same blind spot that produced the original false absence. Yan X et al. 2026
its shown_for_fusion 'nobody has looked in any KMT2A fusion' and absence_class 'never-measured' are refuted by PMID 41501053
the enumeration that found the leukaemia literature, and the three instrument failures that hid it
the paper the absence missed
uses the same enumeration, and inherits its cell-line panel as the control showing why a keyword zero alone is worthless here