Rule 80 measured: the abstract index missed the best KMT2A measurement for three marks out of seven, and confirmed the absence for the other four
also called marks2 sweep · rule 80 measurement · abstract versus full text histone marks
We searched paper titles and abstracts for seven histone tags in KMT2A-fusion leukaemia, then opened nineteen of the actual papers and searched inside them. For three of the seven tags, the single best measurement was inside a paper the abstract search could not find.
The same thing, for a scientist
This beat ran each of seven marks and variants - H3K27ac, H4K16ac, H3K4me2, H3K9me3, H3K9ac, H3.3 and H2A.Z - as a bounded PubMed `[tiab]` intersection against a fixed KMT2A term set, with both operands counted and every querytranslation read back, and then swept **nineteen full texts** of the core KMT2A-rearranged chromatin literature with word-boundary matching. **Rule 80 changed the answer for three of seven.** For H3K27ac the decisive KMT2A::AFF1 experiment (PMID 31243293, H3K27ac ChIP-seq in SEM under DOT1L inhibition, 49 in-text occurrences) is absent from its own bounded query; for H3K9me3 the only KMT2A::AFF1 measurement is in Supplementary Fig. 4b-e of the same paper; for H4K16ac the only locus-resolved observation in a KMT2A-rearranged cell is two sentences and a supplementary panel in PMID 25822366, a paper titled for SIRT1 and DOT1L. For H3.3 the sweep found a use the abstract query missed but it was a **reagent** (an H3.3 K27M transgene in K-562) rather than a measurement. For H3K4me2, H3K9ac and H2A.Z the sweep **confirmed** the abstract-level picture, which is the result that makes the method a measurement rather than a rescue device.
The table
Fixed KMT2A term set: (MLL-rearranged OR KMT2A-rearranged OR MLL-AF9 OR MLL-AF4 OR KMT2A OR "MLL fusion")[tiab] = 3,209 records on 2026-09-20.
| mark / variant | mark term alone | ∩ KMT2A (abstracts) | best KMT2A measurement in the ∩? | rule 80 changed it? |
|---|---|---|---|---|
| H3K27ac | 3,116 | 14 | no - PMID 31243293 absent | yes |
| H3K9me3 | 2,793 | 10 | no - PMID 31243293 absent | yes |
| H4K16ac | 364 | 3 | no - PMID 25822366 absent | yes |
| H3.3 (+4 spellings) | 1,823 | 1 | n/a - the one hit is off-topic | yes, but it found a reagent, not a measurement |
| H3K4me2 | 926 | 9 | yes - PMID 24769646 present | no |
| H3K9ac | 1,380 | 4 | yes - PMID 25822366 present | no |
| H2A.Z (+7 spellings) | 1,233 | 1 | yes - PMID 33967269 is the only paper there is | no |
Three of seven. And the three are not random: they are the marks whose measurements are ChIP panels supporting a claim about a different molecule. The four where the abstract index was adequate are the ones where the mark is either the paper's subject (H3K4me2 at RARalpha, H3K9ac under SIRT1) or where the literature is a single paper (H2A.Z).
What the sweep covered, so the bound is readable
Nineteen documents, all read as stripped full text and checked for length to exclude body-less records:
MLL-AF4 / KMT2A::AFF1 chromatin (6): PMC2607073 (Guenther 2008), PMC5263239 (Kerry 2017), PMC5333536 (Godfrey 2017), PMC6594956 (Godfrey 2019), PMC7787973 (Godfrey 2021), PMC10457349 (Crump 2023), PMC8256865 (Harman 2021) - seven with Harman. Myeloid KMT2A chromatin (5): PMC5474565 (Prange 2017), PMC4390532 (Chen 2015), PMC5955148 (Ropa 2018), PMC7556517 (Ropa 2020), PMC7728597 (Chu 2020). Acetyltransferase and H3K4 (3): PMC5501293 (Valerio 2017), PMC4120120 (Mishra 2014), PMC4003419 (Sakamoto 2014). Menin-inhibitor chromatin (3): PMC7614190, PMC11561541, PMC12632788. Context (1): PMC4119022 (paediatric epigenetic mutation landscape).
What it is not. A purposive sample of papers this graph already cites, weighted to open access. Supplementary spreadsheets, figure source data and methods appendices outside the retrieved article body were not searched, so a mark named only in a supplementary antibody table would still be missed. A nineteen-document bound is stronger than an abstract count and is still a bound.
Two method traps found while running it
1. Substring matching invents chaperones. A naive substring count for HIRA across the nineteen documents returned 9; word-boundary matching returned 0. Every hit was an author surname - Hirashima, Hirai. A histone-chaperone presence claim built on substring counting would have reported HIRA in papers that never mention it. The same exposure applies to MOZ (already logged in data/symbol-collisions.json), to SET, and to any short histone or chaperone token. Count with word boundaries and check what matched.
2. fullTextXML is not the fastest route and sometimes not a route at all. Europe PMC's fullTextXML endpoint returned HTTP 503 for quoted-phrase searches and was slow or unavailable for several PMCIDs; the PMC HTML page (https://pmc.ncbi.nlm.nih.gov/articles/<PMCID>/) succeeded for all nineteen and was faster. This is consistent with the brief's four-route rule and adds an operational note: when sweeping many documents, go to the HTML route first and keep fullTextXML as the fallback, not the other way round.
3. Route 5 returns figure titles even when it does not return the body. For PMID 33967269 (paywalled, no PMCID) the DOI fetch returned 19,501 stripped characters consisting of the abstract, the six figure titles and the reference list. Figure titles alone established that the measured quantity is the ratio of acetylated H2A.Z to total H2A.Z at the Hoxa9 locus - a resolution the abstract does not give. A partial route-5 result is not a failure; read what came back.
Read and not used, with reasons
Per the brief's bounded-rejection standard, the on-topic-looking records that were fetched and rejected:
- PMID 38432146 (aberrant H3K4me3 in CD4+ T cells in systemic lupus) - matched the KMT2A x H3K27ac intersection on an incidental KMT2A mention; autoimmune disease, no leukaemia.
- PMID 30232013 (PRMT6-dependent H3R2 methylation) - matched on H3K27ac; a different residue and no fusion biology.
- PMID 41772004 (VSIR characterisation across haematological malignancies) - the single hit of the menin-inhibitor x seven-mark query; a bioinformatic survey with no chromatin measurement. It is the reason that query's count is 1 rather than 0 and it does not make the absence less real.
- PMID 39806204 (Kmt2c/Kmt2d loss in urothelium redistributes KMT2A-menin to bivalent promoters) - genuinely interesting for the menin arm and squarely in another beat's territory; urothelial, not leukaemia.
- PMIDs 38173411, 37767575, 35836814, 28411381, 23990460 - the H3K9me3 x KMT2A intersection hits that mention the mark without measuring it at fusion-bound loci (adipose stem cells, TWIST1, KDM4 inhibitor, PBX3, CXXC domains).
- PMIDs 42096476, 39909446, 37345101, 35525905, 32599142, 24619213 - the H3K4me2 x KMT2A hits that are not about this mark in this disease (rRNA transcription, a bone-marrow-failure case report, prostate cancer, a WDR5 PROTAC, nucleus pulposus, mouse preimplantation).
- PMID 26847026 (GFI1 in AML/MDS) and PMID 42523331's condensate interpretation - the first rejected as not a mark measurement, the second used for its H3K9ac and ENL-displacement observations and explicitly not for its phase-separation claim, per this project's condensate standard.
- PMID 22464800 (Harris 2012, KDM1A in MLL-AF9 LSCs) - on topic, cited, and not readable: no PMCID confirmed against efetch, and no full text opened, so this beat does not assert what H3K4 measurements it contains.
| script | no script committed; PubMed eutils esearch plus a PMC HTML/fullTextXML fetcher run in the beat scratchpad |
|---|---|
| status | run |
| kmt2a term set | (MLL-rearranged OR KMT2A-rearranged OR MLL-AF9 OR MLL-AF4 OR KMT2A OR "MLL fusion")[tiab] |
| kmt2a term set count | 3209 |
| full texts swept | 19 |
| rule 80 hit rate | 3 of 7 marks; 4 of 7 confirmed the abstract-level picture |
| retrieval note | the PMC HTML route succeeded for all 19 and was faster than fullTextXML, which returned HTTP 503 on quoted-phrase searches; route 5 (DOI to publisher) returned figure titles but not the body for the one paywalled paper |
| result summary | Rule 80 changed the answer for 3 of 7 marks and confirmed the absence for 4. The three rescued are the marks whose KMT2A measurements are supporting ChIP panels for claims about other molecules; the four confirmed are those where the mark is the paper's subject or where the literature is one paper. |
inputs
[ "PubMed E-utilities esearch, 2026-09-20", "19 full texts from PMC / Europe PMC, 2026-09-20", "data/graph/nodes/histone-marks/ana-histone-mark-audit-rejected-candidates.json" ]
per mark — the numbers behind the chart
{
"H3K27ac": {
"alone": 3116,
"intersection": 14,
"rule_80_changed": true,
"missed_pmid": "31243293"
},
"H3K9me3": {
"alone": 2793,
"intersection": 10,
"rule_80_changed": true,
"missed_pmid": "31243293"
},
"H4K16ac": {
"alone": 364,
"intersection": 3,
"rule_80_changed": true,
"missed_pmid": "25822366"
},
"H3.3": {
"alone": 1823,
"intersection": 1,
"rule_80_changed": true,
"found_in_full_text_as": "an H3.3 K27M reagent in K-562 cells, PMID 36635503"
},
"H3K4me2": {
"alone": 926,
"intersection": 9,
"rule_80_changed": false
},
"H3K9ac": {
"alone": 1380,
"intersection": 4,
"rule_80_changed": false
},
"H2A.Z": {
"alone": 1233,
"intersection": 1,
"rule_80_changed": false
}
}substring versus word boundary
{
"token": "HIRA",
"substring_hits": 9,
"word_boundary_hits": 0,
"all_substring_hits_were": "author surnames"
}- A method that recovers measurements an index cannot see makes every existing bounded-absence claim in this graph provisional, including the ones written by this beat - the correct response is to record which part of the document was searched, not to treat any absence as settled.
- Treating a full-text sweep as a rescue device would systematically over-read a thin literature: four of seven marks here gained nothing, and a beat that reported only the three rescues would have described a field that does not exist.
- Would the same sweep over supplementary files rather than article bodies rescue a fourth mark? Supplementary antibody tables and figure source data were not searched here and are where a mark named once in a methods appendix would live.
- Does the rule-80 hit rate hold outside the histone-mark layer? The prediction is that it is highest for measurements that support claims about other molecules - occupancy, accessibility, phosphorylation - and near zero for entities, which have names that reach abstracts.
- How much of this literature is unreachable because the query is scoped to MLL or KMT2A rather than to the partner? The term set used here omits AFF1-only and MLLT3-only framings, and this project's septin case shows a partner-keyed enumeration can carry nine-tenths of a literature.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
Bounded PubMed [tiab] intersections against a fixed KMT2A term set of 3,209 records return 14 for H3K27ac, 10 for H3K9me3, 9 for H3K4me2, 4 for H3K9ac, 3 for H4K16ac, 1 for a five-spelling H3.3 name list and 1 for an eight-spelling H2A.Z name list, with single-operand counts of 3,116, 2,793, 926, 1,380, 364, 1,823 and 1,233 respectively.
NCBI E-utilities esearch over PubMed, run 2026-09-20 by the Q59 marks2 beat, with every querytranslation read back and recorded · US National Library of Medicine 2026What this evidence cannot show
All counts are titles and abstracts only and were taken on 2026-09-20; they will drift. Every querytranslation was read back and none contained a [UID] term, which is this project's signature of a parser rewrite producing a false zero. The KMT2A term set is the one used by the earlier ledger ana-histone-mark-audit-rejected-candidates, chosen for comparability rather than for completeness - it omits partner-gene-only framings such as AFF1 or MLLT3 without MLL or KMT2A, which the septin case in this project's brief shows can carry most of a literature.
Evidence that argues against this (1)
- The prior ledger reported 11 for H3K27ac, 9 for H3K9me3, 6 for H3K36me2, 2 for H3K9ac and 1 for H4K16ac using the same KMT2A term set in 2026-09-19. The counts here are higher for four of them because the mark-side name lists were widened, which demonstrates that even the bounded abstract count is a function of the name list and not a property of the literature. US National Library of Medicine 2026
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
A word-boundary sweep of nineteen full texts of the core KMT2A-rearranged chromatin literature recovers, for three of seven marks, the best available KMT2A measurement from a paper absent from that mark's own bounded abstract query: H3K27ac and H3K9me3 from PMID 31243293 and H4K16ac from PMID 25822366; the same sweep finds zero word-boundary occurrences of H2A.Z, H2AZ1, H2AZ2, H2AFZ or H2AFV in any of the nineteen, and finds H3.3 in two of them, in neither case as a measurement.
Nineteen-document full-text sweep of the core KMT2A-rearranged chromatin literature for seven histone marks and variants, retrieved through the PMC HTML and Europe PMC fullTextXML routes, stripped tags-first and entities-second, and counted with word boundaries · PubMed Central / Europe PMC 2026What this evidence cannot show
Nineteen documents is a purposive sample biased towards papers this graph already cites and towards open-access content reachable through PMC; it is not a census and supplementary files were not searched. Retrieval used the PMC HTML route with Europe PMC fullTextXML as fallback, text stripped tags-first and entities-second per this project's rule, and lengths checked to exclude body-less documents. The three rescues are all from two papers, so the 'three of seven' figure is carried by a small number of information-dense documents rather than by a broad effect.
Evidence that argues against this (1)
- For four of the seven marks the sweep changed nothing, which bounds how much rule 80 can be expected to deliver: a full-text sweep is not a general rescue device, and treating it as one would license reading a literature as richer than it is. The marks it rescued were specifically those whose measurements are supporting panels for claims about other molecules. Godfrey L; Crump NT; Thorne R; Lau IJ; Repapi E; Dimou D; et al. 2019
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
A naive substring count for HIRA across the same nineteen full texts returns nine hits while a word-boundary count returns zero, and every substring hit is an author surname (Hirashima, Hirai).
Nineteen-document full-text sweep of the core KMT2A-rearranged chromatin literature for seven histone marks and variants, retrieved through the PMC HTML and Europe PMC fullTextXML routes, stripped tags-first and entities-second, and counted with word boundaries · PubMed Central / Europe PMC 2026What this evidence cannot show
Demonstrated on one token across one document set. The inverse error - a word-boundary search missing a hyphenated or possessive form - was not separately quantified here, and both directions are real; the point is that the two counts differ by nine to zero on a token this project treats as a gene of interest.
Evidence that argues against this (1)
- This project's registry already logs the same failure shape for MOZ, a three-letter token that occurs inside ordinary words, so the trap is known - what is new here is that it fired on a five-letter gene symbol that looks safe, via author surnames rather than via ordinary English. PubMed Central / Europe PMC 2026
supplies the counted bound and the rule-80 rescue for this mark
supplies the counted bound and the rule-80 rescue for this mark
supplies the counted bound and the rule-80 rescue for this mark
supplies the counted bound; the abstract index was adequate here
supplies the counted bound; the full-text sweep confirmed the absence
supplies the nineteen-document bound behind this variant's absence claim
supplies the nineteen-document bound behind this variant's absence claim
reproduces the prior ledger's arithmetic and overturns the conclusion for three of its six rejected marks