What the Polycomb beat searched, what it counted as absent, what it read and rejected, and the two absences it got wrong before hand-back
also called PRC2 beat enumeration · Q41 bounded absences and rejections · Polycomb arm search record
A record of every search this beat ran, every gap it found, and - importantly - the papers it read and decided not to use. Someone re-running this work next year should be able to repeat the searches and the judgements, not just the searches.
The same thing, for a scientist
This project is good at **bounded absence** (*'this query returns 0 records, here is the query'*) and blind to **bounded rejection** (*the papers read and judged off-beat leave no trace*). This entry records both for the Q41 Polycomb beat, plus a third category the brief names: **found, on topic, and unreadable**. It also records **two absences this beat wrote and then had to correct before hand-back**, because the correction is the most transferable thing in it: in both cases the query was well-formed, the count was accurate, and the conclusion was false. All queries run 2026-09-20 against PubMed `esearch`; counts marked *checked* had every returned title read.
Counted absences that survive
| query | records | checked on-topic | verdict |
|---|---|---|---|
tazemetostat AND (KMT2A OR MLL) | 1 | 0 | No primary study of tazemetostat in KMT2A-rearranged leukaemia. The one record is a pharmacology review (PMID 35753676). Bounded by: the development codes EPZ-6438 and E7438 were not searched, and EPZ-6438 IS used in a leukaemia experiment (PMID 36280659) - so the absence is narrower than it looks. |
(EED226 OR MAK683 OR "EED inhibitor") AND (leukemia OR MLL OR KMT2A) | 0 | 0 | No EED-directed PRC2 inhibitor has been tested in any leukaemia, while 28 records cover the class elsewhere and MAK683 completed a 139-patient phase 1/2 with no leukaemia cohort. Strongest absence in the beat. Bounded by: EEDi-5285, EEDi-5273 and the bivalent degrader were not searched by name. |
(RBBP4 OR RBBP7) AND (KMT2A OR MLL) AND leukemia | 2 | 0 | Neither RBBP paralogue has been studied in this disease. Bounded by: RBAP46, RBAP48, RbAp46, RbAp48 and NURF55 not searched. |
SUZ12 AND (KMT2A OR MLL) AND leukemia | 5 | 1 | Only PMID 22469984. Bounded by: JJAZ1, CHET9, KIAA0160 not searched, and JJAZ1 is the live symbol in the JAZF1-SUZ12 sarcoma literature. |
valemetostat AND (KMT2A OR MLL) | 1 | 1 | The single record is the B-ALL/MLL-AF4 PDX paper. Bounded by: DS-3201 and DS-3201b not searched separately. |
| PRC2.1 versus PRC2.2 in a KMT2A-REARRANGED leukaemia | - | 0 | No study distinguishes the subtypes in this disease. See the corrections below - the first version of this absence was wrong. |
| H3K27me3 measured after menin inhibition, in any system | - | 0 | No menin-inhibitor study read in this beat reports an H3K27me3 measurement, despite the menin-EZH2 link being in print since 2013. This is the gap exp-menin-inhibitor-h3k27me3-timecourse-index-fusion exists to fill. |
Rule 64 was applied throughout. The complex form is consistently larger than the subunit form: "Polycomb repressive complex 2" AND (KMT2A OR MLL) AND leukemia returns 47 against "PRC2" AND (KMT2A OR MLL-rearranged) AND leukemia's 24, and the 47 was title-checked. It recovered the MLL-AF9 genetics, the response-element literature and the trithorax/Polycomb structural work, and no additional study for any absence above.
The two absences this beat got wrong, and how
Both were caught by the subtract-before-you-report rule - grep the repository - and both are logged because the correction generalises.
1. The PRC2.1/PRC2.2 absence. First written as: (JARID2 OR MTF2 OR PHF19 OR AEBP2 OR EPOP) AND (KMT2A OR MLL) AND leukemia returns 1, on-topic 0, therefore nothing distinguishes the subtypes near this axis. The query is accurate and the conclusion was false. Dropping the word leukemia - (... OR "PRC2.1" OR "PRC2.2") AND (KMT2A OR MLL OR menin) - returns 4 and the first is PMID 36635503, whose whole-genome CRISPR screen identifies MTF2-PRC2.1, PCGF1-PRC1.1 and Menin-KMT2A/B as bivalency maintainers. The paper's abstract says cancer cells and pluripotent stem cells, never leukaemia. And this graph already cited it in two entries (gene-pcgf1, protein-prc1-1-complex). Same family as the FACT and Mediator failures: conjoining a disease word to a subunit symbol excludes the paper that frames the finding at a different level.
2. A citation attached to the wrong record, twice. Writing at speed, this beat attached the MLL-AF4 gene-body-spreading and H3K36me3-disruption claim to PMID 19855078, which is the infant promoter-methylation paper; the correct record is PMID 28076791. Separately, two citation blocks were written with author lists and a DOI from recall - PMID 21741597 (Bernt KM, Zhu N, Sinha AU, Vempati S, Faber J, Krivtsov AV, not the names first written) and PMID 29992232 (Du J, Kirk B, Zeng J, Ma J, Wang Q, DOI 10.1093/nar/gky595, not gky438). All three were caught by re-fetching before hand-back and all three were fixed. The generalisation is the brief's own: an identifier already in the graph has the same status as an identifier in memory - unverified until this session fetched it.
3. A mark-name count that misses its own decisive papers. H3K36me2 AND (KMT2A OR MLL) AND leukemia returns 7 and H3K36me3 returns 16 - and both papers this beat actually used are invisible to those queries, because PMID 17589499 writes H3K36 methylation and PMID 27154821 writes Histone H3 Lysine 36 Dimethylation spelled out. A counted absence on a histone mark must enumerate the spelled-out forms as well as the me2/me3 notation. This beat did not apply that lesson to its own H3K4me3 (152) and H3K27me3 (35) counts, which are therefore also under-estimates and are recorded as size indications only, not as literatures.
Bounded rejections - read, on topic in the search, not used
Three lines each: identifier, reason, what would change it.
- PMID 29754954, The Polycomb-Dependent Epigenome Controls beta Cell Dysfunction, Dedifferentiation, and Diabetes (Cell Metab 2018). Returned by the PRC2/KMT2A query because menin biology spans both. Pancreatic beta cells, not haematopoietic. Would change if the graph's endocrine arm needs a PRC2 handle for menin's normal-tissue role - it is the best available.
- PMID 35584667, H3-K27M-mutant nucleosomes interact with MLL1 to shape the glioma epigenetic landscape (Cell Rep 2022). Genuinely about PRC2, H3K27 and MLL1 together, but the system is diffuse midline glioma with an oncohistone. Would change if anyone reports an H3K27M-like dominant-negative effect in leukaemia; the mechanism is directly transferable and the disease is not.
- PMID 33349639, Mll1 is required for Wnt-driven intestinal tumorigenesis and cancer stemness (Nat Commun 2020). Wild-type MLL1, intestinal epithelium. Would change if the project opens a non-haematopoietic KMT2A arm.
- PMID 28180285, HOTAIRM1... temporal collinear HOXA gene activation (NAR 2017) and PMID 26496121, lncRNA SRA with TrxG and PRC2 (PLoS Genet 2015). Both connect PRC2 to HOXA through lncRNAs, which is on-beat in principle. Not used because this graph's RNA-layer arm (
ana-rna-layer-kmt2a-selectivity,mechanism-rna-layer-and-menin-inhibition) owns that question and a Polycomb-beat entry would duplicate it badly. Would change if the RNA-layer arm asks for the PRC2 half explicitly. - PMID 27447986, An Evolutionary Conserved Epigenetic Mark of Polycomb Response Elements Implemented by Trx/MLL/COMPASS (Mol Cell 2016). On-beat and probably important for the response-element question, but Drosophila-anchored and this beat used PMID 29992232 for the mammalian statement instead. Would change if someone writes a dedicated Polycomb-response-element entry; this should be its first citation.
- PMID 41438598, Targeting dysregulated epigenetic and transcription factor networks in KMT2A-rearranged AML using iPSC models (Blood Neoplasia 2026). Squarely on-beat by title and returned by three separate queries. Not read in this beat for time, and it is the single most likely source of new KMT2A-specific PRC2 data. This is a recommended first read for the next Polycomb cycle.
- PMID 38295690, Discovery of a novel, highly potent EZH2 PROTAC degrader for targeting non-canonical oncogenic functions of EZH2 (Eur J Med Chem 2024). A degrader is the right modality for the catalysis-versus-scaffolding question this beat raises for EZH1. Not used because no leukaemia data and the beat had no capacity to evaluate the chemistry. Would change if
exp-eed-allosteric-inhibition-in-kmt2a-aff1is scoped to include a degrader arm. - PMID 31831267, Degradation of Polycomb Repressive Complex 2 with an EED-Targeted Bivalent Chemical Degrader (Cell Chem Biol 2020). Same reasoning; it is named in
gene-eedas existing chemical matter but was not read.
Marks considered and declined. mark-h3k27ac - the reciprocal state on the same residue as H3K27me3, which rises whenever PRC2 is inhibited and is the mechanistic link to the translocation hazard. Declined because this beat read no primary H3K27ac dataset in this disease, and writing it from three PRC2 papers' asides would be building an entry out of other entries' margins. It is the next mark the graph needs. mark-h2ak119ub - PRC1's product and the mark PRC2.2 reads - declined as belonging to the PRC1 arm, which already has protein-prc1-1-complex and gene-pcgf1.
Found, on topic, and unreadable
None. This beat did not require full text for any claim: every conclusion rests on abstracts fetched from NCBI eutils, on database records fetched from UniProt and RCSB, or on the two internal analyses. No paper was put through the five-route test, because none needed to be.
That is itself worth recording as a limitation rather than an achievement. Every biological claim in this beat's entries is abstract-level, and each carries the caveat 'abstract read from the eutils record; full text not read'. Two of them would materially benefit from full text:
- PMID 40722046 (HDAC7/EZH2, the index-disease anchor) - whether EZH2 ChIP or H3K27me3 was measured at HDAC7 is the single fact that decides whether
hyp-prc2-is-the-fusions-repressive-armis a hypothesis or a finding, and the abstract does not say. It is open access (PMC12305908), so route 2 should succeed. - PMID 36280659 (the DS-3201 versus EPZ-6438 head-to-head) - whether H3K27me3 depletion was matched between arms decides whether
exp-eed-allosteric-inhibition-in-kmt2a-aff1is needed as designed. No PMCID was returned in the record fetched here, so this is a route-5 candidate via the DOI.
Both are named as prerequisites in the relevant entries.
| script | none - this is a search and judgement record, not a computation |
|---|---|
| status | run |
| date | 2026-09-20 |
| result summary | Six surviving counted absences, two corrected absences, three corrected citations, ten bounded rejections, two declined marks, zero unreadable papers, two full-text prerequisites. The strongest absence is that no EED-directed PRC2 inhibitor has ever been tested in any leukaemia despite a 139-patient phase 1/2 existing. |
counted absences — the numbers behind the chart
[
{
"query": "tazemetostat AND (KMT2A OR MLL)",
"records": 1,
"on_topic": 0,
"aliases_unsearched": [
"EPZ-6438",
"E7438"
],
"note": "EPZ-6438 IS used in a leukaemia experiment (PMID 36280659); the absence is narrower than the count suggests"
},
{
"query": "(EED226 OR MAK683 OR \"EED inhibitor\") AND (leukemia OR MLL OR KMT2A)",
"records": 0,
"on_topic": 0,
"aliases_unsearched": [
"EEDi-5285",
"EEDi-5273",
"bivalent EED degrader"
],
"note": "strongest absence in the beat; 28 records for the class elsewhere, 139-patient phase 1/2 with no leukaemia cohort"
},
{
"query": "(RBBP4 OR RBBP7) AND (KMT2A OR MLL) AND leukemia",
"records": 2,
"on_topic": 0,
"aliases_unsearched": [
"RBAP46",
"RBAP48",
"RbAp46",
"RbAp48",
"NURF55"
]
},
{
"query": "SUZ12 AND (KMT2A OR MLL) AND leukemia",
"records": 5,
"on_topic": 1,
"aliases_unsearched": [
"JJAZ1",
"CHET9",
"KIAA0160"
]
},
{
"query": "valemetostat AND (KMT2A OR MLL)",
"records": 1,
"on_topic": 1,
"aliases_unsearched": [
"DS-3201",
"DS-3201b"
]
},
{
"query": "H3K27me3 measured after menin inhibition, any system",
"records": 0,
"on_topic": 0,
"note": "no menin-inhibitor study read in this beat reports it; drives exp-menin-inhibitor-h3k27me3-timecourse-index-fusion"
}
]corrected absences
[
{
"absence": "nothing distinguishes PRC2.1 from PRC2.2 near this axis",
"why_wrong": "the query conjoined a disease word to subunit symbols; PMID 36635503 names MTF2-PRC2.1 and Menin-KMT2A/B but never says leukaemia, and was already cited twice in this graph",
"corrected_to": "no study distinguishes PRC2.1 from PRC2.2 in a KMT2A-REARRANGED leukaemia"
},
{
"absence": "H3K36 methylation is barely studied in this disease (7 and 16 records)",
"why_wrong": "both decisive papers spell the mark out and are invisible to the me2/me3 notation queries",
"corrected_to": "the counts are lower bounds; a mark-keyed count must enumerate spelled-out forms"
}
]corrected citations
[
{
"error": "MLL-AF4 gene-body spreading / H3K36me3 disruption attached to PMID 19855078",
"correct": "PMID 28076791; 19855078 is the infant promoter-methylation paper",
"caught_by": "re-fetch before hand-back"
},
{
"error": "PMID 21741597 author list written from recall",
"correct": "Bernt KM, Zhu N, Sinha AU, Vempati S, Faber J, Krivtsov AV, et al.",
"caught_by": "re-fetch before hand-back"
},
{
"error": "PMID 29992232 author list and DOI written from recall (gky438)",
"correct": "Du J, Kirk B, Zeng J, Ma J, Wang Q, et al.; DOI 10.1093/nar/gky595",
"caught_by": "re-fetch before hand-back"
}
]bounded rejections — the numbers behind the chart
[
{
"pmid": "29754954",
"reason": "pancreatic beta cells, not haematopoietic",
"would_change_if": "the endocrine arm needs a PRC2 handle for menin's normal-tissue role",
"title": "The Polycomb-Dependent Epigenome Controls β Cell Dysfunction, Dedifferentiation, and Diabetes.",
"authors": "Lu TT, Heyne S, Dror E, et al.",
"venue": "Cell Metab",
"year": 2018,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
},
{
"pmid": "35584667",
"reason": "H3-K27M glioma, not leukaemia, though mechanistically about PRC2 and MLL1 together",
"would_change_if": "an oncohistone-like dominant-negative effect is reported in leukaemia",
"title": "H3-K27M-mutant nucleosomes interact with MLL1 to shape the glioma epigenetic landscape.",
"authors": "Furth N, Algranati D, Dassa B, et al.",
"venue": "Cell Rep",
"year": 2022,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
},
{
"pmid": "33349639",
"reason": "wild-type Mll1 in intestinal epithelium",
"would_change_if": "a non-haematopoietic KMT2A arm opens",
"title": "The epigenetic regulator Mll1 is required for Wnt-driven intestinal tumorigenesis and cancer stemness.",
"authors": "Grinat J, Heuberger J, Vidal RO, et al.",
"venue": "Nat Commun",
"year": 2020,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
},
{
"pmid": "28180285",
"reason": "lncRNA route to HOXA; owned by this graph's RNA-layer arm",
"would_change_if": "the RNA-layer arm asks for the PRC2 half",
"title": "Reciprocal regulation of chromatin state and architecture by HOTAIRM1 contributes to temporal collinear HOXA gene activation.",
"authors": "Wang XQ, Dostie J",
"venue": "Nucleic Acids Res",
"year": 2017,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
},
{
"pmid": "26496121",
"reason": "same as above",
"would_change_if": "same as above",
"title": "Association of the Long Non-coding RNA Steroid Receptor RNA Activator (SRA) with TrxG and PRC2 Complexes.",
"authors": "Wongtrakoongate P, Riddick G, Fucharoen S, et al.",
"venue": "PLoS Genet",
"year": 2015,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
},
{
"pmid": "27447986",
"reason": "Drosophila-anchored Polycomb response elements; PMID 29992232 used for the mammalian statement instead",
"would_change_if": "a dedicated Polycomb-response-element entry is written - this should be its first citation",
"title": "An Evolutionary Conserved Epigenetic Mark of Polycomb Response Elements Implemented by Trx/MLL/COMPASS.",
"authors": "Rickels R, Hu D, Collings CK, et al.",
"venue": "Mol Cell",
"year": 2016,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
},
{
"pmid": "41438598",
"reason": "NOT READ - squarely on-beat (KMT2A-rearranged AML iPSC models, returned by three queries), dropped for time",
"would_change_if": "nothing; this is a recommended first read for the next Polycomb cycle",
"title": "Targeting dysregulated epigenetic and transcription factor networks in KMT2A-rearranged AML using iPSC models.",
"authors": "Palau A, Thier J, Naughton A, et al.",
"venue": "Blood Neoplasia",
"year": 2026,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
},
{
"pmid": "38295690",
"reason": "EZH2 PROTAC degrader; right modality for the catalysis-versus-scaffolding question but no leukaemia data",
"would_change_if": "exp-eed-allosteric-inhibition-in-kmt2a-aff1 is scoped to include a degrader arm",
"title": "Discovery of a novel, highly potent EZH2 PROTAC degrader for targeting non-canonical oncogenic functions of EZH2.",
"authors": "Velez J, Dale B, Park KS, et al.",
"venue": "Eur J Med Chem",
"year": 2024,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
},
{
"pmid": "31831267",
"reason": "bivalent EED-targeted PRC2 degrader; named in gene-eed as chemical matter but not read",
"would_change_if": "the EED experiment needs a degrader comparator to separate allosteric from structural contributions",
"title": "Degradation of Polycomb Repressive Complex 2 with an EED-Targeted Bivalent Chemical Degrader.",
"authors": "Potjewyd F, Turner AW, Beri J, et al.",
"venue": "Cell Chem Biol",
"year": 2020,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
},
{
"pmid": "29572378",
"reason": "mechanisms of resistance to EZH2 inhibitors in DLBCL; relevant to the drug entries but a different disease and not read in full",
"would_change_if": "a PRC2-directed agent reaches a KMT2A-rearranged trial and a resistance section is needed",
"title": "Mechanisms of resistance to EZH2 inhibitors in diffuse large B-cell lymphomas.",
"authors": "Bisserier M, Wajapeyee N",
"venue": "Blood",
"year": 2018,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
}
]marks declined
[
{
"mark": "H3K27ac",
"reason": "the reciprocal state on the same residue as H3K27me3 and the mechanistic link to the translocation hazard, but this beat read no primary H3K27ac dataset in this disease and would have been building an entry from other papers' asides",
"priority": "highest - this is the next mark the graph needs"
},
{
"mark": "H2AK119ub",
"reason": "PRC1's product and PRC2.2's input; belongs to the PRC1 arm, which already holds protein-prc1-1-complex and gene-pcgf1",
"priority": "medium"
}
]unreadable papers
[]
full text prerequisites
[
{
"pmid": "40722046",
"why": "whether EZH2 ChIP or H3K27me3 was measured at HDAC7 decides whether hyp-prc2-is-the-fusions-repressive-arm is a hypothesis or a finding",
"route": "open access, PMC12305908 - route 2 should succeed",
"title": "HDAC7 induction combined with standard-of-care chemotherapy provides a therapeutic advantage in t(4;11) infant B-cell acute lymphoblastic leukemia.",
"authors": "de Barrios O, Ocón-Gabarró I, Gusi-Vives M, et al.",
"venue": "Biomark Res",
"year": 2025,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
},
{
"pmid": "36280659",
"why": "whether H3K27me3 depletion was matched between the DS-3201 and EPZ-6438 arms decides whether exp-eed-allosteric-inhibition-in-kmt2a-aff1 is needed as designed",
"route": "no PMCID returned; route-5 candidate via DOI 10.1007/s12185-022-03469-8",
"title": "Dual inhibition of EZH1/2 induces cell cycle arrest of B cell acute lymphoblastic leukemia cells through upregulation of CDKN1C and TP53INP1.",
"authors": "Ito J, Yamagata K, Shinohara H, et al.",
"venue": "Int J Hematol",
"year": 2023,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
}
]- Every counted absence in this entry is bounded by an unsearched alias list that is written out beside it. Anyone building a decision on one of them should re-run it over those aliases first - in at least three cases an unsearched alias is already known to change the answer.
- Should this project add a mark-name alias table the way it has an implicit gene-alias discipline? Every histone mark has a notation form (H3K36me2) and one or more spelled-out forms ('Histone H3 Lysine 36 Dimethylation', 'H3K36 methylation'), and this beat found both of its decisive papers invisible to the notation query.
- Can the subtract-before-you-report grep be made a script? It caught the load-bearing error in this beat and it is currently a manual step that an interrupted agent would skip. A `check-absence-vs-bibliography.py` already exists in scripts/ - whether it covers this case was not verified here.
- PMID 41438598 (KMT2A-rearranged AML iPSC models, 2026) was returned by three separate queries in this beat and was not read. It is the most likely source of new KMT2A-specific PRC2 data and should be the next cycle's first action.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
PubMed esearch counts run 2026-09-20 for the Polycomb beat: (EZH2 OR SUZ12 OR EED) AND (KMT2A OR MLL) AND leukemia = 91; 'Polycomb repressive complex 2' AND (KMT2A OR MLL) AND leukemia = 47; 'PRC2' AND (KMT2A OR MLL-rearranged) AND leukemia = 24; H3K27me3 AND (KMT2A OR MLL) AND leukemia = 35; EZH1 AND (KMT2A OR MLL) AND leukemia = 11; EZH2 AND 'MLL-AF9' = 13; EZH2 AND ('MLL-AF4' OR 'KMT2A-AFF1') = 4; SUZ12 AND (KMT2A OR MLL) AND leukemia = 5; (RBBP4 OR RBBP7) AND (KMT2A OR MLL) AND leukemia = 2; (JARID2 OR MTF2 OR PHF19 OR AEBP2 OR EPOP) AND (KMT2A OR MLL) AND leukemia = 1; tazemetostat AND (KMT2A OR MLL) = 1; valemetostat AND (KMT2A OR MLL) = 1; (EED226 OR MAK683 OR 'EED inhibitor') AND (leukemia OR MLL OR KMT2A) = 0.
PubMed esearch counts for the Q41 Polycomb beat, run and recorded 2026-09-20 · NCBI PubMed 2026What this evidence cannot show
Counts marked in the table above as 'checked' had every title read; the 91 and the 35 did not. The 'MLL' token matches free text and will include mixed-lineage-leukaemia mentions unrelated to a rearrangement. Every count is bounded by the symbol set listed with it, and in at least three cases (tazemetostat/EPZ-6438, valemetostat/DS-3201, H3K36me2/'Histone H3 Lysine 36 Dimethylation') an unsearched alias is known to change the answer.
Evidence that argues against this (1)
- A count keyed to one spelling can under-count by an order of magnitude - this project measured SEPT9 at 20 records against SEPTIN9 at 1, and SETD1B at 17 against Set1B's 27. Counts keyed to a disease word additionally exclude papers framed at a different level, which is how the PRC2.1/PRC2.2 absence in this beat was initially wrong. Sparbier CE et al. 2023
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The subunit-plus-disease query used for the PRC2 subtype absence excluded PMID 36635503, which names MTF2-PRC2.1 and Menin-KMT2A/B but whose abstract says 'cancer cells and pluripotent stem cells' and never 'leukaemia'; the paper was already cited in two entries of this graph (gene-pcgf1, protein-prc1-1-complex) before this beat began.
Targeting Menin disrupts the KMT2A/B and polycomb balance to paradoxically activate bivalent genes. · Sparbier CE, Gillespie A, Gomez J, Kumari N, Motazedian A, Chan KL, et al. · Nat Cell Biol 2023What this evidence cannot show
Abstract read from the eutils record; full text not read. The paper's system is not a KMT2A-rearranged leukaemia, so the corrected absence - no study distinguishes PRC2.1 from PRC2.2 in a KMT2A-REARRANGED leukaemia - still stands; what failed was the broader claim that nothing near this axis distinguishes them.
Evidence that argues against this (1)
- That the corrected absence still stands is not a reason to relax. The same query construction would also miss any future paper that studies PRC2 subtypes in KMT2A-rearranged cells while framing its abstract around, say, bivalency or lineage plasticity rather than the disease name - so the surviving absence is bounded by the same weakness that broke the first one. NCBI PubMed 2026
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Both papers this beat used for H3K36 methylation are invisible to the me2/me3 notation queries: PMID 17589499 is titled 'NUP98-NSD1 links H3K36 methylation to Hox-A gene activation and leukaemogenesis' and PMID 27154821 is titled 'ASH1L Links Histone H3 Lysine 36 Dimethylation to MLL Leukemia' - so a counted absence on a histone mark must enumerate the spelled-out forms as well as the notation.
ASH1L Links Histone H3 Lysine 36 Dimethylation to MLL Leukemia. · Zhu L, Li Q, Wong SH, Huang M, Klein BJ, Shen J, et al. · Cancer Discov 2016What this evidence cannot show
Abstract read from the eutils record; full text not read. This citation is used to demonstrate a retrieval failure mode rather than for its biology, which is carried in `mark-h3k36me2-3`. The lesson was NOT retrospectively applied to this beat's H3K4me3 (152) and H3K27me3 (35) counts, which are therefore also under-estimates.
Evidence that argues against this (1)
- The same failure affects PMID 17589499 in the opposite direction - it says 'H3K36 methylation' without a state, so a query for the spelled-out dimethyl form would miss it too. There is no single spelling that captures this literature, and any mark-keyed count should be treated as a lower bound rather than a measurement. Wang GG et al. 2007
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Of the nineteen primary PMIDs this beat used, only two (22396593 and 22469984) were already cited anywhere in this graph, both in mechanism-polycomb-prc1-prc2-cbx8; PMID 36635503 was cited in gene-pcgf1 and protein-prc1-1-complex, PMID 27154821 in eleven LEDGF-arm entries, and PMID 17589499 in fusion-nup98-nsd1 and protein-nup98-domains - so the repository grep prescribed by the subtract-before-you-report rule found three pre-existing citations that bore directly on this beat's conclusions.
Repository citation audit for the Q41 Polycomb beat - grep of data/graph/nodes for each PMID used, run 2026-09-20 · Q41-prc2-arm · Oncofusion knowledge graph, internal analysis 2026What this evidence cannot show
A grep over stored PMID strings finds a paper only if it is cited by identifier; a paper described in prose without its PMID would be missed. The audit was run over `data/graph/nodes` only, not over `docs/`, `analyses/` or the memos, so a citation living in a memo would not have been found.
Evidence that argues against this (1)
- The decisive catch in this beat came from exactly the case a PMID grep handles well - PMID 36635503 stored as an identifier in two entries. The case it would NOT handle is the one that produced the FACT failure, where a paper is relevant under a name no subunit query and no stored identifier would surface. A grep is a necessary check and not a sufficient one. Sparbier CE et al. 2023
the counted absence and its unsearched aliases
zero EED-inhibitor leukaemia studies against 28 for the class elsewhere
the PRC2.1/PRC2.2 absence, in its corrected form
the mark-name retrieval failure that made both decisive papers invisible to the notation query