Analysismoderate

Four counted absences re-run against a second index, and all four hold

also called ledger item M · counted absence recheck · index dependence

When this graph says nobody has studied something, that claim was usually checked against one literature database, and four such claims in this project have previously turned out to be wrong because the search was narrower than the sentence it was used to support. I re-ran four of them against a different database, and this time all four held up - the one apparent exception was a brain-tumour review that happened to mention both search terms for unrelated reasons.

The same thing, for a scientist

Four counted absences for HIRA, SUPT4H1/SPT4, MED23 and H3.3/nucleosome turnover - all originally recorded as PubMed title/abstract zeros - were re-run as EuropePMC free-text queries with positive controls (HIRA 12,449; MED23 937; SUPT4H1 243; KMT2A 10,262). Free-text hit counts of 59-316 resolve to **zero records naming both the gene and KMT2A in title or abstract** on the examined page, which is the expected signature of a true absence rather than a contradiction, because EuropePMC free-text reaches full text. The single apparent hit is a false positive.

Go deeper

Result

entry	free-text hits	naming BOTH in title/abstract	verdict
`gene-hira` E5	159	0	holds
`gene-supt4h1` E3	59	0	holds
`gene-med23` E3	69	0	holds
`hyp-kmt2a-aff1-alters-h3-3-deposition...` E2	316	1	holds (hit is spurious)

The free-text counts are not contradictions. EuropePMC free-text reaches the full text of open-access records, while the original claims were PubMed title/abstract queries. A large full-text count with no title/abstract match is exactly what a real absence looks like: the gene is mentioned in passing in the body of papers about something else.

The one hit is a false positive, checked by reading it: PMID 42131930 is a glioma review in which H3F3A appears as a glioma mutation and MLL appears only inside the phrase 'MLL/SET1 (WRAD) methyltransferase complex' describing WDR5. It is not about H3.3 deposition in KMT2A-rearranged leukaemia and does not bear on the claim.

This is the first check this cycle that confirmed the graph rather than finding a defect, and it is worth stating plainly: the entries audited here recorded a date, the exact query string and the tool, which is the practice that made them re-checkable at all. The absences that previously failed in this project were the ones where the recorded query was narrower than the sentence it licensed.

How this could bite us back

These absences are absences of title/abstract-level reporting, not proof that the experiments were never done. Anyone planning an experiment on the strength of 'nobody has looked at HIRA here' should search full text and conference abstracts before committing reagents.

Still open

Would a genuinely independent corpus - Scopus, Web of Science, or a full-text preprint index - agree? EuropePMC incorporates MEDLINE, so this is a weaker replication than it looks.
Should counted absences carry an expiry date, given that all four were recorded within the last fortnight and the literature moves?

Every source behind this entry (2)

E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.analysis
Counted absences for HIRA, SUPT4H1/SPT4, MED23 and H3.3/nucleosome turnover in KMT2A-rearranged leukaemia, originally recorded as PubMed zeros, return no record naming both the gene and KMT2A in title or abstract when re-run as EuropePMC free-text queries, on the first page of results ranked by relevance.
counted-absence-index-recheck: four recorded absences re-run against a second index with positive controls · Oncofusion knowledge graph (internal analysis) 2026
What this evidence cannot show
Run 2026-09-20. **Only the first page of each result set was examined**, so this is a lower bound on refutation and does not re-establish the absences - a single decisive paper could rank below the examined page, and relevance ranking is not importance ranking. **The two indices are not independent**: EuropePMC incorporates MEDLINE, so agreement between them is weaker evidence than agreement between genuinely separate corpora would be. Moderate for both reasons.
Evidence that argues against this (1)
- A search that finds nothing in title or abstract cannot exclude a result reported only in a figure, a supplementary table or a methods section - which is where negative or incidental findings about a single subunit most often live. These absences are absences of *reporting at the title/abstract level*, and the entries should not be read as claiming the experiments were never done. Jung E et al. 2026
E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The single record returned that named both an H3.3 term and an MLL term in its title or abstract is a glioma review in which H3F3A appears as a glioma mutation and MLL appears only within the phrase describing the MLL/SET1 WRAD methyltransferase complex as the context for WDR5.
Decoding WDR5-Mediated Interactions in Gliomas: Implications for Targeted Therapy · Jung E, Piperi C · Medicinal Research Reviews 2026
PMID 42131930
What this evidence cannot show
Abstract fetched and read back on 2026-09-20. Established as a description of what the abstract contains. Carried because an automated absence checker would have scored this as a contradiction and flipped a counted absence to refuted - **the distinction between co-occurrence and aboutness is exactly what a checker cannot make**, and this graph has now logged that distinction three times.
Evidence that argues against this (1)
- Dismissing a hit as spurious is a judgement, and a reviewer who disagreed could reasonably ask why a WDR5-centred review is irrelevant to a graph in which WRAD is a central complex. The answer is that the claim at issue is specifically about H3.3 deposition in KMT2A-rearranged cells, not about WDR5 - but the reader should be able to check that reasoning, which is why the record is named here. Oncofusion knowledge graph (internal analysis) 2026

Connected to

supports →HIRA - the chaperone that puts new H3.3 into transcribed gene bodies, and therefore the machine that runs this graph's clearance clock
COUNTED ABSENCE RE-RUN AND HOLDS: 159 EuropePMC free-text hits, none naming HIRA and KMT2A in title or abstract (2026-09-20).
supports →SUPT4H1 (SPT4) — zero papers linking it to KMT2A, and the 138th most KMT2A-selective gene out of 17,728
COUNTED ABSENCE RE-RUN AND HOLDS: 59 free-text hits, none in title or abstract with KMT2A.
supports →MED23 — rank 16 of 17,728, survives the lineage split with the B-lineage half stronger, and is not absent from the leukaemia literature after all
COUNTED ABSENCE RE-RUN AND HOLDS: 69 free-text hits, none in title or abstract with KMT2A.
supports →Does KMT2A::AFF1 change histone turnover at its target genes, or does it read a chromatin state it did not make?
COUNTED ABSENCE RE-RUN AND HOLDS: the one apparent hit is a glioma WDR5 review where H3F3A and MLL co-occur for unrelated reasons.
supports →Absence claims: 93% are unbounded in prose, but the evidence claims are 71% bounded - and the query that tested them was itself broken
THE PAYOFF OF BOUNDING: these four were re-checkable precisely because each recorded a date, the exact query and the tool.
related to →Counted absences keyed on a complex subunit miss the papers that name only the complex - demonstrated on our own SSRP1 entry, which declared no KMT2A paper about FACT while this graph cites one about the index fusion in four other entries
THE COMPLEMENTARY FAILURE: that analysis showed subunit-keyed absence ratios are unusable because complex names return ordinary English; this one shows the subunit-keyed zeros themselves survive a second index.