Oncofusion

start / graph / ana-q159-eleven-abstract-level-absences-retested-against-agent-and-model-names-and-none-was-overturned

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q159-retest-the-abstract-level-bounded-absences

Eleven abstract-level bounded absences re-tested with the instrument that overturned the HDAC6 one - ten survive, one is not checkable this way, none overturned

also called Q159 re-test table · the agent-and-model re-run

A previous cycle found one of this project's no-one-has-looked claims to be wrong, by searching for drug names and cell-line names instead of disease names. Doing the same thing to eleven more of those claims did not break any of them: ten held up under the better search and one asks a question no literature search can answer.

The same thing, for a scientist

In cycle 143 `ana-testing-the-refusals-instead-of-the-gaps-overturned-the-first-one-tried` overturned `ana-no-hdac-inhibitor-result-in-this-disease-attributes-to-an-enzyme` by replacing disease terms with **drug names and cell-line names**: a medicinal-chemistry paper reports `MV4-11 IC50 = 28 nM` in a table and has no reason to write *KMT2A* anywhere. This beat applied that instrument systematically to eleven more abstract-level absences. **16 logged queries, 388 records returned, 207 titles hand-read, every `querytranslation` inspected. Zero overturns.** Ten absences survive intact; one (`absence-no-menin-dependence-measurement-in-kmt2a-r-sarcoma`) is **NOT-CHECKABLE by a literature query** because its object is the non-existence of a model, for which Cellosaurus - which that entry already queried - is the matching instrument. The single near-miss is recorded separately (`ana-q159-the-ms67-near-miss-a-chemical-degron-of-a-wrad-subunit-already-exists-in-the-index-fusion-line`) and it also survives, because the corpus had already considered and rejected the paper in question. **The survivors are the result.** One overturn in twelve tested across two cycles is 8.3%, and the rule-of-three upper bound on the true overturn rate given 0 of 11 here is 27%.

Go deeper
The verdict table
absence entryagent/model query runnon-beat after hand-reading titlesverdict
absence-no-proximity-degron-or-crispr-study-of-wrad-in-kmt2a-fusion-leukaemiaWDR5/RBBP5/ASH2L/DPY30 x MV4-11/MOLM-13; WDR5 degrader NAMES (MS67, MS33, OICR-9429, MM-589, MM-102, DDO-2213, piribedil); BioID/TurboID/APEX2 x leukaemia lines7 / 26 / 33 / 61 near-missSURVIVES
absence-ledgf-agent-never-tested-in-menin-inhibitor-resistant-diseaseLEDGIN/ALLINI/BI-224436/pirmitegravir/STP0404/GS-9822/CX14442; and LEDGF x ligand x leukaemia115 / 270SURVIVES
absence-h3k79me2-decay-half-life-never-measuredEPZ-5676/pinometostat/EPZ004777/SGC0946 x washout/recovery/decay/turnover; DOT1L x degron4 / 50SURVIVES
absence-mllt10-kinase-site-relationsMLLT10/AF10 x phospho vocabulary230SURVIVES
absence-enl-yeats-and-condensate-phospho-gatingENL YEATS agent NAMES: SR-0813, TDI-11055, GNE-064, SGC-iMLLT100 on phosphoSURVIVES
absence-no-idh-inhibitor-has-been-reported-in-a-kmt2a-rearranged-cohortivosidenib/enasidenib/olutasidenib/vorasidenib/AG-120/AG-221/FT-2102/AGI-6780 x KMT2A/MLL/MV4-11/MOLM-13/11q2380SURVIVES
absence-nothing-has-ever-been-measured-in-parathyroid-under-a-menin-inhibitor19 menin-inhibitor names x parathyroid/PTH/insulinoma/neuroendocrine20SURVIVES
absence-no-menin-inhibitor-occupancy-dataset-in-an-infant-derived-kmt2a-aff1-line19 menin-inhibitor names x SEM/KOPN-8/RS4;11/MV4-11/ML-2/NOMO-1/THP-1/MOLM-13 x ChIP/CUT&RUN/occupancy10SURVIVES
absence-no-junction-specific-antibody-exists-for-any-kmt2a-fusionreagent vocabulary: antibody/monoclonal/nanobody/scFv/hybridoma/immunogen x fusion/junction/breakpoint/neoepitope400SURVIVES
absence-no-experiment-tests-cxxc-reader-competition-at-a-kmt2a-fusion-targetcompetitor PROTEINS (CFP1/CXXC1/KDM2A/KDM2B/TET1/IDAX/CXXC4/CXXC5/MBD1) x competition vocabulary30SURVIVES
absence-no-menin-dependence-measurement-in-kmt2a-r-sarcoma19 menin-inhibitor names x sarcoma/rhabdomyosarcoma/soft tissue/solid tumour50 for KMT2A-r sarcomaNOT-CHECKABLE

n is the raw record count from a querytranslation-checked esearch; the on-beat column is after reading every title.

Why the sarcoma one is NOT-CHECKABLE and not a survival

The menin-inhibitor name list against sarcoma terms returns 5 records and all five are Ewing sarcoma - menin inhibition impairing metastatic colonisation 4, the PPTC VTP-50469 xenograft evaluation, the serine-biosynthesis paper, and the MLL1/menin dependency paper. Ewing sarcoma is not KMT2A-rearranged. The claim is about a dependency measurement in a KMT2A-rearranged sarcoma, and no such cell line or PDX is known to exist, so a literature query returns zero for the same reason a query for unicorn husbandry does: there is nothing to publish about.

A zero from an instrument that could never have returned anything else is not evidence. Logging it as SURVIVES would inflate the survival count with a case that was never at risk, which is the same defect as scoring an abstention as agreement. The entry already queried Cellosaurus, which is the instrument that can see the object, and the Q113 census lists it 'only for contrast' as the one member of its class with the right tool. Recording it here as NOT-CHECKABLE preserves that.

What I discarded and why - 207 titles read, and the token traps that produced most of the noise

Reading every title was not ceremonial. Four tokens supplied nearly all the off-beat records:

  • SEM - in (BioID OR TurboID OR APEX2 OR proximity labeling) AND (... OR SEM ...), 3 of 6 hits were scanning electron microscopy: a FIB-SEM yeast Erg11 paper, a CoLSSEM neuron paper, and an APEX2/FIB-SEM localisation method. The cell line contributed nothing.
  • p75 - in (LEDGF OR PSIP1 OR p75) AND ligand AND leukaemia, 12 of 27 hits are TNF receptor p75, p75NTR or p75LNTR: sympathetic-neuron apoptosis, dorsal-root sensory neurons, a soluble TNFR-Fc MDS trial, motoneuron survival. p75 is the LEDGF community's own shorthand and it is unusable as a query token.
  • AF10 / AF-10 - in the MLLT10 phospho query, AF-10 is also the name of an IL-6-responsive B cell line, and the bare AF prefix pulled atrial-fibrillation cardiology (connexin in right atrial myocardium, a goat model of atrial dysfunction). At least 13 of 23 discarded. The entry being tested already carries alias_precision_note: 'AF10 is word-like and over-retrieves' - a warning that proved exactly right.
  • MLL - in the antibody-reagent query, 40 records included lethal castration-resistant prostate cancer, a HER2/neu rat vaccine, benzene-induced leukaemia and RNP-masked nicks in genomic DNA.

None of these is new biology and all of them would have been counted as hits by a script that trusted the count. The discipline that matters is not a cleverer query, it is reading the titles - which is what the cycle-143 overturn also rested on.

One near-miss worth naming rather than discarding: PMID 21123134, Detection of fusion genes at the protein level in leukemia patients via the flow cytometric immunobead assay 5. It detects a fusion protein immunologically, which is the closest published object to a junction-specific antibody - but it is a sandwich of two antibodies against the two partner proteins, which is why it is not one and why the absence survives. A future reader who finds this paper should not read it as an overturn.

What the instrument is for, stated so the next beat does not misapply it

The rule from cycle 143 is narrow and it earns its keep only on one shape of claim: 'no agent / no reagent / no assay has been tested in model X'. For that shape, the paper that would refute you is a medicinal-chemistry or methods paper whose abstract names a compound code and a cell line and never names the disease gene. Enumerate compound codes, antibody clones, assay names and cell-line names.

It does not help for:

  • a claim about an experimental design or intent (no experiment has been designed to ask whether WRAD reaches a KMT2A fusion), because a design is not a named object;
  • a claim about whether a model exists (the sarcoma case) - use Cellosaurus;
  • a claim about a database record (no phosphosite mapped) - use PhosphoSitePlus or PRIDE;
  • a measurement that only ever appears in a supplementary figure, which rule 80 already says no synonym list can reach.

Three of the eleven here were of the second and third kinds, and their survival is a property of the instrument as much as of the literature.

Facts and figures
scriptanalyses/q159-retest-the-abstract-level-bounded-absences/probe.py
statusrun
result summary11 abstract-level absences re-tested with agent-name and cell-line-name queries. 10 SURVIVE, 1 NOT-CHECKABLE, 0 OVERTURNED. 16 logged queries, 388 records, 207 titles hand-read, every querytranslation inspected.
absence classliterature-absence-bounded
absence class basisEvery verdict here rests on a PubMed query with its translation recorded, and says nothing about whether the data exists in an archive.
measurements
proportion0.000proportion: 0count11.0count: 11count388.0count: 388
inputs
[
  "analyses/q159-retest-the-abstract-level-bounded-absences/probe-log.jsonl"
]
verdicts
{
  "OVERTURNED": [],
  "SURVIVES": [
    "absence-no-proximity-degron-or-crispr-study-of-wrad-in-kmt2a-fusion-leukaemia",
    "absence-ledgf-agent-never-tested-in-menin-inhibitor-resistant-disease",
    "absence-h3k79me2-decay-half-life-never-measured",
    "absence-mllt10-kinase-site-relations",
    "absence-enl-yeats-and-condensate-phospho-gating",
    "absence-no-idh-inhibitor-has-been-reported-in-a-kmt2a-rearranged-cohort",
    "absence-nothing-has-ever-been-measured-in-parathyroid-under-a-menin-inhibitor",
    "absence-no-menin-inhibitor-occupancy-dataset-in-an-infant-derived-kmt2a-aff1-line",
    "absence-no-junction-specific-antibody-exists-for-any-kmt2a-fusion",
    "absence-no-experiment-tests-cxxc-reader-competition-at-a-kmt2a-fusion-target"
  ],
  "NOT-CHECKABLE": [
    "absence-no-menin-dependence-measurement-in-kmt2a-r-sarcoma"
  ]
}
discarded token traps
{
  "SEM": "3 of 6 hits were scanning electron microscopy (FIB-SEM, CoLSSEM)",
  "p75": "12 of 27 hits were TNF receptor p75 / p75NTR / p75LNTR, not LEDGF",
  "AF10 / AF-10": "an IL-6-responsive B cell line of that name, plus atrial-fibrillation cardiology; >=13 of 23 discarded",
  "MLL": "prostate cancer, a HER2/neu rat vaccine, benzene toxicology, RNP-masked nicks"
}
measurements — the numbers behind the chart
[
  {
    "quantity": "count",
    "subject": "abstract-level bounded absences re-tested with an agent-name and cell-line-name query",
    "value": 11,
    "units": "entries",
    "n": 11,
    "universe": "the 26 pre-census absence-* entries carrying a [tiab] query",
    "null": null,
    "note": "selection was by claim SHAPE (no agent/reagent/assay tested in model X), not at random, so this is not a random sample of the 26"
  },
  {
    "quantity": "proportion",
    "subject": "re-tested absences OVERTURNED by the agent-and-model instrument",
    "value": 0,
    "units": null,
    "n": 11,
    "universe": "11 absences re-tested in beat q159",
    "null": 0.333,
    "note": "0 of 11; rule-of-three 95% upper bound 27%. Null is the cycle-143 result 1 of 3 tried, which is the only prior estimate this project has"
  },
  {
    "quantity": "count",
    "subject": "PubMed records returned across the 16 logged q159 re-test queries",
    "value": 388,
    "units": "records",
    "n": 16,
    "universe": "probe-log.jsonl, 2026-09-24",
    "null": null,
    "note": "207 titles were hand-read; the remainder are the tail of the 115-record LEDGIN query"
  }
]
How this could bite us back
  • Reporting ten survivals as a clean bill of health would be wrong: a survival under a title-level re-read is still bounded by abstracts, and rule 80 says no synonym or compound-name list reaches a measurement that lives only in a supplementary figure.
  • The one absence touching an intervention directly - the LEDGF agent in menin-inhibitor-resistant disease - survives for a reason that should deter, not encourage: no LEDGF-directed agent exists to test, and the entry's own structural argument says the LEDGF grip is TIGHTER in gatekeeper-mutant resistant cells, so the obvious second-line rationale is backwards for that resistance class.
Still open
  1. Fifteen of the 26 pre-census abstract-bounded absences were not re-tested here, selected out because their claim shape does not fit an agent-and-model query. What instrument fits each of them, and is there a shape for which no instrument exists?
  2. The cycle-143 overturn and this beat together give 1 of 12. Is the true overturn rate low, or is the selection biased - was the HDAC6 case picked first precisely because it looked weakest?
  3. Should `p75`, `AF-10` and `SEM` be added to data/symbol-collisions.json? SEM already appears in that registry's discussion but p75 and AF-10 were not found there; I do not own that file and have not edited it.
Every source behind this entry (5)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Querying WDR5 degrader compound names returns a Science Translational Medicine paper reporting a selective WDR5 PROTAC degrader with DC50 3.7 nM in MV4;11 and effects on the chromatin-bound fraction of MLL complex components, but no experiment that distinguishes wild-type KMT2A from a KMT2A fusion.

    A selective WDR5 degrader inhibits acute myeloid leukemia in patient-derived mouse models. · Yu X, Li D, Kottur J, Shen Y, Kim HS, Park KS, Tsai YH, Gong W, Wang J, Suzuki K, Parker J, Herring L, Kaniskan HU, Cai L, Jain R, Liu J, Aggarwal AK, Wang GG, Jin J · Sci Transl Med 2021
    PMID 34586829DOI
    What this evidence cannot show

    Full text read via PMC8500670 on 2026-09-24. The chromatin-fractionation panel uses an MLL antibody that is not stated to discriminate wild-type KMT2A from the MV4;11 fusion, so the experiment cannot settle the WRAD-to-fusion question even though it is a rapid-degradation association experiment in the index-fusion line.

    Evidence that argues against this (1)
    • This corpus already cites this paper in six entries and already states the narrow conclusion: concept-wdr5-what-is-established-for-wild-type-kmt2a-and-what-for-the-fusion says 'No published co-immunoprecipitation, mass spectrometry, structure, degron or proximity experiment reports WDR5 with any KMT2A fusion protein.' Treating this as a new overturn would be rediscovering a decision the corpus had already made. Oncofusion Knowledge Graph 2026
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    A query over the named LEDGF-directed chemical series - LEDGIN, ALLINI, BI-224436, pirmitegravir, STP0404, GS-9822, CX14442 - returns 115 records and every one of the 40 read is HIV-1 integrase pharmacology, with none in leukaemia and none in a menin-inhibitor-resistant model.

    What this evidence cannot show

    40 of the 115 titles were read, not all 115, so this is a bound on the first 40 by PubMed's default relevance order plus the observation that the whole set is an HIV-integrase literature. Two of the twelve compound codes tried - CX05045 and CX-05168 - returned zero and were silently dropped from the translation, so they contributed nothing; see the companion entry on that failure mode.

    Evidence that argues against this (1)
    • The LEDGINs bind HIV-1 integrase at the LEDGF-binding site, not the LEDGF IBD, so finding no leukaemia use of them is weak evidence about a hypothetical IBD-directed agent. The entry being tested says the real bound is that no LEDGF-directed agent exists to test, which no query can strengthen. Adv Exp Med Biol 2021
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    An ENL YEATS agent-name query (SR-0813, TDI-11055, GNE-064, SGC-iMLLT) returns 10 records and none pairs YEATS reader function or ENL condensates with phosphorylation, so the phospho-gating absence survives against the agent-name instrument as well as against the target-name one.

    What this evidence cannot show

    Titles only; a phospho experiment reported in a figure of one of these ten would be invisible, which is rule 80 and is not repaired by switching from target names to agent names. GNE-064 is in fact a SMARCA2/4 bromodomain probe and was an error in my name list - it contributed one off-beat record.

    Evidence that argues against this (1)
    • One of the ten is a 2026 Exp Hematol paper testing SGC-iMLLT in KMT2A::AFF1 leukaemia in vivo with a healthy-HSPC comparison, which is substantive new evidence about the same agent and was absent from every node file in this corpus. The absence survives, but the agent-name instrument still found something the corpus did not have. Exp Hematol 2026
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    All five records returned by nineteen menin-inhibitor names against sarcoma terms concern Ewing sarcoma, which is not KMT2A-rearranged, so the query cannot reach the object of the KMT2A-r sarcoma absence.

    PMID 42690622
    What this evidence cannot show

    Title and journal read from esummary; the abstract of this paper was not opened, because the point it supports is only that the record is about Ewing sarcoma, which the title states. The other four records (PMIDs 41292793, 32333633, 29672864, 27888797) were also title-checked only.

    Evidence that argues against this (1)
    • If a KMT2A-rearranged sarcoma model were ever established, this query would become informative overnight; the NOT-CHECKABLE verdict is a statement about today's model landscape, not a permanent property of the question. Oncofusion Knowledge Graph 2026
  5. E5moderate confidence 3 of 5: One strong study, or several smaller consistent ones.review

    The closest published object to a junction-specific KMT2A fusion antibody is a flow-cytometric immunobead assay that detects fusion proteins using antibodies against the two partner proteins, which is a sandwich and not a junction-specific reagent.

    PMID 21123134
    What this evidence cannot show

    Identified from the title and journal in an esummary record; the full text was not opened. The sandwich-versus-junction distinction is inferred from what an immunobead assay is, not quoted from this paper, and a reader who needs that distinction to be load-bearing should open it.

    Evidence that argues against this (1)
    • If this assay's capture and detection antibodies were ever shown to require the fusion junction epitope rather than the two partner epitopes independently, it would be a junction-specific reagent and the absence would fall. Nothing in the title says either way. Oncofusion Knowledge Graph 2026
Connected to