Oncofusion

start / graph / ana-q124-prediction-5-of-hyp-q117-has-three-clauses-and-an-h3k79me2-arm-can-reach-one

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q124-the-timecourse-and-the-pcgf1-knockout

Prediction 5 of hyp-q117 contains three separable claims - normal eviction, no silencing, and return on withdrawal - and the deposit it names supports one of them only through a clause it never states; the eviction and silencing clauses need tracks the NTvKO arm does not carry, and withdrawal is absent from all 158 samples

also called scoring prediction 5 · which clause an H3K79me2 arm can reach · the reachability declaration, made before the data

A standing prediction in this project says that if you delete a PRC1.1 component, the fusion protein is still removed by the drug but the empty spots are not switched off, and the disease programme comes back when the drug stops. The dataset it points at can only test a fourth thing the prediction takes for granted. Saying which of the three claims was tested is more useful than a verdict.

The same thing, for a scientist

**`hyp-q117`'s prediction 5 is stored twice in its own entry, in two different strings**, and both name the deposit this beat analysed. The long form, in `sections[0].body_md` item 5, verbatim: *"**In a PCGF1 or BCOR knockout, the fusion is evicted normally but the vacated sites are not silenced**, and the programme returns on drug withdrawal. The depositing paper's NTvKO arm is the relevant deposit and this beat did not touch it."* The short form, `data.predictions[4]`, verbatim: *"in a PCGF1 or BCOR knockout the fusion is evicted normally but the vacated sites are not silenced and the programme returns on drug withdrawal."* **Decomposed, it is three claims with three different instruments and three different absence classes** [E1]: | clause | the instrument it needs | is it in GSE260742's NTvKO arm? | absence class | |---|---|---|---| | **1. the fusion is evicted normally** | a fusion-occupancy track - FLAG, MLL1 or MENIN ChIP | **NO.** The NTvKO arm deposits H3K79me2, H2AK119ub, H3K27me3, H3K4me3 and inputs, and no occupancy channel of any kind | `never-measured` in this series | | **2. the vacated sites are not silenced** | H2AK119ub or H3K27me3 gain, or transcription | **YES, and nobody has binned it** - H2AK119ub in three replicate conventions and H3K27me3 in one | **`measured-not-analysed`** | | **3. the programme returns on drug withdrawal** | a washout arm | **NO** - zero washout or withdrawal samples in all 158 records | `never-measured` in this series | **So the arm this prediction names cannot reach two of its three clauses, and this was declared in the pre-registration as refutation RP2 before any file was downloaded** - specifically so that a beat scoring one clause of three could not report "prediction 5 confirmed." `docs/QUEEN_BRIEF.md` rule 96 exists for exactly that error. **What the arm CAN reach is a fourth claim the prediction assumes rather than states: that the DOT1L axis behaves normally in the knockout.** Prediction 5's logic requires the fusion's removal to proceed as usual so that the *only* thing the knockout changes is what happens afterwards. This beat tested the closest measurable proxy - whether either drug's effect on H3K79me2, the mark the fusion's DOT1L recruitment deposits, depends on PCGF1 - and it does not, by 0.0166 and 0.0101 against a 0.0584 two-culture spread. **The proposed cashed-outcome text for the coordinator to append is in `data.proposed_cashed_outcome_text`, ALL-CAPS-led per rule 96, and it must go into BOTH strings**, because leaving the outcome in only one would leave the entry arguing with itself.

Go deeper
The one-beat route to clause 2, with accessions

Clause 2 is the only clause a download can close, and the design is better than the H3K79me2 one because it carries replicates.

H2AK119ub, NTvKO, murine MLL-AF9, mm10 - THREE replicate conventions
  paired: GSM8123403/405/406 (Control DMSO/SGC/VTP), GSM8123399/401/402 (PCGF1 KO DMSO/SGC/VTP)
  rep1:   GSM8123410/411/412 (Control),               GSM8123407/408/409 (PCGF1 KO)
  rep2:   GSM8123420/421/422 (Control),               GSM8123417/418/419 (PCGF1 KO)
  inputs: GSM8123404 (F3_NT paired), GSM8123416 (NT input3), GSM8123413/414 (input1/input2, .filt.bam not .tdf)
          GSM8123400 (F3_C1 paired), GSM8123415 (C1 input)
H3K27me3, NTvKO
  GSM8123426/427/428 (Control DMSO/SGC/VTP), GSM8123423/424/425 (PCGF1 KO DMSO/SGC/VTP)
H3K4me3, NTvKO
  GSM8123432/433/434 (Control DMSO/SGC/VTP), GSM8123429/430/431 (PCGF1 KO DMSO/SGC/VTP)

The prediction that clause 2 makes is unusually sharp and unusually cheap to score: PCGF1 is the PRC1.1 subunit that carries the complex's H2AK119ub activity, so in a PCGF1 knockout the VTP-induced H2AK119ub gain that Q117 measured at the fusion's most-evicted bins should be abolished or strongly reduced. Q117's numbers for the control genotype are already in this graph - RING1B rising 1.38x in the most-evicted quartile against 0.81-0.88x elsewhere, and 16.4x at Meis1, 7.5x at Jmjd1c, 3.6x at Six1. A criterion of the form "the VTP H2AK119ub gain at the fusion's most-evicted quartile falls below 1.10 in the PCGF1 knockout while exceeding 1.25 in the non-targeting control" is scoreable, population-named, and needs twelve downloads.

Note two hazards for whoever runs it. Two of the NTvKO inputs (GSM8123413 and GSM8123414) are deposited as .filt.bam and not as .tdf, so they are not readable by the TDFReader instrument this project uses; GSM8123416 and GSM8123415 are .tdf and are the usable pair. And the three replicate conventions use three different filename families (F3_NT/F3_C1 for paired, bare NT/CLONE1 for rep1 and rep2), so 'replicate' here may mean the same thing it meant in Q122's section 2a - related cultures rather than technical repeats.

Why the decomposition is the finding

A falsifier or prediction that bundles three claims behind one 'and' can be reported as confirmed on its easiest clause, and this corpus has a documented instance of a criterion firing while its consequent was unsupported by the reagent it named - hyp-q117's own HUMAN ARM, which fired on an MLL1-N antibody that cannot distinguish the fusion from wild-type MLL1.

The general form: before running a multi-clause prediction, write out one row per clause with the instrument each needs and whether the deposit carries it, and do it before downloading anything. Here that exercise changed the beat's output from a verdict into a map: one clause scored, one clause priced at twelve downloads with its accessions transcribed, one clause priced at wet work. The prediction's own text was the only thing needed to do it, and it had been sitting in the entry since cycle 68.

Facts and figures
proposed cashed outcome textPARTIALLY SCORED, CYCLE 83 (Q124) - ONE CLAUSE OF THREE, AND THE OTHER TWO ARE UNREACHABLE FROM THE ARM THIS PREDICTION NAMES. The NTvKO deposit named here was analysed: GSE260742 GSM8123438/439/440 (non-targeting-sgRNA control 'NT') and GSM8123435/436/437 (PCGF1-knockout clone 'C1'), a full 2 x 3 of DMSO 72 h / SGC0946 5 uM 72 h / VTP-50469 500 nM 48 h, in murine MLL-AF9 (KMT2A::MLLT3) myeloid cells - 2 kb bins, mm10 verified from each file's own extents, universe 1,227,708, scored at the FLAG-DMSO top-0.1% set defined in NEITHER genotype. **CLAUSE 1, 'the fusion is evicted normally': NOT SCOREABLE FROM THIS DEPOSIT.** The NTvKO arm deposits no FLAG, MLL1 or MENIN track, and H3K79me2 is not fusion occupancy. **CLAUSE 2, 'the vacated sites are not silenced': NOT SCOREABLE FROM AN H3K79me2 ARM**, but it is one beat away - the identical 2 x 3 design is deposited for H2AK119ub in three replicate conventions (GSM8123403/405/406 and 399/401/402 paired; 410/411/412 and 407/408/409 rep1; 420/421/422 and 417/418/419 rep2) and for H3K27me3 (GSM8123426/427/428 and 423/424/425), and none has been binned. **CLAUSE 3, 'the programme returns on drug withdrawal': NEVER MEASURED - no washout or withdrawal arm exists in any of the 158 sample records of GSE260742.** **WHAT WAS SCORED is the clause the prediction assumes rather than states - that the DOT1L axis behaves normally in the knockout - and it HOLDS: the menin inhibitor's H3K79me2 share ratio at the fusion's own sites is 0.8156 in the control culture and 0.7990 in the PCGF1 knockout, a difference of 0.0166, and the DOT1L inhibitor's is 0.1415 against 0.1314, a difference of 0.0101 - both below the 0.0584 that two untreated control cultures of this same line already differ by (Q122's F3-against-F3_NT spread at the same set). F = (1-V)/(1-G) is 0.2148 and 0.2314, inside Q122's 0.0947-0.2500 range, and is a LOWER BOUND because the VTP arm is 48 h and its vehicle is 72 h. **TWO LIMITS THAT MUST TRAVEL WITH THIS OUTCOME.** First, the knockout is UNVERIFIED in the files analysed: the paper's key-resources sheet lists one mouse guide (Mouse PCGF1 sgRNA1, CACCGCTGTGGATTTAGCATTAGT) and the deposit has one clone with no rescue, and no PCGF1 protein or transcript measurement is deposited for these samples - so a null cannot exclude an incomplete knockout. Second, PUBLISHED WORK REPORTS THE OPPOSITE AT THE LEVEL THAT MATTERS: PMID 39158067 (Zhou X et al., Blood 2024, DOI 10.1182/blood.2023023644) shows loss of PCGF1 or BCOR CONFERS MENIN INHIBITOR RESISTANCE by reactivating bivalent noncanonical menin targets including MYC. Both can be true, and together they say something sharper than either: PCGF1 loss changes what a menin inhibitor does to the CELL without changing what it does to H3K79me2, so the mark cannot report the resistance state that PRC1.1 deficiency creates. See ana-q124-the-menin-inhibitors-h3k79me2-effect-is-unchanged-in-a-pcgf1-knockout and ana-q124-prediction-5-of-hyp-q117-has-three-clauses-and-an-h3k79me2-arm-can-reach-one.
do not edit noteProposed only. That entry belongs to the Q117 beat and this beat does not edit it; the coordinator appends.
statusrun
result summaryPrediction 5 decomposes into three clauses with three instruments. Clause 1 (normal eviction) and clause 3 (return on withdrawal) are unreachable from GSE260742 - no occupancy track in the NTvKO arm, no withdrawal arm anywhere in 158 samples. Clause 2 (vacated sites not silenced) is measured-not-analysed and is twelve downloads away, accessions transcribed. The clause actually scored is the one the prediction assumes: the DOT1L axis is PCGF1-independent, and it holds at 0.0166 and 0.0101.
absence classmeasured-not-analysed
absence class noteMixed by clause and recorded per clause in data.clause_reachability rather than collapsed to one value. Clause 2 is measured-not-analysed: the H2AK119ub and H3K27me3 2 x 3 designs are public and unbinned, twelve downloads. Clauses 1 and 3 are never-measured IN GSE260742: no occupancy track exists in that arm and no withdrawal sample exists in the series, so closing them needs new experiments rather than new analysis. The single value on this field is the most optimistic of the three and the field cannot express a per-clause answer, which is itself worth recording.
absence class assigned2026-09-23 (Q124)
clause reachability
{
  "declared_before_the_data": true,
  "clause_a_the_fusion_is_evicted_normally": {
    "reachable": false,
    "why": "There is no FLAG, MLL1 or MENIN track in the NTvKO arm of GSE260742, and H3K79me2 is not fusion occupancy.",
    "absence_class": "never-measured"
  },
  "clause_b_the_vacated_sites_are_not_silenced": {
    "reachable": false,
    "why": "Silencing is H2AK119ub/H3K27me3 gain or transcriptional shutdown; this arm measures neither. The series deposits BOTH for this exact 2x3 design - H2AK119ub GSM8123403/405/406 and 399/401/402 (paired), 410/411/412 and 407/408/409 (rep1), 420/421/422 and 417/418/419 (rep2); H3K27me3 GSM8123426/427/428 and 423/424/425 - so this clause is one beat away.",
    "absence_class": "measured-not-analysed"
  },
  "clause_c_the_programme_returns_on_drug_withdrawal": {
    "reachable": false,
    "why": "No washout or withdrawal arm exists in any of the 158 sample records of GSE260742.",
    "absence_class": "never-measured"
  },
  "what_this_arm_does_reach": "whether the H3K79me2 response to a menin inhibitor and to a DOT1L inhibitor is PCGF1-dependent - the DOT1L-axis clause, which prediction 5 assumes rather than states. Prediction 5 is therefore scored PARTIAL BY CONSTRUCTION."
}
where to append it
[
  "hyp-q117-menin-inhibition-evicts-the-fusion-without-redistributing-it-to-a-recognisable-programme -> data.predictions[4], appended to the existing string",
  "the same entry -> sections[0].body_md, item 5 of the prediction list, appended to the existing sentence"
]
How this could bite us back
  • The decomposition makes prediction 5 look much more expensive than it did. It was written as one prediction with one named deposit; it is three predictions, of which one needs twelve downloads, one needs a fusion-occupancy ChIP that nobody deposited, and one needs a washout experiment that does not exist. A hypothesis whose predictions are priced as one unit will be believed to be nearer testing than it is.
  • If a reader takes 'the DOT1L axis is PCGF1-independent' as confirmation of prediction 5, the error is exactly the one rule 96 was written for, and the ALL-CAPS text proposed here is the only guard against it. Its first clause therefore says PARTIALLY SCORED and names which clause, before any number appears.
Still open
  1. How many other predictions and falsifiers in this corpus are stored in two places, so that an outcome appended to one leaves the other asserting an untested claim? A corpus-wide sweep for the same text appearing in both a data field and a sections body was not run. Rule 96 counted 624 stated falsifiers and 17 recorded outcomes using a detector that had to learn the ALL-CAPS convention; a detector that reads only one of two locations would undercount outcomes the same way.
  2. Does the absence schema need a per-clause form? This entry's absence_class field has to carry one value for a prediction whose three clauses are measured-not-analysed, never-measured and never-measured. The single value chosen is the most optimistic, which is the wrong default for a reader deciding what to run next.
  3. Is 'the programme returns on drug withdrawal' testable from any deposit anywhere, in any KMT2A-rearranged system? This beat bounded it within GSE260742 only and ran no archive query outside this series, which is exactly the weak form of bounded absence this project's rules warn about.
  4. What did the depositing authors conclude about the PCGF1 knockout's effect on the menin inhibitor response? The paper's figures are image-only and outside reach, so this beat scored a prediction against a deposit without reading the depositors' own reading of the same samples.
Every source behind this entry (2)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    GSE260742's NTvKO arm deposits, for the non-targeting control and the PCGF1-knockout clone under DMSO / SGC0946 5 uM / VTP-50469 500 nM: H3K79me2 (GSM8123438/439/440 and 8123435/436/437), H2AK119ub in three replicate conventions (GSM8123403/405/406 and 8123399/401/402 paired; 8123410/411/412 and 8123407/408/409 rep1; 8123420/421/422 and 8123417/418/419 rep2), H3K27me3 (GSM8123426/427/428 and 8123423/424/425), H3K4me3 (GSM8123432/433/434 and 8123429/430/431) and five inputs (GSM8123404/413/414/416 control, GSM8123400/415 knockout, of which GSM8123413 and GSM8123414 are deposited as .filt.bam and not .tdf). It deposits NO FLAG, MLL1 or MENIN track for either genotype, and no washout or drug-withdrawal sample exists in any of the 158 sample records of the series.

    What this evidence cannot show

    This is a claim about a DEPOSIT, not about a study: the paper may contain occupancy or washout experiments whose data were not deposited, and its MOESM1 is image-only with zero extractable characters so its figure panels were not examined. The channel inventory is read from !Sample_title and !Sample_supplementary_file_1 over the 158 records; a track deposited under a title this beat's grep did not anticipate would be missed, and the grep was over the strings 'NTvKO', 'Timecourse', 'K79', 'K119', 'K27', 'K4' and 'INPUT'. The three H2AK119ub replicate conventions use three different filename families (F3_NT/F3_C1 for 'paired', bare NT/CLONE1 for rep1 and rep2), so whether they are technical replicates or related cultures is not settled by the records - Q122's section 2a established that this series' titles and filenames disagree about exactly this.

    Evidence that argues against this (1)
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    hyp-q117's prediction 5 is stored in two different strings inside the same entry - a long form in sections[0].body_md item 5 which names the NTvKO deposit and records that the beat did not touch it, and a short form in data.predictions[4] which does not - so an outcome appended to one of them leaves the other asserting an untested prediction.

    What this evidence cannot show

    The two strings are not contradictory, only unequal in detail, so this is a maintenance hazard rather than an error in that entry. It is reported and NOT edited, per the standing instruction that a beat does not edit another beat's entry. Whether the same duplication exists for the other four predictions in that entry was not checked; a corpus-wide sweep for predictions and falsifiers stored in two places was not run and is named in the open questions.

    Evidence that argues against this (1)
Connected to