Bounded absence: every chromatin measurement under a menin inhibitor is in an adult-derived line, and none is in an infant KMT2A::AFF1 model

Systems were read from the methods, figure legends and GEO sample titles of the five full texts and from the abstracts of the remaining three datasets, so a supplementary arm in one of the three abstract-level papers could contain an infant line. The enumeration itself is bounded by a 61-record PubMed query plus targeted full-text reading, not by an exhaustive search. SEM appears in the pivotal paper as SEMK2 in viability assays, which is a use, just not a chromatin one.

• RS4;11 and MV4;11 both carry the index fusion KMT2A::AFF1, so the fusion itself is represented even though the developmental context is not; a reader who considers the fusion the dominant variable would judge this gap much less important than one who considers cell of origin dominant. Janssens DH et al. 2024

One cell line, one laboratory, peak-overlap statistics sensitive to calling stringency. SEM is an immortalised line from a 5-year-old at relapse of infant-diagnosed disease in the standard characterisation, so 'infant-derived' is itself an approximation that should be checked before it is leaned on. BOUND ADDED AND THE BROAD CLAIM REFUTED, 2026-09-22. PubMed returns 0 for `SEM[tiab] AND (revumenib[tiab] OR "menin inhibitor"[tiab] OR MI-503[tiab] OR VTP-50469[tiab] OR MI-2-2[tiab])`, with controls that fire through the identical code path: `SEM[tiab] AND KMT2A[tiab]` = 8, `revumenib[tiab]` = 103, `SEM[tiab] AND leukemia[tiab]` = 337. But the PubMed zero is a false negative for the DATA question, and GEO settles it: **GSE63664, 'Gene expression induced by DOT1L and Menin inhibition in cell line models of leukemia', contains samples titled `SEM, MI-2-2 treat, biol rep 1` and `biol rep 3` with `SEM, DMSO treat` vehicle controls**, alongside RS4;11 under the same design. So SEM HAS been profiled under a menin inhibitor. What GSE63664 is, by its own title, is EXPRESSION -- not occupancy -- so this entry's headline absence, that no menin-inhibitor OCCUPANCY dataset exists in an infant-derived KMT2A::AFF1 line, is not refuted by it. Note also that `SEM` is a contaminated GEO token: a search for SEM with menin-inhibitor names returns 16 records whose samples are MOLM-13 and REH, matching 'SEM' as a word elsewhere in the record.

• If the uCpG landscape of SEM differs from that of RS4;11, then importing the 20% figure into a discussion of the profiled lines is the same partner-imported reasoning this project keeps finding in its own evidence base. Kerry J et al. 2017

The 36-sample cohort's age distribution was not read for this entry and may include infants; the point stands only for the revumenib-treated arm, which is a single sample undergoing lineage switch. Automated CUT&Tag on primary material requires viable cells at a defined timepoint on therapy, which is an ethical and logistical constraint in infants rather than a technical one.

• If the 36-sample cohort does include infant samples profiled at baseline, the gap is narrower than stated - baseline infant maps would at least supply the uCpG and occupancy denominator against which a treated sample could later be compared. Janssens DH et al. 2021

CORRECTED 2026-09-24 TWICE OVER. This caveat previously said every arm assignment was unsafe because the deposit's titles disagree with its own characteristics; GSE159608 has ZERO such contradictions and the labels are usable [E8]. It also said a cycle-120 agent had named two of four accessions wrongly; THE AGENT WAS RIGHT AND THE COORDINATOR WAS SHIFTED BY ONE - its GSM5139713/14 for VTP and GSM5139709/10 for DMSO are correct. NOTHING IN THIS ENTRY HAS OPENED THE SIGNAL: no library was downloaded, no peak called, no retention computed here, so this establishes only that the experiment was deposited, and the H3K4me3 readout under drug is a histone mark rather than the fusion's or menin's position on chromatin. The SEM menin CUT&Tag samples remain DMSO-only, which is why this entry's occupancy claim survives.

• Treating the H3K4me3 readout as a refutation of an OCCUPANCY absence stretches the term. A histone mark under drug is not the fusion's or menin's position on chromatin, and this entry's title, its enumeration and the argument it feeds are all about occupancy. Read strictly, the entry's central claim was never wrong and only one supporting sentence was - which is a smaller correction than the banner above implies.
• This deposit was already in this project's hands. Cycle 116 parsed all 419 of these sample records while searching 32 series and 2,312 samples for a menin readout under DOT1L inhibition, and the SEM VTP-50469 arms were in the file it read. The absence survived because the question asked then was a different one, which means the cached corpus should be re-queried whenever an absence is asserted, not only when new data arrive.
• The accession list in this evidence has now been wrong once and rewritten twice in two cycles, both times by the same coordinator, and the version that was PRESENTED as a careful re-parse of the primary records was the wrong one. A reader should treat any accession-level detail in this entry as needing a raw check rather than as settled, and note that the only independent confirmations came from agents.

THE REPLACEMENT WAS RUN BY THE AUTHOR OF THE BUG, THE SAME DAY. The two tokeniser normalisations that took the count from 7 to 0 were written AFTER seeing which cases failed, an uncontrolled degree of freedom; both are unambiguous on the raw records, but the sequence was 16.4% to 1.1% to 0% with the instrument edited between the last two steps. A ZERO IS NOT PROOF THE DEPOSITS ARE CORRECT, only that their titles and characteristics agree - both could be wrong together. The comparison is only possible where BOTH fields state something: 666 comparable pairs out of 2,312 samples, so most samples are outside this audit entirely.

• A correction that makes the data look cleaner is the direction that gets checked least, and this one was produced by fixing a single line in the script that was wrong. The same script has now reported two incompatible answers to the same question within twenty-four hours, and nothing external has confirmed either. The defensible position is that the contradiction rate is LOW, not that it is 1.1%.
• The original error was not caught by the pre-registration, by the manual reading of three cases, or by the coordinator at all - it was caught by an agent dispatched on a brief that asserted the wrong number as established fact. The corpus's defence against this class of error is currently 'someone downstream refuses the premise', which is not a control and cannot be relied on.

CORRECTED 2026-09-24: THIS ARM IS n=1 PATIENT. 148752 and 152985 are the same infant before and after a lineage switch, stated in the GEO !Sample_description field as 'paired with' each other; an earlier version of this evidence described them as two patients. THE FUSION ASSIGNMENTS COME FROM A FIELD THIS DEPOSIT SHIFTS. Its characteristics lag its sample titles by one record - 9 of 21 comparable antibody assignments disagree, 42.9% - and the group field is offset too: GSM8371030 is titled SEM_AFF1_1 and carries group KMT2A-EPS15 although SEM is a KMT2A::AFF1 line. The KMT2A-ENL assignment for the two menin-inhibitor patients rests on consistency across their twenty records, which is internal agreement rather than correctness, and if it were wrong this entry's conclusion would fall. NEITHER THE COMPOUND NOR THE DOSE NOR THE EXPOSURE is named in the sample metadata, which says only 'the Menin inhibitor', so this dataset cannot be placed on the dose axis that cycles 115-117 established as where menin-inhibitor chromatin results stop being comparable. BOTH PATIENTS ARE LINEAGE-SWITCHING SAMPLES, a highly selected population, so even the KMT2A-ENL result is narrower than 'the infant experiment has been done'. NO LIBRARY WAS OPENED: no signal downloaded, no peak called, so this establishes what was deposited and not what it shows. Raw data for 51 of the 171 samples are withheld for donor privacy and which 51 was not determined.

• This entry has now been corrected twice in two cycles, both times by material the project already held - GSE159608 in cycle 120 and GSE252378 in cycle 121, the latter cited by one entry here throughout. Its enumeration of eight occupancy datasets was presented as exhaustive and was not, and the surviving claim survives on a fusion-identity technicality. A reader should treat the remaining bound as provisional rather than as twice-tested.
• Read strictly, the entry's title claim was never about histone marks or about non-index fusions, so neither correction touches it, and describing these as refutations overstates. The opposite reading is equally available: an enumeration that misses a ninth dataset on its first pass and a tenth on its second is not reliable evidence of exhaustiveness whatever its headline claim was, and the narrowing is the entry retreating to defensible ground rather than being vindicated.