The lymphoid-restricted DAXX dependency survives three nulls and two comparator restrictions - and ATRX, MEN1 and DOT1L do not survive the same test in the same lines

Run 2026-09-24 against the live Open Targets GraphQL DepMap endpoint; the API reports dataVersion year 26 month 09 with a null iteration, so the release is recorded to the month and not further. Raw Chronos effect sizes only, never a z divided by a comparator SD, per this project's cycle-45 correction. Lineage is binned from DepMap's `diseaseFromSource` string and every line is listed by name, CVCL accession and disease string in `results.json` and `b_lineage_restricted.json` so the binning can be hand-checked. Four cell lines are three donors. No infant KMT2A::AFF1 line exists in DepMap. Moderate rather than strong because the effective n is three and because the gene was selected on a correlated statistic from a 254-gene set.

• The same gene's POOLED difference was already tested and already failed: -0.3048 at permutation p 0.01282, which does not survive Bonferroni over the 254 genes the shortlist was drawn from. Multiplying this analysis's best donor-level p (0.00083) by the same 254 gives 0.21, so the lineage-split result fails the same correction the pooled result failed. A subgroup analysis of an already-selected gene cannot be treated as confirmatory, and the honest reading is that DAXX is a hypothesis worth an experiment rather than a dependency. Oncofusion knowledge graph (internal analysis) 2026

300 genes were drawn and 277 had usable strata; the 23 losses are genes not screened in enough lines, and they are not a random subset (poorly screened genes are systematically unusual), so the background is very slightly enriched for well-screened genes. The background is a sample of the genome, not of genes resembling DAXX in expression or pan-essentiality; a variance-matched background would be a stricter test and was not run. This test bounds the whole-stratum artefact only - it cannot address the selection of DAXX from a 254-gene set.

• Percentile-among-background is the one test here that DAXX's selection history can contaminate. DAXX was chosen as rank 4 of 254 on the pooled index-minus-comparator difference, and the pooled and lymphoid differences are not independent - the lymphoid lines are a subset of the index arm. An honest background test would draw from genes matched on pooled difference, which would shrink the percentile gap by an unmeasured amount. The test as run bounds the stratum artefact and nothing else. Oncofusion knowledge graph (internal analysis) 2026

Fetched 2026-09-24 through the Cellosaurus search endpoint. An earlier attempt through the `cell-line/{name}` endpoint returned a well-formed record with every provenance field empty for all four lines - a clean nothing, which is the failure shape this project distrusts, and it would have let the four lines be treated as four donors. The relationship recovered here is the one Cellosaurus asserts; no independent SNP-fingerprint comparison of SEM and SEM-K2 was performed.

• Collapsing SEM and SEM-K2 changes almost nothing quantitatively - the difference moves from -0.6008 to -0.6054 and the p from 0.00025 to 0.00083 - so in this instance the donor correction is close to cosmetic. The caveat that matters is not the arithmetic but that an n of three cannot support a lineage claim regardless of the p-value, and a permutation null cannot repair a sample; the same point is already on record for the pooled version of this comparison. Oncofusion knowledge graph (internal analysis) 2026

Consistency is not importance. MEN1's effect size is far larger than DAXX's at every comparator (-0.629 against -0.455 at the hardest one) and MEN1's spread is driven by RS4;11 sitting at -0.463, which is a real and separately interesting fact about that line rather than noise. A gene with a smaller, tighter effect will beat a gene with a larger, looser one on any variance-scaled statistic, which is precisely the failure this project corrected in cycle 45 - so this item deliberately reports the raw effect sizes and the separation counts rather than a z.

• This comparison flatters DAXX for a structural reason: menin dependency in KMT2A-rearranged leukaemia is established to be strongest in the myeloid arm and MEN1's myeloid difference here is -0.890, the largest in the whole scan. Restricting to the lymphoid stratum therefore tests MEN1 where it is known to be weakest and DAXX where its only signal is. A reader who takes 'DAXX beats MEN1' out of this entry has inverted it: within the lymphoid stratum DAXX is the more consistent discriminator, and across the whole index arm MEN1 is not remotely challenged. Oncofusion knowledge graph (internal analysis) 2026

The -0.70 threshold was chosen AFTER seeing the gap, which makes the group definition post-hoc; the hypergeometric p is conditional on a six-member deep group and would change under a different cut. It is reported because the gap is 3.7 times the next largest in the distribution and because a rank statement is immune to objections about the Chronos scale, not because the cut-point is principled. Thirteen lines.

• Two of the six deep lines are not KMT2A-rearranged, so the deep group is not a KMT2A group; the correct reading of the same table is that KMT2A-rearrangement is sufficient for membership in this panel and not necessary. If a mechanism is ever found, it has to explain RCH-ACV and YCUB-2 as well - and it cannot be 'a B-ALL fusion', because 697 carries the same TCF3::PBX1 as RCH-ACV and sits at -0.232, on the shallow side of the gap. Cellosaurus, SIB Swiss Institute of Bioinformatics 2026

A Spearman over eight points cannot exclude a moderate age effect; the honest statement is that the data do not support one, not that one is absent. Donor age at sampling is a weak proxy for developmental origin, which is the variable actually of interest, and it is missing for HB11;19 and JM1. Cellosaurus fusion annotation is a curated summary and will not list a fusion nobody reported in that line, so the 'no fusion' rows are uncurated rather than fusion-negative.

• Rejecting these two does not leave KMT2A-rearrangement as the explanation by elimination. A third variable that this analysis cannot see - most obviously telomere maintenance state, since ATRX/DAXX loss is the canonical route to ALT - would sort the lines the same way and has never been measured in any of the thirteen. The deep group should be treated as an unexplained cluster that happens to contain all four index lines. Heaphy CM et al. 2011