The Fanconi signal is in both lineage halves, survives the correlation-proof null, and is still there when lineage is held constant

ana-effect-size-vs-z-reranking left four objections on the table and answered none of them. Three are answerable with the same machinery and one needed a different null.

A max is not a conjunction. The consistency statistic is max(z_myeloid, z_lymphoid). A gene that is depleted only in the seven myeloid lines gets the same score as one depleted in both halves, and the published results file stores only the max. So nothing in the graph said whether the FA signal was lineage-wide. It is: both halves clear the null at p = 5e-05, and the lymphoid half is stronger 1.

The specificity control this data slice cannot run. The intended control was other fusion subtypes, resolved from Cellosaurus exactly as the KMT2A arm was. It fails for a reason worth recording rather than hiding: the Open Targets DepMap leukaemia slice holds 43 lines, and contains one BCR::ABL1 line, one PML::RARA line, one ETV6::RUNX1 line and two RUNX1::RUNX1T1 lines. Nineteen DepMap-registered BCR::ABL1 lines exist in Cellosaurus; one of them survives into this slice with a leukaemia disease string and a stable per-line offset. No fusion-specificity claim can be made from this resource at all, in either direction 1.

So the control became lineage. Non-KMT2A AML and non-KMT2A B-ALL arms are near-complementary halves of the same 43 lines, so together they answer one question — are FA genes more depleted in B-ALL lines than in AML lines — and the answer is yes, sharply: +0.095 against -0.196 on the eight headline genes. Read alone that looks like a lineage explanation for the whole finding.

The within-lineage contrast is the one that decides. Index and comparator drawn from the same disease string, so lineage cannot contribute. KMT2A-rearranged AML against other AML: -0.334. KMT2A-rearranged B-ALL against other B-ALL: -0.308. Both essentially the full effect. Lineage moves the baseline; KMT2A rearrangement moves the genes 1.

A permutation over gene labels asks: is this gene set more depleted than a random gene set of the same size? It assumes the members are independent draws. Physically interacting subunits are not — their gene effects correlate — so drawing 26 independent genes underestimates the variance of a 26-gene complex and the test is anti-conservative. The sweep's own caveat says exactly this. It is not a small objection: the FA/BRCA set is, structurally, one ubiquitin ligase plus its substrate plus its downstream effectors.

The fix is to permute the other margin. Hold the gene set completely fixed. Redraw which leukaemia lines are the index arm, stratified 7 myeloid + 4 lymphoid so the real arm's lineage composition is reproduced exactly, from pools of 32 myeloid and 21 lymphoid leukaemia lines. Every gene-gene correlation, every line's screen quality, and the lineage mix survive untouched. The only thing that varies is which lines are called the index.

For scale, MEN1 gives p = 0.0002 and FLT3 p = 0.0002 on the same null, and ACTB — the offset control — gives p = 0.092 1.

What this null still cannot do. It cannot distinguish a KMT2A effect from anything else those eleven particular lines share, because the draws are from other lines, not from other genotypes of the same lines. That is the limit of every observational dependency comparison and is why exp-replication-stress-readout-kmt2a-r exists.

A gene everything needs is not a selective target, and the earlier scans in this project were contaminated by exactly that. Computed across every screened line, not only leukaemia:

None is a common essential 1. That is a necessary condition for a selective target and nothing more: a gene can be dispensable everywhere and still not be selectively required in KMT2A-rearranged cells.

It does not show that the FA pathway is working in these cells, or working on transcription-replication conflicts, or that any of this is true of a primary blast. Eleven immortalised lines, none infant-derived. It also does not show that the dependency is on the FA pathway as a pathway: the obligate core subunits do not move together, which is set out in mech-fa-core-complex-heterogeneity-kmt2a, and that is the most serious unresolved problem with the whole result.

One more limit that matters for reading the numbers: the background panel here is 1,445 genes, not the 8,269 of the sweep, so ranks are out of roughly 1,400 and are not comparable to the sweep's ranks. Permutation p-values and Chronos differences are comparable; ranks are not.