A null that costs one command, run first rather than afterwards - and SSRP1 is genuinely LESS required in KMT2A-rearranged lines, while the rest of the chaperone layer is not much of anything

SSRP1 is the strongest signal in the set and it points the wrong way. Difference +0.265 - less required in KMT2A-rearranged lines than in other leukaemia lines - with p = 0.0104 and a background percentile of 99.5, meaning more than 99% of random protein-coding genes are more selective toward the index arm than SSRP1 is.

The caveat that governs it was already in the graph and the null does not remove it. gene-ssrp1 records that at a pan-essentiality of -1.271 every arm is dying, so the difference is between two lethal effects and is dominated by where each arm sits against the Chronos floor. The null establishes that the difference is not allocation noise. It does not establish that the difference means the cells need less SSRP1. Those are different claims and the entry should keep both.

And it is SSRP1, not FACT. The beat that flagged this described FACT as the least required gene set, quoting SSRP1 +0.265 and SUPT16H +0.117. SUPT16H's p is 0.2938 - the obligate partner shows nothing. A complex whose two subunits disagree at p = 0.01 and p = 0.29 is not behaving as a complex in this measurement, which is its own small finding and argues against the complex-level reading rather than for it.

What it does not touch: CBL0137 has potent antileukaemic activity against 12 MLL-rearranged lines, 3 PDX populations and an MV4;11 xenograft. A drug that traps FACT into chromatin is not the same perturbation as deleting a subunit, and a gene-effect difference says nothing about it.

CHAF1B and H2AZ1 sit at background percentile 0.3 - as extreme against random genes as anything in the set - with permutation p of 0.2103 and 0.1938. DAXX and NASP sit at the same percentile and are significant.

The two tests are asking different questions and the disagreement is informative rather than a defect. The percentile asks is this difference large compared with random genes; the permutation asks could this gene's own lines have fallen this way by chance. A gene with a large difference and high within-arm variance is extreme on the first and unremarkable on the second. CHAF1B is the clearest case: a -0.187 difference on a gene whose pan-essentiality is -2.314, the deepest in the set, where both arms are far past the floor and the spread is wide.

This is exactly the combination the tool was built to expose - a real-looking difference in a place where differences are common, or a large difference that one arm's spread can produce on its own. An analysis reporting either null alone would have called CHAF1B a hit or called DAXX unremarkable.

CHAF1B's DepMap numbers are uninformative for the reason above. Its published biology is not, and it is the dose lesson of the whole layer: heterozygous Chaf1b deletion, or a CAF-1 dominant-negative allele, is sufficient to suppress leukaemogenesis in vivo without impairing normal haematopoiesis 3.

A CRISPR knockout screen cannot see that. It measures complete loss, where CHAF1B is lethal everywhere; the therapeutic window sits at partial loss, which the screen does not sample. A gene can be pan-essential in DepMap and have a real window in a mouse, and this graph's standing convention - discard anything with pan-essentiality below -0.5 - would have discarded it.

That convention is right for its purpose and wrong as a filter on therapeutic interest. It is recorded here because the same convention has now been applied in three analyses in this project without this caveat attached.