Co-dependency scan: what else does a menin-dependent cell need? (run, not planned)

The top two correlates with MEN1 are KMT2A (r = 0.58) and PSIP1/LEDGF (r = 0.53). Those are precisely the two proteins menin physically bridges in the chromatin tether. The method recovered the known module from dependency data alone, without being told about it 1.

The negative controls behave as they should. ACTB (r = 0.007) and GAPDH (r = 0.044) are deeply essential genes, with mean gene effects of -0.81 and -1.54, yet neither correlates with MEN1. So the correlation is measuring shared biology, not shared essentiality. TP53 sits at r = -0.017.

That combination, a clean positive control and clean negative controls, is what licenses reading the middle of the table.

Every gene in the elongation machinery the fusion hijacks sits at r near zero: CDK9 -0.00, BRD4 0.08, PRMT1 0.07, WDR5 -0.02, MED26 -0.04, CDC73 -0.05, SETD1A -0.02, PAF1 0.04, CTR9 0.03.

This is not a null result, it is a mechanistic one. These genes are required by essentially every cell, so their profile across lines is flat, and a flat profile cannot correlate with anything. The same property that makes them undruggable, no therapeutic window, also makes them invisible to this kind of scan.

The practical consequence: a co-dependency scan will never find you a target inside the pan-essential core. It can only find selective partners. Anyone hoping to identify a new KMT2A-fusion target this way should expect it to surface the tether and the output program, not the machinery. That matches what the interactome beat concluded from the opposite direction, using essentiality scores rather than correlation 2.

MEF2C is the only gene that is both a reasonable correlate (r = 0.27) and strongly leukemia-selective (-0.13, second best in the table). It is a known output-program gene of KMT2A fusions rather than a complex member, so this is consistent with it sitting downstream of the tether rather than beside it.

MYB has the strongest leukemia selectivity in the entire table by a wide margin (-0.41, against -0.13 for the runner-up) while correlating only weakly with MEN1 (0.11). Read together, that says MYB is a genuinely selective leukemia dependency that operates through a different route from the menin tether. For a project looking for an orthogonal combination partner, a strong selective dependency that is not menin-correlated is arguably more interesting than a correlated one, because uncorrelated risks are the ones a combination can actually hedge.

Both of these are hypothesis-generating only. Correlation across cell lines is not evidence of physical interaction, and neither observation has been tested in a KMT2A-rearranged model here.

Cancer cell lines are not infants. Only 152 of 1,208 lines are leukemia of any kind, and the panel is dominated by adult solid tumours, so the leukemia-selectivity column rests on a modest and heterogeneous subset. Correlation across lines reflects shared requirement, which can arise from physical complex membership, shared pathway, shared lineage restriction or shared expression, and this analysis cannot distinguish those. A gene expressed in few lines gives an unstable correlation regardless of biology.

The DepMap release version is not stated by the API, so the exact data vintage is unverified.

The first run silently returned data for the wrong gene: the Ensembl identifier used for LEO1 actually resolves to ITPRID2. It was caught only because the script reports which requested symbols came back empty, and an unexpected symbol appeared in the output table.

This is the same failure mode that has already produced two fabricated-looking citations in this project, where a plausible identifier pointed at an unrelated paper. Identifiers must be resolved against the source, never recalled. The corrected run puts LEO1 at r = 0.09, which changes no conclusion, but it would have been published as a real measurement of the wrong gene.

This entry argued that the elongation machinery shows no co-dependency because pan-essential genes have flat profiles, and a flat profile cannot correlate with anything. A later analysis testing all four WRAD subunits under the identical protocol shows that explanation is right for some genes and wrong for others.

WDR5 and RBBP5 are depleted in almost every line, so my explanation holds for them: there is no variance to correlate with.

ASH2L breaks it. It is depleted in only 21.4% of lines, less than DPY30, so it has more than enough variance to show a correlation if one existed. It shows none. The difference between DPY30 and ASH2L therefore cannot be a statistical artefact of essentiality; it is a claim about biology that needs a biological explanation.

The explanation was not wrong, it was too general. A flat profile does prevent correlation, but the absence of correlation does not imply a flat profile, and I wrote as though it did.