Oncofusion

start / graph / ana-depmap-nup98-men1-power

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Qn-nup98-family

DepMap cannot answer whether MEN1 is selectively required in NUP98-rearranged lines, because it contains none — and it takes only ~3 to answer it

also called NUP98 DepMap MEN1 power analysis · zero NUP98 DepMap lines

We tried to check in a big public database of cancer cell lines whether NUP98-driven leukaemias need menin as much as KMT2A-driven ones do. There are no NUP98 cell lines in the database at all — and the effect is so strong that only about three would be needed to settle it.

The same thing, for a scientist

Run this cycle (`analyses/depmap-nup98-rearranged-men1-power/run.py`). Cell lines were resolved **by DepMap accession from Cellosaurus, never by name**, in both directions: `var:NUP98 AND dr:DepMap` with an HGNC-xref-labelled NUP98 fusion record, and `var:<PARTNER> AND dr:DepMap` over 27 partner and retired-alias symbols keeping any line whose fusion description contains NUP98. **Result: zero NUP98-rearranged DepMap lines.** The same machinery resolves **16 KMT2A-rearranged lines**, so the method works. Among leukaemia lines with MEN1 data (n=54) the KMT2A-rearranged arm (n=11) versus other-leukaemia arm (n=42) gives a MEN1 gene-effect difference of **−0.8455, Cohen's d = −1.879**. At that effect size, 80% power and two-sided α = 0.05 would require about **2.3 NUP98-rearranged lines** against the 42 available comparators, or 4.4 per group balanced. The obstacle is therefore not statistical but material: **three cell lines would do it and the resource has none.** The most valuable by-product is the KAT8 row — see `mechanism-kat8-mof-requirement-is-myeloid-not-fusion-specific`. Per the brief's instruction, no statistic is reported for the NUP98 arm, because reporting one computed on zero lines would be worse than reporting nothing.

Go deeper
What was resolved, and the double-check
index armDepMap lines resolved
NUP98-rearranged0
KMT2A-rearranged (positive control, same function)16
NUP214-rearranged4

The partner-side sweep covered NSD1, KDM5A, HOXA9, HOXA11, HOXA13, HOXD11, HOXD13, DDX10, TOP1, TOP2B, RAP1GDS1, PRRX1, PRRX2, JADE2, KMT2A, SETBP1, PHF23, BPTF, LNP1, POU1F1, IQCG, NSD3 and the retired spellings WHSC1, WHSC1L1, JARID1A, PHF15, PMX1. Not one DepMap-linked line carried NUP98 in any fusion description.

Across all of Cellosaurus (not just DepMap), var:NUP98 returns exactly one cell line — Kasumi-5, NUP98::RAP1GDS1 — and it has no DepMap cross-reference. So the absence is not a DepMap-annotation gap that a different database would fill; the NUP98-fusion cell-line resource barely exists.

The NUP214 lines are ALL-SIL, BE-13, Loucy and Peer. Three are NUP214-ABL1 T-ALL, a kinase fusion with no proposed menin dependency, and one is SET-NUP214. n=4 across two mechanistically different lesions cannot support a dependency claim and none is made — the values are in results.json marked descriptive-only.

The effect sizes, and why a z would have been the wrong number
genemean KMT2A-rmean other leukaemiadifferenceCohen's dn per group for 80% power
MEN1−1.2629−0.4174−0.8455−1.8794.4
KMT2A−0.8113−0.2732−0.5380−1.5256.7
MEIS1−0.4289−0.1324−0.2965−1.17611.4
DOT1L−0.6042−0.3782−0.2260−0.57647.4
ACTB (control)−0.9493−0.7626−0.1866−0.49065.4
HOXA9−0.2707−0.2157−0.0550−0.327147.0
KAT8−1.4888−1.4893+0.00060.003~2 × 10⁶

Cohen's d rather than a z, deliberately: a z measures how unusual a gene looks against a null panel, which depends on how tightly that panel is distributed, and it cannot be converted into a sample size. d can. The ACTB row is the honesty check — a pan-essential control should show no fusion specificity and shows d = −0.49, the residual per-line global sensitivity offset that this project's depmap-line-normalised-specificity analysis exists to remove. MEN1's d is nearly four times ACTB's, so the MEN1 signal comfortably exceeds the offset, but the uncorrected d should be read as an upper bound and the offset-corrected companion analysis is the one to quote for the KMT2A effect itself.

The actionable part: three cell lines

Because the MEN1 effect in KMT2A-rearranged lines is very large, the sample size needed to detect an equal effect in a NUP98-rearranged arm is tiny: about 2.3 lines against the 42 other-leukaemia comparators already in DepMap, or 4.4 per arm if the comparator arm were matched.

That converts an unanswerable question into a small, concrete ask: get three or four NUP98-rearranged leukaemia lines CRISPR-screened and deposited. It is a materials problem with a known price, not a statistics problem. A lab holding NUP98::NSD1 or NUP98::KDM5A models could close it. The same three lines would simultaneously answer the KAT6A/KAT7-versus-menin question and the partner-independence prediction in mechanism-nup98-fusion-menin-mll1-dependency.

The honest caveat on that ask: a cell line adapted to culture from a NUP98-rearranged AML may not retain the dependency, and the reason so few exist is partly that these leukaemias are hard to establish as lines. A power calculation cannot tell you whether the lines you get will be representative.

What this analysis does NOT show
  • It does not show that NUP98-rearranged leukaemia is not menin-dependent. It shows the DepMap route to the question is closed. The preclinical evidence that it is dependent comes from mouse models and PDX, which DepMap does not contain (mechanism-nup98-fusion-menin-mll1-dependency).
  • It does not establish the KMT2A MEN1 effect. That is prior work in this graph; the d is reused here only to set a power scale.
  • It does not report any NUP98 dependency value, because there is nothing to report.
Facts and figures
scriptanalyses/depmap-nup98-rearranged-men1-power/run.py
statusrun
reusesanalyses/depmap-line-normalised-specificity/run.py (kmt2a_rearranged_lines, fetch, LEUKEMIA_HINTS) imported, not copied
n nup98 rearranged depmap lines0
n kmt2a rearranged depmap lines16
n nup214 rearranged depmap lines4
n leukaemia lines with men1 data54
n kmt2a r arm11
n other leukaemia arm42
men1 difference-0.8455
men1 cohens d-1.879
kat8 cohens d0.003
actb control cohens d-0.49
nup98 lines needed for 80pct power2.3
result summaryZero NUP98-rearranged DepMap lines exist, so the question is unanswerable in DepMap; the effect size in the arm that does exist implies only about three NUP98 lines would be needed, making this a materials gap rather than a statistical one. KAT8 shows no fusion specificity at all.
inputs
[
  "Cellosaurus search API (line resolution by DepMap accession, both directions)",
  "Open Targets GraphQL (DepMap Chronos gene effect)",
  "HGNC complete set (gene universe, via the reused module)"
]
How this could bite us back
  • The cheapest experiment in this beat costs three cell lines, and nobody has done it — which should make anyone reading the preclinical NUP98 menin literature ask why the orthogonal check was never run.
  • A NUP98-rearranged line adapted to culture may have lost the dependency that matters, so three lines could give a clean negative that is an artefact of what survives culture; the power calculation cannot warn you about that.
  • Quoting the raw Cohen's d of -1.879 as the MEN1 effect would overstate it, because the ACTB control shows part of it is a per-line offset. Any downstream use must go through the offset-corrected analysis.
Still open
  1. Does any NUP98-rearranged leukaemia line exist anywhere that could be submitted to DepMap? Cellosaurus finds one NUP98-fusion line in total (Kasumi-5, NUP98::RAP1GDS1) and it is not in DepMap; whether it is CRISPR-screenable is unknown.
  2. Would a NUP98-rearranged PDX CRISPR screen substitute for the missing lines, and has anyone run one? The dependency literature uses PDX for drug response but not, as far as this pass found, for genome-scale dependency.
  3. Does DepMap's own OmicsFusion annotation contain a NUP98 fusion that Cellosaurus does not? That file was not cross-checked and is the one remaining way this zero could be wrong.
  4. Does the minimum detectable effect of d ~ 1.0 at these sample sizes mean the KAT8 null is uninformative about modest fusion-specific differences, and what sample size would rule out d = 0.3?
Every source behind this entry (3)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    Cellosaurus resolves zero NUP98-rearranged DepMap cell lines in either resolution direction, while the identical machinery resolves 16 KMT2A-rearranged lines and 4 NUP214-rearranged lines.

    What this evidence cannot show

    This is an absence in Cellosaurus's fusion annotation of DepMap models. A DepMap line whose NUP98 fusion is recorded in DepMap's own OmicsFusion data but not in Cellosaurus would be missed; that file was not cross-checked. The partner-side sweep mitigates but does not eliminate this. It does establish that no NUP98-rearranged line is retrievable by the accession-based route this project mandates.

    Evidence that argues against this (1)
    • Across all of Cellosaurus, var:NUP98 returns exactly one cell line (Kasumi-5, NUP98::RAP1GDS1) with no DepMap cross-reference. That is a check in the same direction but from a different angle, and it argues the gap is in the cell-line resource rather than in the annotation: NUP98-fusion leukaemias are largely studied as primary samples and PDX, not as lines. SIB Swiss Institute of Bioinformatics 2026
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    Among 54 leukaemia lines with MEN1 data, the KMT2A-rearranged arm (n=11) versus other leukaemia (n=42) gives a MEN1 gene-effect difference of -0.8455 with Cohen's d = -1.879, implying about 2.3 NUP98-rearranged lines would suffice for 80% power at two-sided alpha 0.05.

    DepMap CRISPR (Chronos) gene effect via the Open Targets Platform GraphQL API · Open Targets / Broad Institute DepMap · Open Targets Platform 2026
    What this evidence cannot show

    The Open Targets API does not state the DepMap release version, so the analysis is not pinned to a release. Cohen's d is uncorrected for each line's global dependency-sensitivity offset; the ACTB control shows d = -0.49, which bounds the size of that offset. The power calculation assumes equal variance and a normal approximation and is indicative at these sample sizes, not exact.

    Evidence that argues against this (1)
    • The ACTB pan-essential control also shows nominal fusion specificity (d = -0.49), which means part of the MEN1 difference is a per-line offset rather than biology. The offset-corrected companion analysis in this repository exists precisely because of this, and its corrected estimate, not the raw d here, should be quoted for the KMT2A effect. Oncofusion knowledge graph 2026
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    KAT8 shows essentially identical dependency in KMT2A-rearranged and other leukaemia lines (means -1.4888 vs -1.4893; difference +0.0006; Cohen's d = 0.003), while being strongly required in both.

    DepMap CRISPR (Chronos) gene effect via the Open Targets Platform GraphQL API, KAT8 (ENSG00000103510) · Open Targets / Broad Institute DepMap · Open Targets Platform 2026
    What this evidence cannot show

    A null difference at n=11 versus n=42 is consistent with a small real difference the study cannot resolve; the implied minimum detectable effect at this sample size is around d = 1.0, so only a large fusion-specific effect is excluded. Cell lines are not primary leukaemia, and a requirement that is equal across lines in culture could still differ in vivo.

    Evidence that argues against this (1)
    • Absence of differential requirement between KMT2A-rearranged and other leukaemia lines does not test the published claim, which is that MOF catalytic activity is required for both MLL-AF9 and NUP98-HOXA9 leukaemogenesis in mouse models. A gene can be uniformly required in established lines and still be selectively required for transformation, which is a different experiment with a different readout. SIB Swiss Institute of Bioinformatics 2026
Connected to