Oncofusion

start / graph / ana-non-menin-dependency-cross-validation

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q9-landscape

Analysis: cross-validating the single-group non-menin KMT2A-fusion dependencies in one framework

also called non-menin dependency audit · orphan dependency cross-validation

Several labs have each reported a different molecule that KMT2A-fusion leukemia cannot live without. Nobody has checked them against each other or against the big public screening data, so we do not know which ones are real.

The same thing, for a scientist

A public-data-only audit of every reported non-menin dependency of KMT2A-rearranged leukaemia, scored in a single consistent framework: DepMap CRISPR (Chronos) effect in KMT2A-r versus non-KMT2A-r lines, common-essentiality status, normal-tissue expression (GTEx/HPA), human loss-of-function constraint (gnomAD), existing chemical matter (ChEMBL), and the model system and fusion partner in which the original claim was made. The output is a ranked, comparable target list replacing a scatter of mutually incomparable single-group claims. It directly serves the lab's central thesis that menin is not the only obligate interface, and it is the cheapest way to find out whether that thesis has any candidates.

Go deeper
Why this is worth doing and why nobody has1 deeper

Each of these dependencies was reported once, by one group, in one model, with a different readout:

  • PSMB8 (immunoproteasome catalytic subunit) in KMT2A-complex-driven leukaemia, acting via BASP1 repression of fusion target genes, with a claim that normal haematopoietic cells are spared 1.
  • PTBP1 and PKM2-mediated metabolic dependency in MLL-ENL leukaemia 2.
  • ARID2 and ARID1B (SWI/SNF targeting subunits) with opposite, stage-dependent effects in MLL-AF9 3.
  • C/EBP family as a strict requirement for MLL-driven myeloid transformation 4.

They are in different fusion backgrounds (KMT2A-complex generally, MLL-ENL, MLL-AF9, MLL-ENL), different species, and different phases (initiation versus maintenance). Nobody has asked the obvious question: placed side by side under one scoring scheme, which of these has the essentiality-versus-dependency profile that menin has - selectively required by the fusion, tolerated by normal cells? That is precisely the criterion META_KING lists as a standing open question for the project.

It has not been done because it is nobody's paper. Each group's incentive is to advance its own target, not to rank it against four others.

Scoring framework

For each candidate, and for menin as the positive control and CDK9/WDR5/BRD4 as common-essential negative controls:

  1. Selectivity: mean Chronos in KMT2A-r lines minus mean Chronos in non-KMT2A-r lines.
  2. Tolerability proxy: DepMap common-essential flag, fraction of all lines depleted, gnomAD pLI/LOEUF.
  3. Normal-tissue footprint: GTEx/HPA expression, with particular attention to haematopoietic and fetal tissue where available.
  4. Evidence quality: replicated independently? in-vivo? human primary cells? correct fusion partner?
  5. Chemical tractability: ChEMBL matter against the target or its family; whether the dependency is enzymatic (druggable by inhibition) or interface-based (druggable only as a PPI).

The analysis explicitly records which candidates fail, because a candidate that disappears under cross-validation is as useful a result as one that survives, and it prevents the lab from building a programme on the next DOT1L.

The panel composition, now enumerated

This entry flagged that the DepMap KMT2A-rearranged line count had not been counted and should be before the "small, AML-weighted" claim was asserted. The coordinator enumerated it in ana-depmap-kmt2a-rearranged-specificity using Cellosaurus gene-fusion annotations cross-referenced to DepMap identifiers.

Eleven KMT2A-rearranged lines have MEN1 gene-effect data, out of 54 leukemia lines:

LineagenLines
AML7THP-1, NOMO-1, OCI-AML-2, MV4-11, MOLM-13, SHI-1, MOLM-14
B-ALL4SEMK2, HB11;19, RS4;11, SEM

So "AML-weighted" is correct at 7 of 11 (64%), and the B-ALL arm is four lines. By partner, only four carry KMT2A::AFF1 (SEMK2, RS4;11, SEM, MV4-11) and one of those, MV4-11, is AML rather than B-ALL. The panel contains three KMT2A::AFF1 B-ALL lines.

That is the number this entry's structural argument needed, and it supports it: a dependency specific to KMT2A::AFF1 infant B-ALL is being asked of three cell lines, none infant-derived. A null DepMap result for such a candidate should indeed be reported as uninformative rather than as a failed cross-validation.

Facts and figures
scriptanalyses/non-menin-dependency-cross-validation/run.py
statusplanned
result summaryNot run. All inputs are public; no access agreement or collaboration is required.
cost tierlow
expected outputA ranked table of non-menin candidate dependencies with a selectivity score, a tolerability score, an evidence-quality grade and a chemical-tractability note, plus an explicit list of claims that fail cross-validation.
inputs
[
  "DepMap CRISPR (Chronos) gene effect matrix and cell-line annotation (KMT2A-r status, lineage)",
  "gnomAD constraint (pLI, LOEUF) per candidate gene",
  "GTEx and Human Protein Atlas normal-tissue expression",
  "ChEMBL bioactivity and mechanism records per candidate target",
  "Candidate list: PSMB8, PTBP1, PKM, ARID2, ARID1B, CEBPA/CEBPB, plus MEN1 (positive control) and CDK9/WDR5/BRD4/PAF1 (common-essential negative controls), extensible as new single-group claims appear"
]
How this could bite us back
  • If every non-menin candidate fails cross-validation, the honest conclusion is that menin may genuinely be the only tractable obligate interface for this fusion, which weakens the lab's differentiating claim.
  • Cell-line essentiality screens systematically miss dependencies that require a microenvironment, a developmental context or a fetal cell state - all of which are central to infant KMT2A-r disease - so a negative result here is weaker than it looks.
  • Publishing a cross-validation that fails other groups' targets creates enemies among exactly the second-ring labs the survey recommends partnering with; framing matters.
Still open
  1. PARTLY ANSWERED: the panel is 11 rearranged lines, 7 AML and 4 B-ALL, with only three KMT2A::AFF1 B-ALL lines and none infant-derived. The remaining question is whether three lines can falsify anything, and what the minimum detectable effect is at that n.
  2. Does any non-menin candidate reproduce menin's profile - not common-essential, sharply depleted in KMT2A-r lines? If none does, the lab's 'other obligate interfaces' thesis has no current candidates and needs new discovery rather than curation.
  3. Should the candidate list be generated systematically from the KMT2A::AFF1 interactome in this graph rather than from whatever happens to have been published?
Every source behind this entry (5)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    PSMB8 was identified proteomically as a specific vulnerability in KMT2A-complex-driven leukaemia, with genetic and pharmacologic inactivation impairing murine and human leukemic proliferation while normal haematopoietic cells reportedly remain unaffected, acting through increased BASP1 which represses KMT2A-fusion target genes.

    PMID 38049829DOI
    What this evidence cannot show

    Single group; the normal-cell sparing claim rests on in-vitro and murine readouts. Immunoproteasome inhibition has systemic immunological consequences a leukaemia-focused study would not detect, which matters in infants. No contradicting study was located, and that is itself the finding: the PSMB8 result is a single-group report that no independent laboratory has attempted to reproduce in the intervening period, and the 'normal haematopoietic cells remain unaffected' clause is the part most likely to fail on replication, because it rests on short-term assays rather than competitive repopulation and because the immunoproteasome is not dispensable for normal immune function. The absence of a published rebuttal here means the claim is untested, not that it has survived testing - which is precisely why it is in the cross-validation list.

    Evidence that argues against this (1)
    • A structural limit on this analysis that the plan should state up front, sourced to the affiliation record of the claim it is meant to cross-validate rather than to any critique of it. The PSMB8 dependency, like most of the candidate list, was reported by a single group. The group's principal investigator is also an author on the registration-enabling phase 2 of ziftomenib in relapsed/refractory NPM1-mutant AML (PMID 40997296, DOI 10.1200/JCO-25-01694), and his current affiliation on that 2025 paper is Hannover Medical School rather than the Greifswald address carried elsewhere in this graph. Neither fact bears on whether PSMB8 is a real dependency. Both bear on what this analysis can conclude: cross-validating single-group claims in public data can show which claims fail, but it cannot show that a claim which survives is correct, because the public data (DepMap, GTEx, gnomAD, ChEMBL) were not generated in the model system where the claim was made. For PSMB8 specifically, a proteomic dependency discovered in KMT2A-complex-driven murine and human leukaemia will not necessarily register in a DepMap panel that is AML-weighted and contains few KMT2A::AFF1 B-ALL lines, so a null DepMap result should be reported as uninformative rather than as a failed cross-validation. The analysis plan should pre-specify that asymmetry - which candidates can be falsified by public data and which cannot - before it is run, or it will generate a ranked table whose low scores mean two different things. Caveats: this is a methodological argument, not a result; the DepMap KMT2A-rearranged line count was not enumerated in this pass and should be before the claim about panel composition is asserted in the entry body. Read: author affiliations of the PubMed record; abstract read, not full text. Wang ES et al. 2025
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    PTBP1, a direct MLL-ENL target, is rate-limiting for proliferation and controls PKM2 splice isoform selection, making MLL-ENL leukaemia selectively sensitive to glucose deprivation, energetic decoupling and serine-biosynthesis inhibition.

    Exploitable metabolic dependencies in MLL-ENL-induced leukemia · Garcia-Cuellar MP, ... Slany RK · Blood Advances 2020
    PMID 32766858DOI
    What this evidence cannot show

    MLL-ENL (KMT2A::MLLT1), not KMT2A::AFF1; partner-specific metabolic wiring is plausible. Metabolic dependencies measured in culture are notoriously sensitive to medium composition and rarely survive transfer to in-vivo nutrient environments.

    Evidence that argues against this (1)
    • This is a different fusion in a different lineage. MLL-ENL and MLL-AF9 in murine or human myeloid systems are the models this dependency was found in, and the index disease here is human infant KMT2A::AFF1 B-lymphoid leukaemia. A dependency that has not been lineage-split in this project's own DepMap framework should be treated as a candidate. Selective sensitivity to glucose deprivation and serine-biosynthesis inhibition is also measured in culture, where nutrient availability is set by the medium; marrow is not a nutrient-limited compartment in the same way. Oncofusion knowledge graph (internal analysis) 2026
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    The SWI/SNF targeting subunits Arid2 and Arid1b have distinct, stage-dependent roles in MLL-AF9 leukaemogenesis: Arid2 loss enhances initiation but Arid2 is required for maintenance, while Arid1b loss promotes leukaemogenesis in both phases.

    PMID 35022500DOI
    What this evidence cannot show

    Opposite effects at initiation versus maintenance mean a continuously dosed drug could be net harmful; conditional mouse genetics cannot resolve what steady pharmacological inhibition would do.

    Evidence that argues against this (1)
    • This is a different fusion in a different lineage. MLL-ENL and MLL-AF9 in murine or human myeloid systems are the models this dependency was found in, and the index disease here is human infant KMT2A::AFF1 B-lymphoid leukaemia. A dependency that has not been lineage-split in this project's own DepMap framework should be treated as a candidate. The Arid2 result is additionally stage-dependent in opposite directions — loss enhances initiation but Arid2 is required for maintenance — which makes it a poor cross-validation target: a drug given to an established leukaemia and a lesion present from the start are testing different halves. Oncofusion knowledge graph (internal analysis) 2026
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Compound deletion of C/EBP family members almost entirely abrogates growth and survival of MLL-transformed cells, with rare escapees characterised by high expression of the resident gene.

    Myeloid transformation by MLL-ENL depends strictly on C/EBP · Wesolowski R, ... Leutz A · Life Science Alliance 2020
    PMID 33144337DOI
    What this evidence cannot show

    C/EBPalpha is required for normal myelopoiesis, so even a real dependency here has a poor prima facie therapeutic window. The indexed abstract has gene symbols stripped and the exact family members must be read from the paper.

    Evidence that argues against this (1)
    • This is a different fusion in a different lineage. MLL-ENL and MLL-AF9 in murine or human myeloid systems are the models this dependency was found in, and the index disease here is human infant KMT2A::AFF1 B-lymphoid leukaemia. A dependency that has not been lineage-split in this project's own DepMap framework should be treated as a candidate. Compound deletion of a transcription-factor family is a strong perturbation with no pharmacological equivalent, and the rare escapees characterised by high expression of the resident gene are the informative part: the requirement is dose-dependent and escapable. Oncofusion knowledge graph (internal analysis) 2026
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    The benchmark the analysis scores against exists: MEN1 is not DepMap common-essential (mean Chronos about -0.36 across 1,258 lines) yet is sharply depleted in KMT2A-rearranged and NPM1-mutant leukaemia lines, whereas CDK9, WDR5, BRD4, PAF1 and others are common-essential with mean Chronos below -0.8.

    DepMap CRISPR (Chronos) gene effect, gnomAD constraint and Human Protein Atlas expression · Broad Institute DepMap; gnomAD; Human Protein Atlas; Open Targets · Databases 2026
    What this evidence cannot show

    These figures are taken from this graph's own gene-men1 node, where they were computed from the Open Targets API rather than a DepMap release table and the release version is unstated. Cell-line essentiality is a poor model of normal-tissue tolerance, and the KMT2A-r line panel in DepMap is small and skewed toward AML rather than infant ALL.

    Evidence that argues against this (1)
    • The candidate list this analysis is built on is a moving target, and one new entrant already has the exact profile the analysis is designed to search for. In January 2026 Nature Communications reported that the RNA-binding protein AKAP95 is required for MLL-rearranged leukemogenesis: its phase separation and RNA binding modulate Pol II recruitment into condensates genome-wide, it interacts with the MLL1 translocated fragment with partial co-condensation strengthening its binding at MLL-AF9 target genes, its loss downregulates MLL-AF9 targets and impairs MLL-AF9-driven leukemogenesis, and - the decisive phrase for this analysis - 'AKAP95 is not essential for normal and regenerative hematopoiesis, rendering it an ideal target for leukemia therapeutics'. A designed peptide that disrupts AKAP95 phase separation inhibits MLLr leukaemia cell proliferation. Non-essential in normal haematopoiesis, required in the MLL-fusion context, with a tool compound: that is the menin-like profile this analysis exists to find, and it was not in the candidate list. Two consequences. (1) The analysis should be built to ingest new claims rather than to score a fixed list, and its status should record the literature cut-off date, because a ranked table drawn on one date is stale on the next. (2) It also supplies a concrete answer to the entry's first open question - whether any non-menin candidate reproduces menin's profile - by nominating one, which is a reason to run the analysis rather than to abandon the thesis. Caveats: the profile claim rests on the paper's own murine and in-vitro readouts, in MLL-AF9 rather than KMT2A::AFF1 and not in infant B-ALL; AKAP95 is a broadly expressed nuclear RNA-binding protein, so 'not essential' needs to be checked against DepMap and gnomAD before it is believed, which is precisely what this analysis would do. Read: abstract only. Yan X et al. 2026
Connected to