Analysismoderateking

Two of the three top new dependencies are myeloid identity, not KMT2A

also called lineage split test · myeloid TF confound

The cell lines used to find genes this leukaemia needs are mostly one blood-cell type. So genes that any cell of that type needs look like they are specific to our disease. Splitting the lines by cell type separates the two, and two of the three new top hits turn out to belong to the cell type.

The same thing, for a scientist

The genome-wide scan contrasts KMT2A-rearranged against other leukaemia lines, but the index panel is MYELOID-ENRICHED (8 AML, 5 B-ALL), so a myeloid-identity gene scores as KMT2A-selective for reasons unrelated to KMT2A. Splitting the index arm by lineage and requiring the dependency in BOTH halves: **CEBPA (-2.79 myeloid, +0.15 B-lineage) and SPI1 (-1.82, +0.92) are myeloid-identity artefacts with NO dependency in KMT2A-rearranged B-ALL.** MYB is weak once split (-1.12, -0.56). What survives is the menin module itself — **MEN1 -2.29/-2.28 with a split gap of 0.01**, PSIP1 -2.78/-1.72, KMT2A -1.48/-1.22 — plus **BRD9 (-1.11/-1.62)**, which is a genuine new lineage-independent dependency.

Go deeper

The split

Standard deviations below the other-leukaemia arm. A real KMT2A dependency should appear in both halves.

gene	KMT2A myeloid	KMT2A B-lineage	OCI-AML3	U-937	verdict
MEN1	-2.29	-2.28	-2.8	0.36	lineage-independent
PSIP1	-2.78	-1.72	-2.63	0.52	lineage-independent
KMT2A	-1.48	-1.22	-2.05	-0.32	lineage-independent
BRD9	-1.11	-1.62	-2.13	-0.01	lineage-independent
MYB	-1.12	-0.56	-2.25	-1.08	weak or ambiguous
CEBPA	-2.79	0.15	-5.38	-2.09	MYELOID-IDENTITY ARTEFACT
SPI1	-1.82	0.92	-2.07	-2.3	MYELOID-IDENTITY ARTEFACT
EP300	-0.45	-0.68	-0.14	0.9	weak or ambiguous
DOT1L	-0.53	-0.88	-1.76	-0.53	weak or ambiguous
CREBBP	0.37	0.26	-0.93	0.19	weak or ambiguous

MEN1's split gap is 0.01. It is the only gene in the panel whose dependency is indistinguishable between KMT2A-rearranged AML and KMT2A-rearranged B-ALL, which is what a dependency on the fusion rather than on the cell should look like. PSIP1/LEDGF and KMT2A itself behave the same way, so the module that survives is precisely the menin-LEDGF-KMT2A interface.

BRD9 is the genuinely new survivor and is not explained by the programme lines either: U-937 sits at -0.01. It came out of interactome round 4 with an on-target swap-allele proof behind it, and it is the one output-layer nomination that holds up.

What this corrects, and what it does not

Interactome round 4 reported an inversion — chromatin complexes rejected as anti-dependencies while myeloid transcription factors topped the ranking at MYB 17, CEBPA 25, SPI1 43 — and flagged, correctly, that none of it had been lineage-conditioned. Two of the three do not survive.

This does NOT say CEBPA and SPI1 are unimportant. A myeloid-identity dependency can be a perfectly good target in myeloid KMT2A-rearranged AML. It says they are not evidence about KMT2A, and that the beat's framing ("the data kept nominating the output layer") describes the panel's lineage composition rather than a finding about the fusion.

It also does not rescue the complexes. Seven chromatin regulators were rejected as anti-dependencies and nothing here revisits that.

Facts and figures

script	analyses/lineage-confound-myeloid-tf-dependencies/run.py
status	executed
run date	2026-09-19
headline	CEBPA and SPI1 are myeloid-identity artefacts with no dependency in KMT2A-rearranged B-ALL; MEN1, PSIP1, KMT2A and BRD9 survive the split.
caveat	Five B-lineage and eight myeloid lines, so each half is small and a single line moves a mean. The split is also by lineage LABEL, not by an expression signature, and KMT2A-rearranged disease includes genuinely ambiguous-lineage cases that a label forces into one bin. This bounds a dependency's lineage SPECIFICITY; it does not test whether the gene matters therapeutically, and a myeloid-identity dependency can still be a good target in myeloid KMT2A-rearranged AML - it is simply not evidence about KMT2A.

z by gene

index split

{
  "myeloid": [
    "ML-2",
    "MOLM-13",
    "MOLM-14",
    "NOMO-1",
    "OCI-AML-2",
    "SHI-1",
    "THP-1"
  ],
  "b_lineage": [
    "HB11;19",
    "KOPN-8",
    "RS4;11",
    "SEM",
    "SEMK2"
  ]
}

z by gene — the numbers behind the chart

{
  "MYB": {
    "kmt2a_myeloid": -1.12,
    "kmt2a_b_lineage": -0.56,
    "split_gap": -0.57,
    "verdict": "weak or ambiguous"
  },
  "CEBPA": {
    "kmt2a_myeloid": -2.79,
    "kmt2a_b_lineage": 0.15,
    "split_gap": -2.94,
    "verdict": "MYELOID-IDENTITY ARTEFACT"
  },
  "SPI1": {
    "kmt2a_myeloid": -1.82,
    "kmt2a_b_lineage": 0.92,
    "split_gap": -2.74,
    "verdict": "MYELOID-IDENTITY ARTEFACT"
  },
  "BRD9": {
    "kmt2a_myeloid": -1.11,
    "kmt2a_b_lineage": -1.62,
    "split_gap": 0.51,
    "verdict": "lineage-independent"
  },
  "EP300": {
    "kmt2a_myeloid": -0.45,
    "kmt2a_b_lineage": -0.68,
    "split_gap": 0.23,
    "verdict": "weak or ambiguous"
  },
  "CREBBP": {
    "kmt2a_myeloid": 0.37,
    "kmt2a_b_lineage": 0.26,
    "split_gap": 0.11,
    "verdict": "weak or ambiguous"
  },
  "MEN1": {
    "kmt2a_myeloid": -2.29,
    "kmt2a_b_lineage": -2.28,
    "split_gap": -0.01,
    "verdict": "lineage-independent"
  },
  "DOT1L": {
    "kmt2a_myeloid": -0.53,
    "kmt2a_b_lineage": -0.88,
    "split_gap": 0.35,
    "verdict": "weak or ambiguous"
  },
  "PSIP1": {
    "kmt2a_myeloid": -2.78,
    "kmt2a_b_lineage": -1.72,
    "split_gap": -1.06,
    "verdict": "lineage-independent"
  },
  "KMT2A": {
    "kmt2a_myeloid": -1.48,
    "kmt2a_b_lineage": -1.22,
    "split_gap": -0.26,
    "verdict": "lineage-independent"
  }
}

lineage artefacts

[
  "CEBPA",
  "SPI1"
]

survivors

[
  "BRD9",
  "MEN1",
  "PSIP1",
  "KMT2A"
]

How this could bite us back

This analysis was run by the coordinator, on a panel the coordinator chose, after reading a result the coordinator wanted to check - and it concluded that the coordinator's preferred hypothesis is the one that survives. That is the shape of a confirmatory analysis and should be weighted accordingly.
Labelling CEBPA and SPI1 as artefacts could remove two genuine myeloid targets from consideration. The claim is narrow: their genome-wide RANK is not evidence about KMT2A. It is not a claim that they do not matter in KMT2A-rearranged AML.

Still open

Should the genome-wide selective scan be re-run with the index arm lineage-balanced, or reported as two scans (KMT2A-r AML and KMT2A-r B-ALL) rather than one? Every ranking this project has published from that scan carries this confound.
BRD9 survives the split and is not explained by the programme lines. Is its dependency reproducible under permutation and line-normalisation, which this repo has scripts for and nobody has run on it?
MEN1's split gap is 0.01 across two lineages. Does any other gene in the genome match that, and would a genome-wide search for lineage-independent KMT2A dependencies find anything the hand-picked panel missed?

Every source behind this entry (1)

E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.analysis
Splitting the KMT2A-rearranged DepMap panel by lineage (8 myeloid, 5 B-lineage) and requiring a dependency in both halves separates KMT2A-specific dependencies from myeloid-identity ones. CEBPA scores -2.79 SD in KMT2A-rearranged myeloid lines and +0.15 in KMT2A-rearranged B-lineage lines; SPI1 scores -1.82 and +0.92. Neither has any dependency in KMT2A-rearranged B-ALL, so their genome-wide rank reflects the index panel being myeloid-enriched. MEN1 scores -2.29 and -2.28, a split gap of 0.01; PSIP1 -2.78 and -1.72; KMT2A -1.48 and -1.22; BRD9 -1.11 and -1.62. The menin-LEDGF-KMT2A module and BRD9 survive the split; CEBPA and SPI1 do not, and MYB, EP300, CREBBP and DOT1L are weak or ambiguous once split.
Lineage-split test of myeloid transcription-factor dependencies in KMT2A-rearranged leukaemia · Oncofusion knowledge graph (internal analysis); DepMap Chronos via Open Targets 2026
What this evidence cannot show
Run by the coordinator on 2026-09-19 after interactome round 4 flagged that its DepMap rankings had not been lineage-conditioned. Five B-lineage and eight myeloid lines, so each half is small and one line moves a mean. The split is by lineage LABEL, not by an expression signature, and KMT2A-rearranged disease includes genuinely ambiguous-lineage cases that a label forces into one bin - this project already holds an entry on KMT2A-rearranged MPAL for that reason. Chronos measures knockout, not interface blockade. The test bounds lineage SPECIFICITY and says nothing about therapeutic value.
Evidence that argues against this (3)
- The B-lineage half is five lines and two of them, SEM and RS4;11, are the most-used KMT2A::AFF1 models in the field, which means they are also the most adapted to culture. A dependency absent in five heavily passaged B-ALL lines is weaker evidence of true lineage restriction than the clean numbers suggest, and CEBPA's +0.15 rests on those five. Oncofusion knowledge graph (internal analysis); DepMap Chronos via Open Targets 2026
- The result is convenient in the same direction as the coordinator's other analysis this cycle: both conclude that the menin axis is the specific thing and the alternatives are confounded. The panel and the lineage assignment were both chosen by the coordinator after seeing the beat's headline, which is the order of operations most likely to produce a confirmatory split. Oncofusion knowledge graph (internal analysis); DepMap Chronos via Open Targets 2026
- Calling CEBPA and SPI1 artefacts risks discarding real biology. KMT2A-rearranged AML is a substantial fraction of the disease, SPI1 mutations with transcriptional-network inactivation are reported in a subset of MLL-fusion AML, and a target that works only in the myeloid half is still a target for half the patients. The finding is about what the genome-wide RANK means, not about whether these genes matter. Oncofusion knowledge graph (internal analysis); DepMap Chronos via Open Targets 2026

Connected to

← related toETV6::RUNX1 (TEL-AML1), t(12;21)(p13;q22)
Counter-evidence: Jeon 2023 (PMID 36045062) reports FISH- and RT-PCR-confirmed ETV6::RUNX1 in an adult AML-M5a, so B-lineage restriction is near-complete rather than absolute
← related toNPM1 (nucleophosmin) — the nucleolar shuttle whose mutant form binds chromatin with KMT2A and buys the second menin-inhibitor indication
NPM1 itself shows no KMT2A selectivity on a lineage-split DepMap run; the dependency is conferred by the mutant, not the gene.
← related toLineage-split dependency for the round 5 panel: 23 genes, 2 myeloid-identity artefacts caught, and 3 survivors of which one has no literature at all
Applies the myeloid-confound method established there to this beat's 23-gene panel.
refutes →Interactome round 4: what was counted, what was written, what was rejected, and the two alias traps this beat fell into and caught
Two of the three myeloid transcription factors this enumeration nominated as top dependencies do not survive a lineage split of the index panel.
related to →MYB — rank 17 of 17,728 for KMT2A selectivity, and the only node in this beat where partial, transient suppression eradicated leukaemia without stopping normal blood production
MYB's KMT2A-selectivity weakens substantially once the index arm is split by lineage (-1.12 myeloid, -0.56 B-lineage).
supports →BRD9 — dispensable in the average cell, rank 39 of 17,728 in KMT2A-rearranged lines, and with an on-target proof most epigenetic targets never get
BRD9 survives the lineage split (-1.11 myeloid, -1.62 B-lineage) and is not explained by the pathway-equivalent lines.
related to →EP300 — the acetyltransferase whose paralogue split is the finding: p300 is the 108th most KMT2A-selective gene, CBP is the 16,950th
EP300's dependency is weak in both halves once split, which bounds the EP300-versus-CREBBP paralogue claim.
supports →MEN1 / menin — the adaptor the fusion cannot work without
MEN1 is the only panel gene with an essentially zero lineage-split gap, which is what a dependency on the fusion rather than the cell should look like.
supports →PSIP1 / LEDGF p75 — the tether with the best therapeutic index in the whole interactome
LEDGF/PSIP1 survives the lineage split alongside MEN1 and KMT2A.
related to →Running the HOXA programme is not enough to need menin
Companion test: that one separates programme from rearrangement, this one separates lineage from rearrangement.
refutes →The whole protein-coding genome: menin is still first, and a metabolic pathway appears (run)
The scan's index arm is myeloid-enriched, so a myeloid-identity gene ranks as KMT2A-selective; CEBPA and SPI1 are shown to do exactly that.
← supportsBRD9 survives the null; CEBPA sits at the median of random genes
Discharges the blocker that entry recorded: BRD9 survives line-normalisation and a random-gene null at the 0.62nd percentile.
← related toSPI1 / PU.1 — a myeloid dependency that a KMT2A panel mistakes for its own
SPI1 is one of the two genes shown to be a myeloid-identity artefact of the index panel.
← related toCEBPA — ranks 25th on the KMT2A scan and sits at the median of random genes
CEBPA is one of the two genes shown to be a myeloid-identity artefact of the index panel.
← supportsExactly one gene in 8,248 beats menin, and it is FLT3
CEBPA (47.7th percentile) and SPI1 (96.2nd) remain rejected at genome scale.
← related toNothing beats menin after all, and the statistic that said otherwise ranks narrowness
← related toOne paper holds up 70 entries, and one dataset holds up 167
← related toThe deferred KMT2A partners: what the counts actually are, what was written, what was rejected, and three numbers the previous enumeration got wrong
applies that analysis's lineage split to all six partners in this beat before any dependency number is quoted
← related toLineage-split DepMap: neither USP2 nor MAML2 is a dependency in KMT2A-rearranged lines, and USP22 is
applies the same myeloid-enrichment correction to this beat's genes; no artefacts found in this panel