4,589 genes with no prior, and menin comes first (run)
also called unbiased KMT2A-selectivity scan · discovery scan · MEN1 rank 1
**Extended, not superseded:** `ana-depmap-kmt2a-selective-scan-full` re-ran this over the whole protein-coding genome rather than the 4,589 genes here, and **menin still came first** — so this scan's headline survives the wider search. The -full run additionally surfaced a haem-biosynthesis cluster that this one could not see. Until now every dependency analysis here started from a shortlist taken from the literature, which can confirm but never discover. This one scanned nearly five thousand genes with no shortlist and asked which are most specifically needed by leukemia cells carrying the fusion. Menin came first.
The same thing, for a scientist
Unbiased scan of 4,589 protein-coding genes ranked by difference in mean DepMap gene effect between 11 Cellosaurus-annotated KMT2A-rearranged leukemia lines and 43 other leukemia lines. MEN1 ranks **1 of 4,589** (difference -0.866, empirical p 0.00044) and is not pan-essential (mean effect across all lines -0.363). FLT3 ranks 2, PSIP1 6, MEF2C 7, KMT2A 8. The scan independently rediscovered ZFP64 at rank 12, a published MLL-rearranged-specific dependency that was not on the watchlist. The seven genes struck in the cycle-23 target audit rank between the 67th and 98th percentiles.
Why a scan and not another candidate test
Every dependency analysis in this project has been confirmatory. A candidate list was taken from the literature — menin, LEDGF, DOT1L, the WRAD subunits, PAF1C — and each member was scored. That design cannot discover anything, and it cannot address the threat the cycle-23 target audit left standing:
If the interactome is systematically undruggable — pan-essential, paralogue-buffered, or a germline tumour suppressor — menin isn't the best handle, it's the only one, and the project rests on one interface with a known resistance mechanism.
That is a claim about genes nobody has looked at, so a candidate list cannot test it.
MEN1 was placed in the scan unflagged, competing with 4,588 others on its own merits. If the menin dependency is as central as this project assumes, MEN1 should rank near the top without help. If it did not, that would be the most important negative result the project could produce, and the design was chosen so that outcome would be unmissable 1.
The result: MEN1 ranks first3 deeper
More negative means more required in rearranged lines. Pan-essentiality is the mean effect across all ~1,200 lines in the scan and is shown for every hit, because the easiest way to be fooled is to rediscover genes every cell needs.
| Rank | Gene | difference | KMT2A-r | other | pan-essential? | note |
|---|---|---|---|---|---|---|
| 1 | MEN1 | -0.866 | -1.263 | -0.397 | no (-0.36) | the project's premise |
| 2 | FLT3 | -0.843 | -0.962 | -0.119 | no (-0.15) | partly FLT3-ITD lines |
| 3 | DNAJC9 | -0.645 | -2.333 | -1.688 | yes (-1.39) | discard |
| 4 | CCND3 | -0.623 | -1.527 | -0.904 | no (-0.35) | not on any list here |
| 5 | TINF2 | -0.575 | -2.206 | -1.631 | yes (-1.51) | discard |
| 6 | PSIP1 (LEDGF) | -0.565 | -0.553 | +0.012 | no (+0.03) | the tether |
| 7 | MEF2C | -0.560 | -0.843 | -0.283 | no (-0.04) | |
| 8 | KMT2A | -0.541 | -0.811 | -0.270 | no (-0.21) | |
| 9 | MYB | -0.475 | -1.925 | -1.451 | no (-0.38) | see below |
| 11 | ZEB2 | -0.464 | -1.205 | -0.740 | no (-0.36) | not on any list here |
| 12 | ZFP64 | -0.410 | -0.392 | +0.018 | no (-0.005) | rediscovered, see below |
| 15 | UROS | -0.387 | -0.823 | -0.436 | no (-0.20) | not on any list here |
| 19 | GFI1 | -0.312 | -0.558 | -0.247 | no (-0.02) | not on any list here |
Ranks 3, 5, 10, 13, 16, 17, 18 and 20 are DNAJC9, TINF2, RPL13A, LSM2, RPS6, RPL9, RPS24 and RRM2 — ribosomal and core-machinery genes with pan-essentiality between -1.4 and -2.6. They are more depleted in rearranged lines because those lines are more sensitive overall, not because of the fusion. The pan-essentiality column exists to catch exactly this, and it did.
The scan rediscovered a known MLL-specific dependency it was not told about
ZFP64 ranks 12 with a pan-essentiality of -0.005 — a gene almost no cell line needs, yet selectively required in the rearranged lines. It was not on the watchlist and appears in this graph only as a stub.
It should have been. Lu et al. identified ZFP64 by domain-focused CRISPR screening as an essential transcription factor in MLL-rearranged leukemia, whose critical function is to maintain MLL expression by binding the MLL promoter — the most enriched location of ZFP64 occupancy in the human genome, explained by an exceptional density of ZFP64 motifs there 2.
This is the strongest available check that the scan works. It recovered, from dependency data alone and with no prior, a published MLL-rearranged-specific dependency with a clean mechanism. That the same scan puts MEN1 first is therefore harder to dismiss as an artefact of the design.
The struck genes are at the bottom, which is independent support for the audit
The cycle-23 target audit struck seven candidates on reasoning that, on re-checking, was wrong or materially incomplete in four cases. This scan had no knowledge of that audit.
| Gene | rank of 4,589 | percentile | difference |
|---|---|---|---|
| CDC73 | 3,064 | 66.8 | +0.020 |
| KDM4C | 3,782 | 82.4 | +0.047 |
| RBBP5 | 3,818 | 83.2 | +0.049 |
| MLLT3 | 3,883 | 84.6 | +0.052 |
| ASH2L | 4,460 | 97.2 | +0.125 |
| KDM6A | 4,497 | 98.0 | +0.143 |
| CTR9 | 4,500 | 98.1 | +0.146 |
All seven have positive differences — marginally less required in rearranged lines than in other leukemia. The audit reached the right list by reasoning that was partly wrong; an unbiased scan reaches the same list directly. DPY30 ranks 674 (14.7th percentile), unremarkable for the fifth time across five independent designs.
Three results that cut against the graph's current position
HOXA9 ranks 961, the 21st percentile. The canonical downstream effector of KMT2A fusions is not selectively required in rearranged lines in this data. The likely explanation is paralogue buffering — HOXA10 and the rest of the cluster — which is exactly the mechanism hyp-paralogue-buffering-hides-dependencies predicts, so this is a test that hypothesis passes. But anyone citing HOXA9 dependency from DepMap should not.
MYB ranks 9. This project's lineage-conditioned analysis previously overturned a MYB co-dependency claim. Here MYB is among the most selectively required genes in rearranged lines. The two are not formally contradictory — one is correlation with MEN1 across lines, the other is a group difference — but the graph should stop treating MYB as settled in either direction.
DOT1L ranks 47, the top 1%. That is more favourable than the group-permutation reading in ana-depmap-kmt2a-rearranged-specificity, which put its p at 0.177. Rank among 4,589 genes and a two-group permutation on 11 versus 43 lines are answering different questions at very different power, and DOT1L is the gene where they diverge most.
What this is not
Not genome-wide. 4,589 of roughly 19,000 protein-coding genes, chosen at random after seeding the watchlist. A stronger hit may sit in the unscanned 75%.
Not independent of the earlier specificity analysis in method. It is the same difference statistic on the same 11 versus 43 lines. What is new is the absence of a prior and the ranking context: knowing MEN1 beats 4,588 other genes is a different fact from knowing it beats 918 random ones.
Eleven lines, none infant-derived. Unchanged and unfixable with this data.
Selective dependency is not druggability. CCND3, ZEB2, GFI1 and UROS are transcription factors, a cyclin and a biosynthetic enzyme. Ranking highly here says nothing about whether any of them has a ligandable pocket, and three of the four have no entry in this graph at all.
| script | analyses/depmap-kmt2a-selective-scan/run.py |
|---|---|
| status | run |
| run date | 2026-09-18 |
| seed | 20260918 |
| n scanned | 4589 |
| n kmt2a rearranged | 11 |
| n other leukemia | 43 |
| men1 rank | 1 |
| men1 percentile | 0.022 |
| men1 empirical p | 0.000436 |
| men1 pan essentiality | -0.3631 |
| figure | analyses/depmap-kmt2a-selective-scan/results.md |
top selective non pan essential
[ "MEN1", "FLT3", "CCND3", "PSIP1", "MEF2C", "KMT2A", "MYB", "ZEB2", "ZFP64", "UROS", "GFI1" ]
pan essential contaminants in top 20
[ "DNAJC9", "TINF2", "RPL13A", "LSM2", "RPS6", "RPL9", "RPS24", "RRM2" ]
struck gene ranks — the numbers behind the chart
{
"CDC73": 3064,
"KDM4C": 3782,
"RBBP5": 3818,
"MLLT3": 3883,
"ASH2L": 4460,
"KDM6A": 4497,
"CTR9": 4500
}notable low ranks — the numbers behind the chart
{
"DPY30": 674,
"HOXA9": 961,
"ELL2": 975
}novel hits with no entry
[ "CCND3", "ZEB2", "GFI1", "UROS" ]
- MEN1 ranking first is the single most favourable result this project has produced, which is exactly why it should be held to the strictest reading. It is 11 cell lines, none infant-derived, on 24% of the protein-coding genome, using the same statistic as the analysis it is meant to corroborate.
- Eight of the top twenty are ribosomal genes. Strip the pan-essentiality column and this becomes a list that would waste a year. Any ranking of this kind without that column should be distrusted, including future ones from this project.
- A selective dependency is not a drug target. Three of the four novel hits are transcription factors, the protein class with the worst record of direct ligandability.
- The scan agreeing with the cycle-23 strike list is reassuring but not independent confirmation of the audit's reasoning — it is a second route to the same list, and the audit's reasoning was separately shown to be wrong in four of ten cases.
- CCND3, ZEB2, GFI1 and UROS rank in the top 20 as selective, non-pan-essential dependencies and have no entry in this graph. Are any of them documented KMT2A-fusion biology, and which is the most tractable?
- ZFP64 acts by maintaining MLL transcription, but this graph also holds that MLL fusions are abnormally stable and degradation-resistant. Does lowering fusion transcription actually lower fusion protein on a useful timescale?
- Scanning the remaining ~14,000 protein-coding genes would cost roughly four hours of API time. Is anything above MEN1 hiding there?
- HOXA9 at rank 961 implies DepMap cannot see paralogue-buffered dependencies. Which other canonical KMT2A-fusion effectors are paralogue-buffered, and does that explain other low ranks here?
- MYB ranks 9 here and was overturned by this project's lineage-conditioned analysis. Which is right, and does the difference between a correlation statistic and a group-difference statistic explain it?
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
In an unbiased scan of 4,589 protein-coding genes ranked by difference in mean DepMap gene effect between 11 KMT2A-rearranged and 43 other leukemia lines, MEN1 ranks 1 (difference -0.866, empirical p 0.00044 against the scan's own distribution) with a pan-essentiality of -0.363 across all lines. FLT3 ranks 2, CCND3 4, PSIP1 6, MEF2C 7, KMT2A 8, MYB 9, ZEB2 11, ZFP64 12, GFI1 19, DOT1L 47, DPY30 674 and HOXA9 961. Seven genes struck in the cycle-23 target audit rank between 3,064 and 4,500, all with positive differences. Ranks 3, 5, 10, 13, 16, 17, 18 and 20 are ribosomal or core-machinery genes with pan-essentiality between -1.4 and -2.6.
DepMap CRISPR (Chronos) gene effect via the Open Targets Platform GraphQL API, scanned across 4,589 HGNC protein-coding genes and split by Cellosaurus KMT2A gene-fusion annotation cross-referenced to DepMap identifiers; analysis in analyses/depmap-kmt2a-selective-scan/ · Broad Institute DepMap; Cellosaurus (SIB); HGNC; Open Targets Platform; analysis run by this project · Databases and derived analysis 2026What this evidence cannot show
Our own computation, not peer reviewed. 4,589 of roughly 19,000 protein-coding genes, so not genome-wide. Same difference statistic and same 11 versus 43 lines as ana-depmap-kmt2a-rearranged-specificity, so this is not methodologically independent of it; what is new is the absence of a prior and the ranking context. Empirical p is the fraction of scanned genes with an equally extreme difference, which is a background rather than a formal test. Eleven lines, none infant-derived. Cellosaurus fusion annotation may be incomplete, which biases toward the null. DepMap release version not stated by the API.
Evidence that argues against this (2)
- Eight of the top twenty hits are ribosomal or core-machinery genes that are simply more depleted in a more sensitive panel. Without the pan-essentiality column this scan would have produced a plausible-looking list of false leads, and any ranking of this kind published without that column should be distrusted. It is also a reminder that MEN1's own rank is a rank among genes subject to the same panel effect. analysis run by this project 2026
- HOXA9, the canonical downstream effector of KMT2A fusions, ranks 961 of 4,589 — not selectively required. If this scan is trusted for MEN1 at rank 1, it must also be trusted here, and the straightforward reading is that DepMap single-gene knockouts cannot see a dependency buffered by paralogues. That limitation applies to every gene in the scan with close paralogues, which is not a small set. analysis run by this project 2026
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Domain-focused CRISPR screening identified ZFP64 as an essential transcription factor in MLL-rearranged leukemia, whose critical function is to maintain MLL expression by binding the MLL promoter — the most enriched location of ZFP64 occupancy in the human genome, accounted for by an exceptional density of ZFP64 motifs embedded in that promoter.
A Transcription Factor Addiction in Leukemia Imposed by the MLL Promoter Sequence · Lu B, Klingbeil O, Tarumoto Y, Somerville TDD, Huang YH, Wei Y, et al. (Vakoc CR, senior author) · Cancer Cell 34:970-981.e8 2018What this evidence cannot show
Read as abstract only. The mechanism is transcriptional maintenance of MLL expression, so ZFP64 acts upstream of the fusion rather than as a component of it — a different kind of target from a partner interface, and one whose therapeutic logic is lowering fusion levels rather than disrupting a complex. Note that a separate finding in this graph holds that MLL fusions are abnormally stable and resistant to degradation, which makes the consequences of lowering their transcription an open rather than obvious question.
Evidence that argues against this (1)
- ZFP64 is a C2H2 zinc-finger transcription factor, a protein class with a poor record of direct small-molecule ligandability, so a genuine and selective dependency here may still not be actionable. ZFP64 also regulates fetal haemoglobin: knocking it down raises HBG expression in HUDEP-2 cells, which indicates functions in normal erythroid biology that an inhibitor would also disturb. Huang S et al. 2025
Tests and refutes the pan-essentiality objection to this scan's ranking: essentiality explains r^2=0.004 of the difference statistic, with a positive slope.
- ← supportsThe whole protein-coding genome: menin is still first, and a metabolic pathway appears (run)
answers that entry's open question: across all 17,728 protein-coding genes nothing outranks MEN1
- supports →The assumption the whole project rests on, tested at last: menin dependency is fusion-specific (run)
same statistic without a prior: MEN1 ranks 1 of 4,589 rather than beating 918 random genes
most selectively required gene in KMT2A-rearranged lines out of 4,589 scanned, and not pan-essential
rediscovered at rank 12 with no prior; ZFP64 maintains MLL expression by binding the MLL promoter
all seven struck genes rank between the 67th and 98th percentile with positive differences, reached independently of the audit's reasoning
HOXA9 ranks 961 of 4,589: the canonical effector is invisible to single-gene knockout, as that hypothesis predicts
- refutes →DPY30 — the smallest COMPASS subunit, and the only one that behaves like a selective dependency
unremarkable at the 14.7th percentile, the fifth independent design to say so
tension to resolve: MYB ranks 9 here but was overturned as a MEN1 co-dependency there
the interface the programme targets sits on the single most fusion-selective dependency found without a prior
tension to resolve: USP22 is a reported KMT2A-r vulnerability that the unbiased scan did not surface
rank 4 of 4,589; selective and not pan-essential (difference -0.623, pan-essentiality -0.348)
the scan hit that led to this target being looked up at all
rank 11 of 4,589; selective and not pan-essential (difference -0.464, pan-essentiality -0.358)
rank 19 of 4,589 — the weakest of the four novel hits (difference -0.312, pan-essentiality -0.021)
UROS at rank 15 has no leukemia literature at all; one clean miss in four novel hits, recorded as calibration for the scan