Every DOT1L dependency number this project has computed, in one place, lineage-split and in raw Chronos units
also called DOT1L DepMap synthesis · DOT1L raw effect vs z · DOT1L lineage split
This project has measured how much cancer cell lines need DOT1L about a dozen times, in different runs, and the answers look different depending on which statistic you read. Put side by side and split by whether the leukaemia is myeloid or lymphoid, the picture is a real but moderate dependency that the available cell lines are too few to prove.
The same thing, for a scientist
No new computation: a join of persisted results across nine of this project's own DepMap runs, reported the way its own rules require - **lineage-split, in raw Chronos units, with the z beside the effect**. The numbers. DOT1L is **not common-essential**: pan-cancer Chronos mean **-0.148**, **12.1%** of 1,208 lines depleted. In the KMT2A-rearranged arm the mean is **-0.604** (n=11) against **-0.373** in other leukaemia (n=43), a raw difference of **-0.231**. **The lineage split runs the opposite way to this project's known artefacts**: the lymphoid arm (n=4) gives **-0.310** and the myeloid arm (n=7) gives **-0.186**, so DOT1L is *not* a myeloid-lineage artefact of the kind that produced the spurious CEBPA and SPI1 hits. On the graph's own re-ranking, DOT1L is **61st of 8,269 on raw effect and 335th on the z**, gaining 274 places when the comparator-SD denominator is removed. And the honest limits: the permutation p against random genes is **0.088** (AML-only **0.142**), the lineage-consistency scan's verdict is **'weak or ambiguous'**, and a power calculation on the observed Cohen's d of **-0.576** says **47 lines per group** are needed for 80% power against the **11** that exist. So DOT1L is a graded, lineage-consistent, moderately sized dependency that DepMap is structurally unable to establish at conventional significance.
The table the project should have had
| quantity | value | source run |
|---|---|---|
| pan-cancer Chronos mean | -0.148 | rna-layer-kmt2a-selectivity |
| % of 1,208 lines depleted | 12.1% | rna-layer-kmt2a-selectivity |
| KMT2A-r arm mean (n=11) | -0.604 | depmap-kmt2a-rearranged-specificity |
| other-leukaemia mean (n=43) | -0.373 | depmap-kmt2a-rearranged-specificity |
| raw difference (combined) | -0.231 | depmap-kmt2a-rearranged-specificity |
| raw difference, lymphoid arm (n=4) | -0.310 | rna-layer-kmt2a-selectivity |
| raw difference, myeloid arm (n=7) | -0.186 | rna-layer-kmt2a-selectivity |
| z, lymphoid / myeloid | -0.773 / -0.464 | rna-layer-kmt2a-selectivity |
| rank on raw effect (d) | 61 of 8,269 | effect-size-vs-z-reranking |
| rank on the z | 335 of 8,269 | effect-size-vs-z-reranking |
| comparator-arm SD | 0.401 | effect-size-vs-z-reranking |
| permutation p (all leukaemia) | 0.088 | depmap-kmt2a-rearranged-specificity |
| permutation p (AML only) | 0.142 | depmap-kmt2a-rearranged-specificity |
| empirical p vs random genes | 0.013 | depmap-kmt2a-rearranged-specificity |
| Cohen's d | -0.576 | depmap-nup98-rearranged-men1-power |
| n per group for 80% power | 47.4 (have 11) | depmap-nup98-rearranged-men1-power |
| co-dependency with MEN1, leukaemia-only r | 0.681 | depmap-lineage-conditioned |
| co-dependency, within-tissue r | 0.544 | depmap-lineage-conditioned |
Every one of these is a persisted number from a run this project already executed; this entry adds no fetch and no model 1.
The three readings that follow, in order of how much they change
1. The lineage split exonerates DOT1L from the artefact that caught CEBPA and SPI1. The KMT2A-rearranged arm in DepMap is myeloid-enriched, so a gene that is simply a myeloid dependency scores well by accident; CEBPA and SPI1 were shown to move in the wrong direction in the B-lineage half 2. DOT1L moves the right way in both halves and more strongly in the lymphoid half (-0.310 against -0.186). For a project whose disease of interest is infant B-ALL, that is the useful direction and it has not been stated anywhere in the graph.
2. The z was hiding a real effect, exactly as the reranking predicted. DOT1L sits 61st of 8,269 on raw Chronos difference and 335th on the scaled statistic, a loss of 274 places caused entirely by a comparator SD of 0.401 3. In the depleted tail the z ranks narrowness at Spearman -0.92 against comparator SD, so DOT1L's penalty is the signature of a gene that matters somewhat in many leukaemias rather than one that matters nowhere else. That is a different and more interesting kind of dependency than the z presents, and it is consistent with the two-regulatory-unit model of H3K79 methylation: an endogenous DOT1L unit operating in most leukaemias plus a fusion-directed one on top.
3. DepMap cannot settle this and should stop being asked to. With 11 KMT2A-rearranged lines and an observed Cohen's d of -0.576, the panel is short by a factor of four of the 47 per group needed for 80% power 4. A permutation p of 0.088 is therefore the expected output of an underpowered test of a real moderate effect, not evidence of absence. This is the same structural limit the project hit for NUP98 rearrangement, where the available line count is zero.
What this says about the target, stated carefully
A graded dependency with a favourable pan-essentiality profile (12.1% of lines depleted, mean -0.148) is exactly the shape of target for which partial, tolerable inhibition should have produced partial benefit - and pinometostat's clean safety with an unreached maximum tolerated dose is the profile of an agent whose exposure was never taken to where that partial benefit would appear.
The contrary reading is equally available from the same numbers: a difference of -0.231 Chronos between KMT2A-rearranged and other leukaemia lines is a moderate selectivity, roughly a quarter of what MEN1 shows, and a moderately selective dependency is a moderately good drug target. This entry does not choose; it puts the numbers where the argument can be had on them.
| status | run |
|---|---|
| run date | 2026-09-19 |
| computation | none - join of persisted results |
| gene | DOT1L |
| raw difference combined | -0.231 |
| verdict from lineage scan | weak or ambiguous; survives_lineage_split false; lineage_artefact false |
| result summary | A real, lineage-consistent, moderate dependency that is stronger in the lymphoid arm than the myeloid one, ranked 61 of 8,269 on raw effect against 335 on the z, and structurally unprovable in DepMap at the available n. |
inputs
[ "analyses/rna-layer-kmt2a-selectivity/results.json", "analyses/depmap-kmt2a-rearranged-specificity/results.json", "analyses/depmap-interactome5-lineage-split/results.json", "analyses/effect-size-vs-z-reranking/results.json", "analyses/depmap-nup98-rearranged-men1-power/results.json", "analyses/depmap-lineage-conditioned/results.json", "analyses/countermeasure-target-ranking/results.json", "analyses/age-restricted-vulnerability-panel/results.json" ]
pan cancer
{
"chronos_mean": -0.148,
"pct_lines_depleted": 12.1,
"n_lines": 1208,
"common_essential": false
}kmt2a rearranged
{
"mean": -0.6042,
"n": 11
}other leukaemia — the numbers behind the chart
{
"mean": -0.3732,
"n": 43,
"sd": 0.4007
}lineage split
{
"lymphoid": {
"n": 4,
"index_mean": -0.683,
"difference": -0.3098,
"z": -0.773
},
"myeloid": {
"n": 7,
"index_mean": -0.5592,
"difference": -0.186,
"z": -0.464
}
}lineage consistent statistic
{
"d": -0.178,
"z": -0.444,
"definition": "max(myeloid, lymphoid) i.e. the weaker arm"
}genome ranks — the numbers behind the chart
{
"raw_effect_d": 61,
"scaled_z": 335,
"n_scored": 8269,
"places_gained_by_removing_denominator": 274
}significance — the numbers behind the chart
{
"perm_p_all_leukaemia": 0.08829,
"perm_p_aml_only": 0.14189,
"empirical_p_vs_random_genes": 0.01306,
"holm_p": 0.0392
}power — the numbers behind the chart
{
"cohens_d": -0.576,
"n_per_group_for_80pct": 47.4,
"n_available": 11,
"shortfall_fold": 4.3
}men1 codependency — the numbers behind the chart
{
"pooled_r": 0.5703,
"within_tissue_r": 0.5439,
"leukaemia_only_r": 0.6808,
"fraction_retained": 0.954
}- A dependency that is moderate everywhere rather than absolute in one genotype predicts a narrow therapeutic index, because the comparator cells that would be spared are themselves partially dependent.
- Post-hoc power calculations are a standard route to talking oneself past a non-significant result; this entry reports p 0.088 in the summary line specifically so that the power argument cannot be quoted without it.
- Reporting a favourable pan-essentiality profile as a safety signal repeats the error this project has already written down about gnomAD and GTEx: 12.1% of lines depleted is a statement about cancer cell lines in culture, not about a patient's marrow.
- The lymphoid arm shows the larger difference (-0.310 from n=4) but no run reports an interval on it. Is the lymphoid-versus-myeloid difference real, or is DOT1L simply lineage-indifferent within noise? For a project focused on infant B-ALL this is the single most valuable number to firm up.
- DepMap is short by a factor of four for this effect size. Is there any orthogonal dependency dataset - a primary-sample CRISPR screen, a PDX panel - with enough KMT2A-rearranged lymphoid material to close the gap?
- If DOT1L's z penalty comes from mattering somewhat in many leukaemias, is the 'somewhat' the endogenous MLLT10-directed DOT1L unit? A per-line association between DOT1L dependency and MLLT10 expression would test that directly.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
Across this project's persisted DepMap runs, DOT1L has a pan-cancer Chronos mean of -0.148 with 12.1% of 1,208 lines depleted and is not common-essential; the KMT2A-rearranged arm mean is -0.604 (n=11) against -0.373 in other leukaemia (n=43), a raw difference of -0.231, which splits into -0.310 in the lymphoid arm (n=4) and -0.186 in the myeloid arm (n=7), with z values of -0.773 and -0.464 against a comparator-arm SD of 0.401.
DepMap CRISPR (Chronos) gene effect retrieved through the Open Targets Platform GraphQL API, as persisted across the Oncofusion knowledge graph's DepMap analyses · DepMap / Open Targets Platform 2026What this evidence cannot show
No new computation: a join of persisted results.json files from analyses/rna-layer-kmt2a-selectivity, depmap-kmt2a-rearranged-specificity, depmap-interactome5-lineage-split, effect-size-vs-z-reranking, depmap-nup98-rearranged-men1-power, depmap-lineage-conditioned, countermeasure-target-ranking, depmap-line-normalised-specificity and age-restricted-vulnerability-panel. The DepMap release version is not stated in the Open Targets API response and is therefore unverified, a caveat this project's own DepMap entries carry. The runs differ slightly in arm definition - the KMT2A-rearranged arm is variously 10 or 11 lines and the myeloid half 6 or 7 depending on which run resolved Cellosaurus annotations - so the numbers here are consistent to about one line rather than exactly. Cell lines are not patients, and a CRISPR knockout is not a drug.
Evidence that argues against this (1)
- The permutation test on this difference does not reach conventional significance: p 0.088 across all leukaemia and 0.142 restricted to AML, and the lineage-consistency scan's own verdict for DOT1L is 'weak or ambiguous' with survives_lineage_split false. The numbers in this entry describe a real direction with an inconclusive test. Oncofusion knowledge graph (internal analysis) 2026
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis
The KMT2A-rearranged arm in DepMap is myeloid-enriched, so a difference between that arm and a comparator mixes fusion status with lineage; CEBPA ranked 25th and SPI1 43rd on this project's own scan and both are artefacts, moving +0.15 and +0.92 in the B-lineage half. DOT1L moves in the depleting direction in both halves and more strongly in the lymphoid half.
Lineage confound in myeloid transcription-factor dependencies: splitting the KMT2A-rearranged arm (Oncofusion KG analysis) · Oncofusion knowledge graph (internal analysis) 2026What this evidence cannot show
The lymphoid arm is four cell lines. A difference of -0.310 estimated from n=4 has a wide interval that none of these runs reports, and the fact that it exceeds the myeloid arm's -0.186 is not established as a real lineage difference - the two are within sampling noise of each other. The claim this entry makes is only the negative one: DOT1L is not a myeloid-lineage artefact, because the lymphoid half does not move the wrong way.
Evidence that argues against this (1)
- Passing a lineage split is a necessary and not a sufficient check. DOT1L's lineage-consistent statistic - which takes the weaker of the two arms by design - is -0.178 in raw Chronos units and its verdict in the interactome lineage-split scan is 'weak or ambiguous'. Not being an artefact is not the same as being a strong dependency. Oncofusion knowledge graph (internal analysis) 2026
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.analysis
Scoring every gene twice on the same numerator, DOT1L ranks 335 of 8,269 on the z-consistency statistic and 61 of 8,269 on the d-consistency statistic in raw Chronos units, gaining 274 places when the comparator-arm SD of 0.401 is removed from the denominator; among the 2,199 genes actually depleted, places gained by the scaling correlate with comparator SD at Spearman -0.92, so the scaled ranking is close to a ranking of narrowness.
Is the lineage-consistency ranking a ranking of effect size or of narrowness? (Oncofusion KG analysis) · Oncofusion knowledge graph (internal analysis) 2026What this evidence cannot show
Both statistics take max(myeloid, lymphoid) - that is, the weaker of the two index arms - so DOT1L's d-consistency of -0.178 is the myeloid arm's line-normalised value and is deliberately conservative. Ranks are over 8,269 scored genes from one seeded sweep with one panel draw; a different seed would move a rank of 61 by an unreported amount. A high rank on raw effect is a statement about effect magnitude, not about statistical significance, which E1's counter-evidence addresses separately.
Evidence that argues against this (1)
- Removing the denominator does not make the effect large in absolute terms. A lineage-consistent raw difference of -0.178 Chronos is about a fifth of what MEN1 shows, and the reranking's own headline is that MEN1 is 1st of 8,269 on raw effect while DOT1L is 61st. Rank 61 is a good position in a genome and a modest dependency in a cell. Oncofusion knowledge graph (internal analysis) 2026
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis
For DOT1L the observed Cohen's d between the KMT2A-rearranged arm (n=11) and other leukaemia (n=42) is -0.576, which requires 47.4 lines per group for 80% power; the panel therefore falls short by roughly a factor of four and the resulting permutation p of 0.088 is the expected output of an underpowered test rather than evidence of no effect.
Power of the DepMap panel to detect a MEN1-scale dependency in rearranged arms (Oncofusion KG analysis) · Oncofusion knowledge graph (internal analysis) 2026What this evidence cannot show
A power calculation performed on the observed effect size is circular as a hypothesis test and is used here only to describe the panel's resolution, not to rescue a non-significant result. Post-hoc power is not evidence that the effect is real. The correct reading is that DepMap cannot distinguish 'moderate real effect' from 'no effect' at this n, in either direction.
Evidence that argues against this (1)
- Underpowered is not the only available explanation. DOT1L's empirical p against random genes is 0.013 and its scan rank is 184 with percentile 1.04, which the countermeasure-ranking run classifies as NOT selective at its own threshold. A gene can be genuinely moderately selective and still not be a good target, and the power argument should not be used to convert a weak result into a strong one. Oncofusion knowledge graph (internal analysis) 2026
- E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis
DOT1L's co-dependency with MEN1 across DepMap lines is r 0.570 pooled, 0.544 after within-tissue centring and 0.681 restricted to leukaemia lines, retaining 95.4% of the pooled correlation after lineage adjustment and exceeding a variance-matched null at the 100th percentile.
Does the MEN1 co-dependency signal survive lineage adjustment? (Oncofusion KG analysis) · Oncofusion knowledge graph (internal analysis) 2026What this evidence cannot show
A co-dependency correlation says the two genes' effects track across lines; it does not say they act in one complex or that inhibiting one substitutes for inhibiting the other. Correlation is computed over 1,208 lines pooled and 54 leukaemia lines, and the leukaemia-only figure rests on the smaller set. Pan-essential genes cannot appear as co-dependencies at all, which biases the whole method toward genes with dynamic range.
Evidence that argues against this (1)
- A strong MEN1-DOT1L co-dependency predicts that menin and DOT1L inhibition should be substantially redundant, which is the opposite of the complementary and synergistic activity reported preclinically for the two drug classes - and the mechanistic account of that synergy has them acting on PRC1.1 by different routes, targeted versus genome-wide. Neville D et al. 2026
lineage-split raw Chronos synthesis: -0.310 lymphoid, -0.186 myeloid, rank 61/8269 on effect
MEN1-DOT1L co-dependency r 0.681 leukaemia-only, 95.4% retained after lineage adjustment
- related to →KMT2A-rearranged acute lymphoblastic leukaemia
the larger arm difference is lymphoid, which is the project beat
DOT1L is the worked example: 335 on the z, 61 on raw effect