Live DepMap on fourteen genes: CDK6's pan-essentiality is a lineage artefact of pooling, and not one gene of the homeodomain cofactor code is a dependency anywhere
also called CDK6 vs CDK4 lineage split · homeodomain code dependency null
We asked a public database, live, how badly each of these fourteen genes is needed across 1,258 cancer cell lines and across the 57 blood-cancer lines specifically. Two answers came back: the cell-cycle gene CDK6 is needed everywhere a bit and in blood cancers a lot, while its twin CDK4 is the reverse - and none of the HOX/PBX/MEIS genes this beat is about is needed anywhere at all.
The same thing, for a scientist
A single live query against DepMap CRISPR (Chronos) gene effect through the Open Targets GraphQL API on 2026-09-20, for fourteen genes, reporting mean and median across all lines, the fraction of lines at or below -0.5 and -1.0, and the mean restricted to leukaemia-annotated lines. Two results. **(1) The CDK6 pan-essentiality flag in `docs/coverage.md` priority 7 survives as arithmetic and dissolves as an argument.** CDK6's all-lines mean is -0.6321 (past the project's -0.5 threshold) but its leukaemia mean is -1.3767, while CDK4 is the *more* pan-essential gene (-0.7371) and the *less* required one in leukaemia (-0.4464). The pooled statistic averages across a tissue partition. **(2) Not one gene of the homeodomain cofactor code is a dependency in any leukaemia line.** PBX3 +0.0418, PBX1 +0.0391, MEIS1 -0.1791, HOXA9 -0.2230, HOXA10 -0.0997, BMI1 -0.0103, WT1 +0.1400, ID2 -0.1131 - all leukaemia-line means. The deepest PBX paralogue is PBX2 (-0.3464), which the literature says is the *wrong* one.
The table
DepMap CRISPR (Chronos) gene effect via Open Targets GraphQL depMapEssentiality.screens, retrieved 2026-09-20. Leukaemia subset = screens whose diseaseFromSource contains leukemia, leukaemia, lymphoblastic or myeloid (n=57 for every gene except MEIS1, n=56).
| gene | n lines | mean, all | median, all | <= -0.5 | <= -1.0 | mean, leukaemia |
|---|---|---|---|---|---|---|
| CDK6 | 1,258 | -0.6321 | -0.5267 | 51.9% | 23.1% | -1.3767 |
| CDK4 | 1,258 | -0.7371 | -0.5645 | 58.4% | 22.3% | -0.4464 |
| CCND3 | 1,258 | -0.3459 | -0.2503 | 17.0% | 6.4% | -1.0076 |
| PBX2 | 1,257 | -0.2524 | -0.2393 | 6.3% | 0.2% | -0.3464 |
| HOXA9 | 1,254 | -0.1727 | -0.1587 | 3.8% | 0.2% | -0.2230 |
| ID2 | 1,257 | -0.1226 | -0.1080 | 2.9% | 0.1% | -0.1131 |
| HOXA10 | 1,255 | -0.0628 | -0.0610 | 0.4% | 0.1% | -0.0997 |
| MEIS1 | 1,257 | -0.0528 | -0.0430 | 1.3% | 0.1% | -0.1791 |
| BMI1 | 1,258 | +0.0585 | +0.0738 | 1.4% | 0.2% | -0.0103 |
| PBX1 | 1,258 | +0.0110 | +0.0155 | 0.5% | 0.0% | +0.0391 |
| PBX3 | 1,258 | +0.0297 | +0.0265 | 0.08% | 0.0% | +0.0418 |
| WT1 | 1,251 | +0.0825 | +0.0910 | 0.8% | 0.5% | +0.1400 |
| CDKN2A | 1,255 | +0.1570 | +0.1496 | 0.2% | 0.0% | +0.1309 |
| RB1 | 1,258 | +0.1801 | +0.1632 | 0.2% | 0.0% | +0.1127 |
Result 1 - the pan-essentiality flag and the selectivity claim are both true
The ledger's priority 7 says CDK6 comes back pan-essential, which it calls an argument about the therapeutic window that the priority as written does not make. Both halves check out:
- Pan-essential, yes. Mean -0.6321 across 1,258 lines clears the project's own -0.5 threshold, and 51.9% of all cancer lines lose fitness on CDK6 knockout. Anyone told only that CDK6 is selectively required in KMT2A-r has been misled.
- And CDK4 is more pan-essential still (-0.7371, 58.4% of lines), which nobody in this graph had written down.
- The resolution is lineage. CDK6's leukaemia mean is -1.3767 and CDK4's is -0.4464, a separation of 0.93 gene-effect units. CDK6 is the haematopoietic G1 kinase; CDK4 is the kinase of most solid tissues. Pooling averages across the partition and produces a number that describes neither side.
This is the same failure the project already documented for its z-score scan and for lineage-unsplit selectivity: a pooled statistic over a heterogeneous population. The correction it implies is not that CDK6 is a good target - it is that the window is a LINEAGE window, and the lineage is the dose-limiting tissue. That argument is developed in gene-cdk6.
Result 2 - the homeodomain cofactor code has no dependency anywhere, and PBX3 is the emptiest
Eight of the fourteen genes are the transcription-factor code this beat was dispatched to write. None of them is a fitness requirement of any leukaemia line. PBX3 is the most extreme: one line in 1,258 falls below -0.5, and none below -1.0, and the leukaemia mean is positive.
Four readings, and the entry does not choose between them:
- Wrong assay. Every published positive for these genes is transformation, LSC frequency, serial replating or transplantation. A 21-day CRISPR fitness screen in an immortalised line sees none of those. This is the standard and probably correct explanation.
- Paralogue buffering. PBX1/PBX2/PBX3 share a DNA-binding mode; the HOXA cluster is the textbook redundant family; BMI1 is one of six PCGFs. Single-gene knockout is buffered by construction in every case. Note the internal evidence for this: PBX2, the paralogue the literature excludes, has the deepest effect of the three (-0.3464), which is what a partially redundant family with uneven expression looks like, not what a clean dependency looks like.
- Wrong lines. Eleven KMT2A-rearranged lines, myeloid-enriched, none infant-derived.
- The literature is over-read. Kept on the list because it is a live possibility: the human evidence for PBX3 is co-expression and prognosis, and the functional evidence is murine.
What is NOT available as a reading: that this table refutes the literature. It does not. It bounds one assay, and it says so.
What the tumour-suppressor rows are and are not
CDKN2A (+0.157), RB1 (+0.180) and WT1 (+0.083) all score positive - knockout is neutral to marginally favourable. These rows should not be read as biology. A large fraction of DepMap lines carry homozygous CDKN2A or RB1 deletions, so the guide has nothing to cut in those lines, and CRISPR scores at deleted or amplified loci are dominated by copy-number and cutting artefacts. The copy-number distribution was not checked by this analysis, so the positive values are recorded and explicitly not interpreted.
They are included because leaving them out would have made the table look like a clean dependency ranking, and it is not one.
| script | scratchpad query against https://api.platform.opentargets.org/api/v4/graphql; not persisted under analyses/ because this beat may not write outside its own three paths |
|---|---|
| status | run |
| run date | 2026-09-20 |
| n genes | 14 |
| n lines all | 1251-1258 depending on gene |
| n lines leukaemia | 56-57 depending on gene |
| result summary | CDK6's pan-essentiality flag is arithmetically correct and dissolves as an argument once split by lineage: CDK6 -0.6321 all lines vs -1.3767 in leukaemia; CDK4 -0.7371 all lines vs -0.4464 in leukaemia, so CDK4 is the broader dependency and CDK6 the deeper haematopoietic one, separated by 0.93 gene-effect units. Separately, none of the eight homeodomain-code genes is a dependency in any leukaemia line, with PBX3 the most extreme at +0.0418 (leukaemia) and 1 line in 1,258 below -0.5. |
| internal consistency check | CDK6 all-lines mean from this retrieval (-0.6321, n=1,258) against the persisted unwritten-interactor triage (-0.6323, n=1,208): agreement to 0.0002 across two independent retrievals on different line counts. |
| what this does not show | It does not refute the PBX3, MEIS1, HOXA9 or BMI1 literature. Every published positive for those genes is a transformation, self-renewal, LSC-frequency or in vivo result, and this assay is 21-day proliferation. It bounds ONE assay. |
inputs
[ "DepMap CRISPR Chronos gene effect via Open Targets depMapEssentiality.screens", "14 Ensembl gene ids", "screen diseaseFromSource strings for the leukaemia subset" ]
results table — the numbers behind the chart
{
"CDK6": {
"n": 1258,
"mean_all": -0.6321,
"median_all": -0.5267,
"frac_le_0_5": 0.5191,
"frac_le_1_0": 0.2313,
"mean_leukaemia": -1.3767
},
"CDK4": {
"n": 1258,
"mean_all": -0.7371,
"median_all": -0.5645,
"frac_le_0_5": 0.5835,
"frac_le_1_0": 0.2226,
"mean_leukaemia": -0.4464
},
"CCND3": {
"n": 1258,
"mean_all": -0.3459,
"median_all": -0.2503,
"frac_le_0_5": 0.1701,
"frac_le_1_0": 0.0644,
"mean_leukaemia": -1.0076
},
"PBX2": {
"n": 1257,
"mean_all": -0.2524,
"median_all": -0.2393,
"frac_le_0_5": 0.0628,
"frac_le_1_0": 0.0016,
"mean_leukaemia": -0.3464
},
"HOXA9": {
"n": 1254,
"mean_all": -0.1727,
"median_all": -0.1587,
"frac_le_0_5": 0.0375,
"frac_le_1_0": 0.0016,
"mean_leukaemia": -0.223
},
"ID2": {
"n": 1257,
"mean_all": -0.1226,
"median_all": -0.108,
"frac_le_0_5": 0.0286,
"frac_le_1_0": 0.0008,
"mean_leukaemia": -0.1131
},
"HOXA10": {
"n": 1255,
"mean_all": -0.0628,
"median_all": -0.061,
"frac_le_0_5": 0.004,
"frac_le_1_0": 0.0008,
"mean_leukaemia": -0.0997
},
"MEIS1": {
"n": 1257,
"mean_all": -0.0528,
"median_all": -0.043,
"frac_le_0_5": 0.0127,
"frac_le_1_0": 0.0008,
"mean_leukaemia": -0.1791,
"n_leukaemia": 56
},
"BMI1": {
"n": 1258,
"mean_all": 0.0585,
"median_all": 0.0738,
"frac_le_0_5": 0.0143,
"frac_le_1_0": 0.0024,
"mean_leukaemia": -0.0103
},
"PBX1": {
"n": 1258,
"mean_all": 0.011,
"median_all": 0.0155,
"frac_le_0_5": 0.0048,
"frac_le_1_0": 0,
"mean_leukaemia": 0.0391
},
"PBX3": {
"n": 1258,
"mean_all": 0.0297,
"median_all": 0.0265,
"frac_le_0_5": 0.0008,
"frac_le_1_0": 0,
"mean_leukaemia": 0.0418
},
"WT1": {
"n": 1251,
"mean_all": 0.0825,
"median_all": 0.091,
"frac_le_0_5": 0.008,
"frac_le_1_0": 0.0048,
"mean_leukaemia": 0.14
},
"CDKN2A": {
"n": 1255,
"mean_all": 0.157,
"median_all": 0.1496,
"frac_le_0_5": 0.0024,
"frac_le_1_0": 0,
"mean_leukaemia": 0.1309,
"do_not_interpret": "copy-number confounded at a commonly deleted locus; CN distribution not checked"
},
"RB1": {
"n": 1258,
"mean_all": 0.1801,
"median_all": 0.1632,
"frac_le_0_5": 0.0016,
"frac_le_1_0": 0,
"mean_leukaemia": 0.1127,
"do_not_interpret": "copy-number confounded at a commonly deleted locus; CN distribution not checked"
}
}- This table is the kind of result that gets quoted without its caveats. The two numbers most likely to be misquoted are PBX3 +0.0418, which does NOT mean PBX3 is unimportant, and CDK6 -1.3767, which does NOT mean CDK6 is KMT2A-specific - the leukaemia subset is not conditioned on fusion status.
- A lineage split with n=57 on one side is a thin comparison, and no confidence interval was computed. The direction is large and consistent with two prior analyses; the magnitude should not be quoted to four decimal places as if it were precise, which is exactly what the table above invites.
- The analysis was run in a scratchpad and its script is not persisted under analyses/, because this beat is confined to three paths. Anyone re-running it should re-fetch rather than trusting these numbers, and the DepMap release is unstated so the values can move.
- Is the leukaemia CDK6 signal carried by myeloid or lymphoid lines? The 57-line subset was not split by lineage within haematopoiesis, and the index disease is B-lineage. This is the single most valuable follow-up and it is one more grouping of the same retrieved data.
- Does DepMap's own common-essential flag agree with this project's -0.5 threshold for CDK6? The flag was not retrieved, and the project convention is not a DepMap definition.
- What are the copy-number states behind the positive CDKN2A and RB1 rows? Until that is checked those two numbers should not be quoted at all.
- Would a combinatorial PBX1/PBX2/PBX3 knockout show a dependency the singles cannot? No such screen exists in DepMap and the paralogue-buffering explanation is currently untested rather than supported.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
DepMap CRISPR (Chronos) gene effect retrieved live on 2026-09-20 for fourteen genes gives the following means across all lines and across leukaemia-annotated lines: CDK6 -0.6321/-1.3767; CDK4 -0.7371/-0.4464; CCND3 -0.3459/-1.0076; PBX2 -0.2524/-0.3464; HOXA9 -0.1727/-0.2230; ID2 -0.1226/-0.1131; HOXA10 -0.0628/-0.0997; MEIS1 -0.0528/-0.1791; BMI1 +0.0585/-0.0103; PBX1 +0.0110/+0.0391; PBX3 +0.0297/+0.0418; WT1 +0.0825/+0.1400; CDKN2A +0.1570/+0.1309; RB1 +0.1801/+0.1127. Line counts are 1,251-1,258 for the all-lines column and 56-57 for the leukaemia column.
DepMap CRISPR (Chronos) gene effect via the Open Targets Platform GraphQL API, depMapEssentiality.screens, for 14 Ensembl gene ids · Broad Institute DepMap; Open Targets Platform; query written and run by this project 2026-09-20 · Database query 2026What this evidence cannot show
A database fact opposable on scope. (1) The leukaemia subset is a case-insensitive substring match on the screen's own diseaseFromSource field for leukemia/leukaemia/lymphoblastic/myeloid - a coarse text match, not a curated lineage annotation, and NOT conditioned on fusion status, so it says nothing about KMT2A rearrangement. (2) The DepMap release version is not stated by the API, so this table is not reproducible to a specific release. (3) Chronos gene effect is 21-day CRISPR knockout fitness in immortalised lines: it cannot see self-renewal, differentiation, LSC frequency or any in vivo phenotype. (4) No infant-derived line is in the panel. (5) Tumour-suppressor rows are confounded by copy-number at deleted loci and the copy-number distribution was NOT checked. (6) Ensembl ids were taken from the query script and were not independently re-resolved against HGNC this cycle; the returned approvedSymbol was used as the match key, so a wrong id would have surfaced as a wrong symbol.
Evidence that argues against this (1)
- This project's own persisted genome-wide scan, computed on a different line partition (11 KMT2A-rearranged vs 43 other leukaemia lines), gives CDK6 a difference of -0.5266 at rank 13 and CDK4 +0.0716 at rank 16,052 - consistent in direction with this table and computed on a different comparator, which is the closest thing to independent replication available here. analysis run by this project 2026
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
The coverage ledger's priority 7 states that CDK6 has no entry of its own and comes back pan-essential in the unwritten-interactor triage, which is an argument about the therapeutic window that the priority as written does not make; the underlying triage records CDK6 with an index mean of -1.7918, other-leukaemia mean -1.2652, difference -0.5266 and pan-essentiality -0.6323 across 1,208 lines, and BMI1 with index mean -0.0361 and pan-essentiality +0.0591.
docs/coverage.md priority 7 and analyses/unwritten-interactors-dependency-triage/results.json (all_rows) · Oncofusion Knowledge Graph coordinator; triage run by this project · Internal ledger and derived analysis 2026What this evidence cannot show
Read from the persisted results file on 2026-09-20, not re-run. The triage's pan-essentiality threshold is a mean at or below -0.5 across all lines with data, which is a project convention and not a DepMap definition of common-essentiality; DepMap's own common-essential flag was not retrieved. The triage used 1,208 lines against this analysis's 1,258, and CDK6's value agrees to 0.0002 (-0.6323 vs -0.6321), which is a useful internal consistency check across two retrievals.
Evidence that argues against this (1)
- The triage classified CDK6 as pan-essential and therefore 'not selective at any effect size', which is the correct call under its own threshold and is what this analysis shows to be an artefact of pooling across a tissue partition - so the instrument was working and the conclusion it licensed was too strong. analysis run by this project 2026
the dependency null that explains why no in vitro potency assay exists for this target
live lineage-split DepMap: -0.6321 all lines, -1.3767 in 57 leukaemia lines
CDK4 is the broader dependency (-0.7371) and the shallower one in leukaemia (-0.4464)
bounded negative: PBX3 knockout is fitness-neutral in every leukaemia line in the panel
bounded negative: BMI1 knockout is fitness-neutral, consistent with both primary papers' in vitro arms
bounded negative: HOXA10 knockout is fitness-neutral; paralogue-buffered by construction
supplies the lineage split the existing entry's DepMap block did not have