The KMT2A-rearranged DepMap arm is not MTAP-deleted, and the gene that looked like the test cannot answer it
also called MTAP deletion in the KMT2A-rearranged arm · MAT2A rank 911 is not a panel artefact
Last cycle we worried that a few of this project's best-looking drug-target numbers might be an accident of which cell lines we happened to compare, because cells missing a gene called MTAP are unusually sensitive to that whole chemical pathway. The answer is no: the most reliable warning signs point the other way, so the numbers are not an artefact of the panel.
The same thing, for a scientist
Cycle 40 flagged that MAT2A's selectivity rank (911 of 17,728) could be a property of the 11-line KMT2A-rearranged arm rather than of the fusion, if that arm were enriched for MTAP deletion. Tested against the cached full scan. MTAP's own Chronos score is uninformative BY CONSTRUCTION - a deleted gene has nothing to cut and scores ~0 - so the test is the synthetic-lethal partners. PRMT5 (rank 17,006) and its obligate partner WDR77 (15,455) are both LESS depleted in the rearranged arm, the opposite of what MTAP deletion predicts. No evidence of enrichment. MAT2A's rank is not an MTAP artefact - and is also not a strong selectivity finding, being a -0.11 arm difference on a -1.35 pan-essential requirement.
Why the answer is 'no' rather than 'inconclusive'
The worry was right to be raised, and resolving it cost one script and no API calls
| analysis dir | analyses/mtap-deletion-in-the-kmt2a-arm/ |
|---|---|
| source scan | analyses/depmap-kmt2a-selective-scan-full/results.json |
| source scan generated | 2026-09-19T10:12:06Z |
| n kmt2a rearranged lines | 11 |
| n other leukemia lines | 43 |
| genes scanned | 17728 |
| answer | No evidence of MTAP-deletion enrichment in the KMT2A-rearranged arm. |
| shown for fusion | NOT APPLICABLE - this is a property of a DepMap cell-line panel, not a claim about any fusion protein. The 11 rearranged lines are adult and paediatric leukaemia lines; none is infant-derived. |
panel — the numbers behind the chart
{
"PRMT5": {
"rank": 17006,
"mean_kmt2a_r": -1.2344,
"mean_other": -1.3403,
"difference": 0.106,
"pan_essentiality_all_lines": -1.1967,
"reads_out": "canonical MTAP-deletion dependency",
"direction": "AGAINST MTAP deletion"
},
"WDR77": {
"rank": 15455,
"mean_kmt2a_r": -1.5903,
"mean_other": -1.6503,
"difference": 0.06,
"pan_essentiality_all_lines": -1.354,
"reads_out": "PRMT5 obligate partner (MEP50)",
"direction": "AGAINST MTAP deletion"
},
"MAT2A": {
"rank": 911,
"mean_kmt2a_r": -1.5424,
"mean_other": -1.431,
"difference": -0.1114,
"pan_essentiality_all_lines": -1.3485,
"reads_out": "SAM supply to PRMT5 - the gene the worry was about",
"direction": "consistent with MTAP deletion"
},
"RIOK1": {
"rank": 969,
"mean_kmt2a_r": -1.1249,
"mean_other": -1.0169,
"difference": -0.108,
"pan_essentiality_all_lines": -0.998,
"reads_out": "PRMT5 substrate/adaptor",
"direction": "consistent with MTAP deletion"
},
"MTAP": {
"rank": 7038,
"mean_kmt2a_r": 0.0711,
"mean_other": 0.0872,
"difference": -0.0161,
"pan_essentiality_all_lines": 0.0306,
"reads_out": "UNINFORMATIVE BY CONSTRUCTION - Chronos cannot see a deletion",
"direction": "not interpretable"
}
}- MAT2A and PRMT5 inhibitors are in clinical development, and this entry must not be read as support for using one in KMT2A-rearranged leukaemia. The result here is a NEGATIVE about panel composition, not a positive about a target: it says MAT2A's rank is not an MTAP artefact, and in the same breath says that rank is a -0.11 difference on a -1.35 pan-essential requirement, which is a weak basis for anything.
- A pan-essential gene is a bad drug target for the reason the number makes obvious - it is required at similar depth in the comparator lines, which in a patient are the normal proliferating tissues. Nothing here addresses therapeutic index.
- All of this is cell-line data from 11 adult and paediatric lines, none infant-derived, in a disease whose defining presentation is infant. No claim here should be carried to an infant patient.
- MAT2A and RIOK1 move together (-0.111, -0.108) while PRMT5 and WDR77 move together (+0.106, +0.060). SAM supply and SAM consumption diverging within one pathway is not MTAP biology and has no explanation here. Is it noise on n=11, or is there a SAM-supply requirement in KMT2A-rearranged lines that is independent of PRMT5?
- A direct copy-number lookup on the 11 rearranged lines would settle the deletion question outright rather than inferring it from phenotype. CORRECTED 2026-09-21 by Q73: I wrote that 'the line identifiers are in the source scan'. THEY ARE NOT. `analyses/depmap-kmt2a-selective-scan{,-full}/results.json` record only `n_kmt2a_rearranged: 11` and `n_other_leukemia: 43`, and run.log has no list either. That is precisely WHY this question had to be answered by phenotype inference rather than by a copy-number lookup - the lookup is not one API call away, it needs the scan re-run with the resolved line ids persisted. Third instance of rule 90 in one cycle: the identities were computed and only the count was written.
- The four FA-pathway ranks that triggered the cycle-40 worry alongside MAT2A - FANCF 24, FANCI 52, BRIP1 63, FANCD2 465 - are untouched by this result. The replication-stress half of that worry is still open and needs its own test.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.
MTAP deletion makes cancer cells dependent on PRMT5. This is the observation the entire worry rested on, and it is the reason PRMT5 is the right read-out: if the 11 KMT2A-rearranged lines were enriched for MTAP loss, PRMT5 knockout would be MORE damaging in that arm. Measured in the cached scan it is LESS damaging - arm means -1.2344 against -1.3403, difference +0.106, rank 17,006 of 17,728.
MTAP deletion confers enhanced dependency on the PRMT5 arginine methyltransferase in cancer cells. · Kryukov GV, Wilson FH, Ruth JR · Science 2016What this evidence cannot show
The PRMT5 dependency of MTAP-null cells is established in solid-tumour and mixed-lineage panels; this analysis applies it to a leukaemia-only comparison, where the baseline PRMT5 requirement is already deep (pan-essentiality -1.20) and the dynamic range for detecting extra dependency is correspondingly compressed. Rank 17,006 is a difference of means with NO computable confidence interval: the cached file stores arm means, not per-line Chronos values. The direction is what is load-bearing, not the magnitude (rule 87). 11 rearranged lines against 43 comparators. A 0.1 difference on n=11 is not distinguishable from zero without per-line data.
Evidence that argues against this (1)
- Mavrakis et al. showed the same dependency arises from disordered methionine metabolism in MTAP/CDKN2A co-deleted cancers (PMID 26912361), and CDKN2A is co-deleted with MTAP in most cases because the loci are adjacent at 9p21. This graph's scan puts CDKN2A at rank 17,552 with POSITIVE Chronos in both arms, which Q69 correctly refused to read as anti-selectivity because CDKN2A is a tumour suppressor. That means the CDKN2A row cannot corroborate the MTAP call either way, and one of the two loci that usually travel together is therefore uninformative here. Mavrakis KJ et al. 2016
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.
The synthetic-lethal axis created by MTAP deletion is specifically MAT2A/PRMT5/RIOK1 - all three, named together. This is why the test used four genes rather than one: MTAP deletion predicts that every member becomes MORE required. Observed, the axis SPLITS - MAT2A (-0.111) and RIOK1 (-0.108) go one way while PRMT5 (+0.106) and WDR77 (+0.060) go the other. A coherent MTAP-deletion signal cannot split down the middle of its own axis.
MTAP Deletions in Cancer Create Vulnerability to Targeting of the MAT2A/PRMT5/RIOK1 Axis. · Marjon K, Cameron MJ, Quang P · Cell Rep 2016What this evidence cannot show
This PMID is already cited in `gene-mat2a`; it is cited rather than restated here, per rule 82. The new content is the measurement against this graph's own arm split, not the axis itself. Treating a 2-2 split as 'incoherent' assumes the four members should move together, which is true for a deletion-driven effect but not necessarily for any other cause of MAT2A dependency. All four genes are deep shared requirements (pan-essentiality -1.00 to -1.35). Every difference discussed sits on a compressed scale, which is the category error Q71 named for HCFC1: report the arm means, not the rank.
Evidence that argues against this (1)
- The split could be read the opposite way - that MAT2A and RIOK1 are genuinely more required in the rearranged arm and PRMT5 is not, which would be a real finding about SAM supply rather than a refutation of MTAP enrichment. This analysis cannot exclude that reading, and records it as the open question rather than resolving it. What it does exclude is the specific explanation the cycle-40 worry proposed, because that explanation predicts all four move together. Marjon K et al. 2016
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.
MTAP's own Chronos score cannot answer this question and reading it would have been a category error. A CRISPR knockout score measures the consequence of cutting a gene; a gene that is already homozygously deleted has no target to cut and scores near zero regardless of how important its absence is. MTAP scores +0.0711 in the rearranged arm and +0.0872 in the comparators - positive in both, i.e. not a dependency anywhere - and that is exactly what would be observed whether the lines are deleted or intact.
mtap-deletion-in-the-kmt2a-arm: are the 11 KMT2A-rearranged DepMap lines enriched for MTAP deletion, which would make MAT2A's selectivity rank a property of the panel? · This repository (analysis, reproducible from the cached scan) 2026What this evidence cannot show
This is a methodological claim about what Chronos measures, argued from the design of the assay rather than from a citation that states it about MTAP specifically. It is the fourth instance in this graph of the same category error - CDKN2A, SIRT1 and ATM were each nearly read as anti-selective when the correct statement was that the rank does not apply. MTAP is a different flavour of the same mistake: not 'positive Chronos means tumour suppressor' but 'near-zero Chronos means nothing was cut'. The test used here - infer deletion from the synthetic-lethal partners - measures the PHENOTYPE of MTAP loss, not copy number. A direct copy-number lookup on the 11 lines would be strictly better and is still worth doing.
Evidence that argues against this (1)
- A near-zero Chronos score is also what an intact but genuinely dispensable gene produces, so the score is consistent with the lines being MTAP-intact as well. That is precisely the point - the row is uninformative in both directions - but it means this evidence item establishes only that MTAP's own score must be excluded from the test, not that the lines are intact. The conclusion that they are not enriched for deletion rests on E1 and E2. This repository (analysis, reproducible from the cached scan) 2026
THE ENTRY THE WORRY WAS ABOUT. MAT2A's selectivity rank of 911 was the number at risk: if the 11 KMT2A-rearranged lines were enriched for MTAP deletion, that rank would be a property of the panel rather than of the fusion. Tested and not supported - PRMT5 and WDR77, the two most direct read-outs of MTAP status, are both LESS depleted in the rearranged arm. The rank survives as not-an-artefact, and separately fails to be impressive: -0.11 on a -1.35 pan-essential requirement.
THE CO-DELETED LOCUS THAT CANNOT CORROBORATE. MTAP and CDKN2A sit adjacent at 9p21 and are usually lost together, so CDKN2A status would normally be a second read on MTAP deletion. It cannot be one here: CDKN2A has POSITIVE Chronos in both arms because it is a tumour suppressor, and Q69 already established that its rank of 17,552 must not be read as anti-selectivity. One of the two loci that travel together is therefore uninformative, which is why the test rests on the synthetic-lethal partners instead.
- related to →HCFC1 (HCF-1) — a founding member of the wild-type KMT2A complex that the fusion cannot bind
THE SAME CATEGORY OF READING ERROR, one cycle apart. Q71 showed HCFC1's rank is uninterpretable because both arms sit at -1.8 and a -0.073 difference on a deep shared requirement is not selectivity at any rank. Every gene in this MTAP panel is in that same condition - pan-essentiality -1.00 to -1.35, differences around 0.1 - so the same discipline applies: report the arm means, not the rank.
- related to →FANCI — ID2 module; rank 15 of 8,269 for KMT2A-selective dependency, and no KMT2A literature at all
THE HALF OF THE CYCLE-40 WORRY THAT IS STILL OPEN. That worry had two limbs - MTAP deletion and elevated baseline replication stress - and named FANCF 24, FANCI 52, BRIP1 63, FANCD2 465 and MAT2A 911 as the five ranks that would evaporate together if either were true. This analysis closes the MTAP limb only. The replication-stress limb is untouched and still needs its own test, which would be a gamma-H2AX panel rather than a lookup.
SAME LESSON FROM THE OTHER DIRECTION. That analysis found a repeated DepMap proportion was the null; this one finds a repeated DepMap concern was answerable from a file already on disk. Both turn on refusing to read a rank as a result without asking what the number would look like if nothing were going on - there, the median; here, what Chronos does when a gene is already deleted.
asked whether this project's 11-line KMT2A-rearranged DepMap arm occupies this state; the answer is no, and the analysis also establishes that the deleted gene's own Chronos score cannot answer it