All 46 comparator lines screened: the two we tripped over are the only two we can find
also called comparator arm screen · control group audit
Last cycle we found that two of the 'healthy comparison' leukaemia cell lines used in every calculation here secretly run the same biology as the disease we are studying. The obvious next question was how many more there are. We checked all of them: the same two, and no others we can detect — though four lines could not be checked at all, and the method can only find what it is looking for.
The same thing, for a scientist
All 46 non-KMT2A-rearranged leukaemia lines in the DepMap comparator were resolved to Cellosaurus BY DEPMAP ACCESSION and screened for HOXA/MEIS1-axis drivers (PICALM-MLLT10, NUP98 fusions, NPM1c, SET/DEK-NUP214, CDX2, direct HOXA9/MEIS1 lesions, and missed KMT2A rearrangements). 42 resolved; **2 flagged — U-937 (PICALM-MLLT10) and OCI-AML-3 (NPM1 p.Trp288Cysfs*12)**, exactly the two already known. 4 lines could not be resolved.
The answer, and what it is not
2 of 42 resolved comparator lines carry a pathway-equivalent lesion, and they are the two the earlier audit had already found by hand. No third contaminant surfaced.
| line | DepMap | Cellosaurus | lesion |
|---|---|---|---|
| OCI-AML3 | ACH-000336 | CVCL_1844 | NPM1c |
| U-937 | ACH-000406 | CVCL_0007 | PICALM-MLLT10 |
This is a floor, not a clean bill of health. Cellosaurus coverage is uneven: a line with no sequence-variation entry cannot be flagged however well it is characterised elsewhere. 4 lines — YCUB-2, CMS, CHRF-288-11, MY — could not be resolved at all. And the screen only finds lesions on its list, so a HOXA-activating mechanism nobody has named is invisible to it.
Two lines that look like hits and are not
P31/FUJ carries KMT2A p.Arg158fs*13 and F-36P carries KMT2A p.Glu766Gln. An earlier version of this screen flagged both on the bare gene symbol.
They are not contamination. A point mutation or frameshift in KMT2A is not a KMT2A rearrangement — no fusion partner, no retained N-terminus fused to an elongation scaffold, no reason to expect the HOXA/MEIS1 programme. Both belong exactly where they are, in the comparator. The screen now requires a gene fusion variation type rather than the symbol.
Worth stating because the opposite error — moving them into the index arm to be safe — would have quietly redefined the disease being measured.
The tool returned zero against known positives, three times, for three different reasons
This is recorded because each failure produced a plausible answer, which is the dangerous kind, and because this project keeps logging the same class.
1. Wrong field. The first run returned 0 flagged — against a comparator containing two lines already proven positive. Requesting fields=id,ac,di,cc,sy omits sequence-variation-list, which is where Cellosaurus actually stores fusions and mutations. The fusion text was simply not in the response being searched.
2. Wrong line. Searching id:OCI-AML3 returned CVCL_D6C9, "HyCyte OCI-AML3 KO-hFGL2" — a knockout derivative — rather than the parent CVCL_1844. Name matching across these databases is not safe. Searching the DepMap accession resolves exactly: ACH-000336 → CVCL_1844, ACH-000406 → CVCL_0007.
3. Fields that never touch. Even with the right record, NPM1c did not match: Cellosaurus keeps the gene symbol in an xref and the change in mutation-description, so "p.Trp288Cysfs12"* never sits next to "NPM1" — and serialising the raw JSON does not help, because field order puts them arbitrarily far apart. The screen now pairs gene, variation type and description explicitly.
A zero from a database query is a claim about the query until proven otherwise. The only reason these were caught is that two positives were known in advance — which is an argument for always running a new screen against a known-positive control before believing its negatives.
| script | analyses/comparator-arm-screen/run.py |
|---|---|
| status | executed |
| run date | 2026-09-19 |
| n leukaemia lines | 57 |
| n index | 11 |
| n comparator | 46 |
| n resolved | 42 |
| n flagged | 2 |
| headline | The two already found are the only two detectable. Four lines unresolved. The screen returned zero against known positives three times first. |
| caveat | A regex over Cellosaurus prose is a screen, not a genotype. Cellosaurus records are uneven - a line with no karyotype comment cannot be flagged however it is characterised - so this produces a FLOOR on contamination, never a clean bill of health. The HOXA9/MEIS1 pattern in particular will fire on any comment merely mentioning those genes. Every flag needs reading before it is acted on. |
flagged
[
{
"depmapId": "ACH-000336",
"name": "OCI-AML3",
"cellosaurus": "CVCL_1844",
"disease": "Acute Myeloid Leukemia",
"flags": [
"NPM1c"
],
"evidence_snippet": "VAR NPM1 [Mutation] p.Trp288Cysfs*12 (c.860_863dup"
},
{
"depmapId": "ACH-000406",
"name": "U-937",
"cellosaurus": "CVCL_0007",
"disease": "Acute Myeloid Leukemia",
"flags": [
"PICALM-MLLT10"
],
"evidence_snippet": "PICALM-MLLT10"
}
]unresolved
[
{
"depmapId": "ACH-003590",
"name": "YCUB-2"
},
{
"depmapId": "ACH-002709",
"name": "CMS"
},
{
"depmapId": "ACH-003273",
"name": "CHRF-288-11"
},
{
"depmapId": "ACH-002060",
"name": "MY"
}
]lesions screened
[ "PICALM-MLLT10", "NUP98 fusion", "NPM1c", "SET-NUP214", "DEK-NUP214", "CDX2", "direct HOXA9/MEIS1", "KMT2A rearrangement missed by name" ]
not contamination
{
"P31/FUJ": "KMT2A p.Arg158fs*13 - a point mutation, not a rearrangement",
"F-36P": "KMT2A p.Glu766Gln - a point mutation, not a rearrangement"
}tool failure modes
[ "wrong field: fields=id,ac,di,cc,sy omits sequence-variation-list", "wrong line: id:OCI-AML3 resolves to a knockout derivative, not the parent", "fields that never touch: gene symbol in xref, change in mutation-description" ]
unresolved followup 2026 09 19
{
"YCUB-2": {
"accession": "CVCL_A082",
"unique_match": true,
"disease": "TCF3-HLF B-ALL",
"variants_recorded": 2,
"flags": [],
"verdict": "screened clean"
},
"CHRF-288-11": {
"accession": "CVCL_A280",
"unique_match": true,
"disease": "childhood AMKL",
"variants_recorded": 0,
"verdict": "no variant record; uninformative"
},
"CMS": {
"name_hits": 3,
"verdict": "name-ambiguous; not resolved"
},
"MY": {
"name_hits": 15,
"verdict": "name-ambiguous; not resolved"
}
}index arm verified
{
"method": "Cellosaurus by accession, gene-fusion variation type",
"result": "12/12 confirmed KMT2A fusions",
"partners": {
"MLLT3": [
"THP-1",
"MOLM-13",
"MOLM-14",
"NOMO-1"
],
"AFF1": [
"SEM",
"MV4-11",
"RS4;11",
"SEMK2"
],
"AFDN": [
"SHI-1",
"OCI-AML-2",
"ML-2"
],
"MLLT1": [
"HB11;19"
]
}
}- Reporting 'no third contaminant' after a tool that failed three times invites more confidence than the result carries. The honest summary is that the screen can now see the two lesions it was built around, and has never been shown to see anything else.
- Moving P31/FUJ and F-36P into the index arm would have felt like caution and would have been an error - it would have redefined KMT2A-rearranged disease to include point mutations, changing every selectivity number in the project while appearing to make the comparator cleaner.
- Four comparator lines could not be resolved in Cellosaurus by DepMap accession. What are they and what do they carry?
- The index arm is still split by NAME, which this analysis found unsafe in the other direction. Should it be re-split by accession?
- Should a new screen be required to pass a known-positive control before its negatives are reported anywhere in this project?
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.analysis
Screening all 46 non-KMT2A-rearranged leukaemia lines in the DepMap comparator, resolved to Cellosaurus by DepMap accession, finds 2 carrying a pathway-equivalent lesion: U-937 (PICALM-MLLT10, a DOT1L-dependent HOXA/MEIS1 driver) and OCI-AML-3 (NPM1 p.Trp288Cysfs*12, the canonical NPM1c). These are the two identified by the earlier hand audit; no third was found. 4 lines could not be resolved in Cellosaurus. P31/FUJ and F-36P carry KMT2A POINT mutations, not rearrangements, and remain correctly in the comparator.
Comparator-arm screen for pathway-equivalent lesions across all DepMap leukaemia lines · Oncofusion knowledge graph; Cellosaurus + DepMap via Open Targets 2026What this evidence cannot show
Run by the coordinator on 2026-09-19. A FLOOR on contamination, never a clean bill of health: Cellosaurus records are uneven and a line with no sequence-variation entry cannot be flagged however well characterised it is elsewhere; four lines resolved to nothing; and the screen only detects lesions on its list, so an unnamed HOXA-activating mechanism is invisible. Regex over curated prose is a screen, not a genotype. The index arm is still split by NAME, which is the same class of matching this analysis found unsafe in the other direction.
Evidence that argues against this (1)
- Finding no third contaminant is the weakest possible reassurance, because the tool returned zero against two KNOWN positives on its first run and needed three separate fixes to see them. A screen whose negatives were wrong three times should not have its negatives trusted on the fourth without an independent check - and the only check performed here was against the two lines that prompted it. A line carrying, say, a cryptic NUP98 fusion absent from its Cellosaurus record would look identical to a clean line. Oncofusion knowledge graph; Cellosaurus + DepMap via Open Targets 2026
- E_four_unresolved_followed_upmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.database
The four lines the screen could not resolve all EXIST in Cellosaurus; they simply carry no DepMap cross-reference, so accession-based lookup fails. Followed up by name: **YCUB-2** (CVCL_A082, unique match) is TCF3-HLF-rearranged B-ALL with two recorded variants and no HOXA/MEIS1-axis lesion - safely screened and correctly a comparator. **CHRF-288-11** (CVCL_A280, unique match) is childhood acute megakaryoblastic leukaemia with ZERO recorded sequence variants, so 'no flags' means nothing is recorded rather than nothing is there. **CMS** and **MY** could not be resolved safely at all - their names return 3 and 15 Cellosaurus hits respectively, and picking the first would repeat the exact error that returned a knockout derivative for OCI-AML3.
Cellosaurus records for the four comparator lines lacking a DepMap cross-reference · Cellosaurus, SIB 2026What this evidence cannot show
Follow-up run by the coordinator on 2026-09-19. Only ONE of the four is genuinely screened: YCUB-2. CHRF-288-11 has an empty variant record, and CMS and MY are name-ambiguous, so three of four remain unscreened in substance even though all four now have candidate accessions. Two of the three are megakaryoblastic leukaemia lines, which is a lineage with no expected HOXA/MEIS1-axis driver, but that is an argument from lineage rather than from their records.
Evidence that argues against this (1)
- This narrows the gap without closing it, and the residue is the awkward part. An empty Cellosaurus variant list is indistinguishable from a clean one, so CHRF-288-11 contributes nothing to the contamination estimate either way. And resolving CMS or MY by taking the first of 3 or 15 name hits would manufacture a confident answer from an ambiguous query - which is precisely how this screen produced a knockout derivative in place of a parent line earlier in the same session. Leaving them unresolved is the correct outcome, not a failure to finish. Cellosaurus, SIB 2026
- E_index_arm_verifiedstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
The INDEX arm, which is still split by name, verifies cleanly against Cellosaurus by accession: all twelve candidate lines carry a confirmed KMT2A gene fusion with the partner named. THP-1, MOLM-13, MOLM-14 and NOMO-1 carry KMT2A-MLLT3; SEM, MV4-11, RS4;11 and SEMK2 carry KMT2A-AFF1; SHI-1, OCI-AML-2 and ML-2 carry KMT2A-AFDN; HB11;19 carries KMT2A-MLLT1. No line in the index arm is there by mistake.
Cellosaurus sequence-variation records for the KMT2A-rearranged index lines · Cellosaurus, SIB 2026What this evidence cannot show
Confirms no FALSE POSITIVES in the index arm - every line there is a genuine KMT2A fusion. It does not by itself exclude false NEGATIVES, a KMT2A-fusion line sitting in the comparator because its name is not on the list; that is what the screen's gene-fusion pattern covers, and it found only KMT2A point mutations. Cellosaurus fusion annotation is itself curated from publications and inherits their errors.
Evidence that argues against this (1)
- A verified index arm makes the panel's OTHER limitation more visible rather than less: the eleven lines are genuine KMT2A fusions, but they are eleven lines from ten patients, none infant-derived, weighted to AML and to three partners. Confirming that a small unrepresentative panel is correctly labelled does not make it representative. Cellosaurus, SIB 2026
- ← related toEWSR1::FLI1 (EWS-FLI1), t(11;22)(q24;q12)
Counter-evidence: Brennan 2022 (PMID 36522207) randomised 640 patients between two regimens that BOTH contain ifosfamide and etoposide, 3-year EFS 61% vs 67%, so benefit tracks the regimen not the two added drugs
Closes its open question: screening all 46 comparator lines finds the same two contaminants and no third.
- related to →The whole protein-coding genome: menin is still first, and a metabolic pathway appears (run)
Bounds the contamination in this scan's comparator at two known lines, with four unresolved.
The analysis whose control arm prompted the audit.
The hand audit that found the first two; this is its systematic version.
Supplies the two pathway-equivalent lines this analysis splits out, and shows they are not equivalent to each other.