Six gene bodies are AFF1-exclusive and recurrent at gene scale; the authors' own pipeline confirms one of them, refutes three and has never measured two - and IRX1 is the one it confirms
also called q154 class audit · the six: ELF1, GNAQ, H3C15, IRX1, PAN3, SPAG7
Exactly six genes in the whole genome are hit by the fusion protein in three or more of the eight AFF1 leukaemias and in none of the other eighteen patients. Five of those six have a non-AFF1 patient sitting just under the cut-off, and the scientists who made the data call three of them in non-AFF1 patients too. IRX1 is the one where the nearest non-AFF1 patient is eight times below the bar and the original pipeline agrees it is AFF1-only.
The same thing, for a scientist
The gene-body AFF1-exclusive recurrent class - occupied in >=3 of 8 KMT2A::AFF1 samples and 0 of 18 non-AFF1 - has exactly six members: **ELF1, GNAQ, H3C15, IRX1, PAN3, SPAG7** [E1]. Two independent audits separate IRX1 from the rest. **Margin.** The highest non-AFF1 min-of-two-replicates share at each member, against the 2.0x gate: SPAG7 **1.807** (90% of the gate), H3C15 **1.973** (99%), ELF1 **1.605** (80%), PAN3 **1.337** (67%), GNAQ **0.911** (46%), **IRX1 0.252 (13%)**. **Five of the six have a non-AFF1 sample within a factor of 2.2 of the threshold; IRX1's nearest non-AFF1 competitor is 7.9-fold below it** [E1]. **Independent instrument.** Looked up in Janssens et al.'s own published per-sample target table: **ELF1, GNAQ and PAN3 are all called in non-AFF1 samples** (ELF1 in the KMT2A-ENL samples p160/p186/p214; GNAQ in p186, TB971992, TB995291 and the KMT2A-AF9 A95941; PAN3 in p160/p186/p214, TB995291 and A95941), so their exclusivity is **refuted**. H3C15 and SPAG7 appear nowhere in that table - **unmeasured**. **IRX1's row there is exactly `['p279','p318','p454']` - confirmed** [E2]. **Threshold stability.** Re-scoring the whole class at gates of 1.25, 1.5, 1.75, 2.0, 2.5 and 3.0 gives class sizes 8, 6, 9, 6, 6 and 3, with membership turning over almost completely; only **GNAQ and PAN3** are members at every gate from 1.5 to 3.0. **IRX1 is a member from 1.25 to 2.0 and drops out at 2.5**, because p318's IRX1 is 3.052x and 2.167x and fails a 2.5 gate - its *exclusivity* (0 of 18) holds at every gate tested, only its recurrence tier moves [E1]. **The class as a set is a property of the ruler; IRX1's exclusivity is not.**
The six, audited three ways
| gene | AFF1 samples | highest non-AFF1 share (gate 2.0) | margin | in the authors' table | authors' non-AFF1 calls | verdict |
|---|---|---|---|---|---|---|
| IRX1 | 3 (p279, p318, p454) | 0.252 (A73051) | 7.9x | yes | none | confirmed exclusive |
| GNAQ | 5 | 0.911 (A80956) | 2.2x | yes | p186, TB971992, TB995291, A95941 | refuted |
| PAN3 | 7 (all but TB920488) | 1.337 (A95941) | 1.5x | yes | p160, p186, p214, TB995291, A95941 | refuted |
| ELF1 | 5 (all Texas Children's) | 1.605 (A95941) | 1.25x | yes | p160, p186, p214 | refuted |
| SPAG7 | 3 | 1.807 (A93949) | 1.11x | no | - | unmeasured |
| H3C15 | 3 | 1.973 (p214) | 1.01x | no | - | unmeasured |
H3C15's nearest non-AFF1 sample misses the gate by 1.4%. SPAG7's misses by 9.7%. Neither should be described as exclusive.
What this does to the class-level null
The class of six sits far above its permutation null (0.88 unstratified, 0.86 under an exact institution-stratified permutation), which on its own reads as a real partner-exclusive programme. The member-level audit says most of that signal is a sensitivity effect rather than an exclusivity one: four of the six have a non-AFF1 sample within 50% of the threshold, and the three that the authors' more sensitive, FDR-based caller can see are all called in non-AFF1 samples by it.
That is the direction of the artefact and it was declared in advance: a fold-change gate applied to samples that differ in oncoprotein signal will convert a gradient into an exclusivity, and the paper this deposit comes from is titled MLL oncoprotein levels influence leukemia lineage identities. The label permutation cannot remove it, because the gradient is real and is correlated with the label.
What survives the audit is IRX1 alone - and PAN3 is the informative counter-example, because PAN3 is on this project's own list of canonical KMT2A-fusion program genes recovered by the q143 filter-free ranking. A gene already known to be a general KMT2A-fusion target appearing as AFF1-exclusive is a direct demonstration that the gate manufactures exclusivity.
| script | analyses/q154-gene-scale-occupancy-null/analyze.py |
|---|---|
| status | run |
| result summary | Class of six at gate 2.0; margins 0.252 (IRX1) to 1.973 (H3C15); authors confirm IRX1, refute ELF1/GNAQ/PAN3, have no row for H3C15/SPAG7; class membership turns over across gates 1.25-3.0 while IRX1's exclusivity does not. |
inputs
[ "Q154 gene-body occupancy matrix, 19,402 x 133", "Janssens 2024 Supplementary Data 2" ]
class members gate 2
[ "ELF1", "GNAQ", "H3C15", "IRX1", "PAN3", "SPAG7" ]
gate sweep class sizes — the numbers behind the chart
{
"1.25": 8,
"1.5": 6,
"1.75": 9,
"2.0": 6,
"2.5": 6,
"3.0": 3
}members at every gate 1 5 to 3 0
[ "GNAQ", "PAN3" ]
measurements
[
{
"quantity": "margin_to_the_nearest_non_index_sample",
"subject": "IRX1 gene body: ratio of the 2.0x gate to the highest non-AFF1 min-of-two-replicates share",
"value": 7.9,
"units": "fold",
"n": 18,
"universe": "the 18 non-AFF1 GSE252378 samples; highest is A73051 at 0.252x an average promoter",
"null": 1,
"note": "the other five class members have margins of 1.01x (H3C15), 1.11x (SPAG7), 1.25x (ELF1), 1.50x (PAN3) and 2.20x (GNAQ)"
},
{
"quantity": "n_class_members_confirmed_exclusive_by_the_source_pipeline",
"subject": "the six gene-body AFF1-exclusive recurrent genes, looked up in Janssens 2024 Supplementary Data 2",
"value": 1,
"units": "genes of 6",
"n": 6,
"universe": "the authors' 1,091-gene per-sample oncoprotein-target table over the same 26 samples",
"null": null,
"note": "1 confirmed (IRX1), 3 refuted (ELF1, GNAQ, PAN3), 2 never measured (H3C15, SPAG7)"
}
]- If a partner-specific target list were drawn from this class, four of the six entries would be genes the data's own authors call in other partners - and PAN3 is a canonical pan-KMT2A-fusion target. A therapy selected on that list would be aimed at general fusion biology while being sold as AFF1-specific.
- GNAQ is occupied in 5 of 8 AFF1 with a 2.2-fold margin and is refuted only by the authors' calls in four non-AFF1 samples. It is the second-best candidate in this class and has no entry in this project. What is GNAQ's role in a KMT2A-rearranged blast?
- H3C15 is a 518 bp gene body and its nearest non-AFF1 sample misses the gate by 1.4%. Should gene bodies below about 2 kb - the promoter window width - be excluded from a gene-body analysis entirely, and how many of the 19,402 are they?
- PAN3 is 38 kb from FLT3 and the authors' PC1 feature table treats PAN3 and FLT1 as one interval. Is the 13q12 signal one domain spanning FLT3-PAN3-FLT1, and does it behave differently in AFF1 than in other partners as a domain rather than as three genes?
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
The gene-body AFF1-exclusive recurrent class at a 2.0x gate has six members - ELF1, GNAQ, H3C15, IRX1, PAN3, SPAG7 - whose highest non-AFF1 min-of-two-replicates shares are 1.605, 0.911, 1.973, 0.252, 1.337 and 1.807; re-scoring at gates 1.25, 1.5, 1.75, 2.0, 2.5 and 3.0 gives class sizes 8, 6, 9, 6, 6, 3 with only GNAQ and PAN3 members throughout, and IRX1 a member from 1.25 to 2.0.
Q154 gene-body class membership, margins and threshold sweep · Iqbal Lab oncofusion knowledge graph, beat Q154 · this project 2026What this evidence cannot show
The margin is computed as the largest min-of-two-KMT2A-N-replicates share among the 18 non-AFF1 samples, so it ignores the per-library floor, which never binds at this scale anyway. H3C15 is a 518 bp gene body - the shortest of the six by an order of magnitude - so its share is the noisiest and its 1.4% margin should be read as no margin at all.
Evidence that argues against this (1)
- A threshold sweep that changes class membership is not by itself evidence that the 2.0x gate is wrong. Every threshold produces a different set, and the pre-registered gate was 2.0 because that is q147's and q152's. What the sweep shows is that no statement about the class SET should be made without the gate attached - not that a particular member is spurious. Iqbal Lab oncofusion knowledge graph et al. 2026
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
In Janssens et al.'s published per-sample oncoprotein-target table, ELF1 is called in p160, p186 and p214, GNAQ in p186, TB971992, TB995291 and A95941, and PAN3 in p160, p186, p214, TB995291 and A95941 - all non-AFF1 - while H3C15 and SPAG7 appear nowhere in the table and IRX1's row is exactly ['p279','p318','p454'].
MLL oncoprotein levels influence leukemia lineage identities · Janssens DH, Duran M, Otto DJ, Wu W, Xu Y, Kirkey D, Mullighan CG, Yi JS, Meshinchi S, Sarthy JF, Ahmad K, Henikoff S · Nature Communications 15:9341 2024What this evidence cannot show
Read from Supplementary Data 2 (41467_2024_53399_MOESM4_ESM.xlsx) by direct row lookup. H3C15 is the current HGNC symbol; the table predates the histone renaming and uses HIST-era symbols, so H3C15's absence was also checked under HIST2H3D and is still absent. SPAG7 has no alias risk.
Evidence that argues against this (1)
- The authors' caller is more sensitive at these loci, not necessarily more correct. A peak caller that nominates a site in a non-AFF1 sample at low signal is making a call this project's fold gate deliberately refuses. Which instrument is right at ELF1, GNAQ and PAN3 is not settled by either of them. Janssens DH et al. 2024
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
PAN3 is already on this project's own list of canonical KMT2A-fusion program genes, recovered by the q143 filter-free promoter ranking alongside HOXA3/7/9/10, MEIS1, MYB, CDK6 and JMJD1C - so a gene known to be a general KMT2A-fusion target appears here as AFF1-exclusive.
Q143 - a filter-free promoter ranking recovers forty of the fifty published oncoprotein target genes · Iqbal Lab oncofusion knowledge graph, beat Q143 · this project 2026What this evidence cannot show
PAN3 appears in that entry inside a gene list and has never had an entry or a characterisation in this project. Its presence in a recovered-target list is evidence that it is a KMT2A-fusion target, not that it is a general one across partners - though the authors' table, which calls it in ENL and AF9 samples, is.
Evidence that argues against this (1)
- PAN3 sits 38 kb downstream of FLT3 on 13q12 and 5 kb upstream of FLT1; the authors' own PC1 feature table carries the interval as 'PAN3, FLT1'. A gene-body call at PAN3 may be reading a domain that belongs to the FLT3 locus rather than to PAN3, and FLT3's own gene body is occupied in only 1 of 27 here, which argues against but does not exclude it. Janssens DH et al. 2024
inside the shared universe IRX1 is the only survivor of the four
IRX1 is the only member both above background on N/C and exclusive in the authors caller
IRX1 is the weakest member on independence, the criteria never applied to it
the class is significant as a count and mostly threshold-driven as a set: 5 of 6 members have a non-index sample within a factor of 2.2 of the gate and 3 are called in non-AFF1 samples by the deposit paper's own caller
- related to →IRX1 — Iroquois homeobox 1, the gene that names the HOXA-negative half of infant t(4;11) ALL
IRX1 is the only one of the six with a 7.9-fold margin and an independent confirmation
the margin and cross-instrument audit that motivates the model
the table the cross-instrument audit reads