AFF1-exclusivity describes 378 of 719 genes and 72% of them rest on one patient - and the MEIS1 comparison I used to call IRX1 indistinguishable was an artefact of scoring MEIS1 with the weaker instrument
also called q169 patient-support distribution · exclusivity is the majority category · MEIS1 has 6 patients not 2 · IRX1 is 60.5th percentile overall
Last cycle I said IRX1 could not be told apart from MEIS1, a gene nobody doubts. That was wrong: I had measured MEIS1 with this project's own weak method, which found it in 2 patients when the published method finds it in 6. But the bigger finding is that being 'unique to this fusion' is not special - more than half the genes in the published table have it, and most of those were seen in a single patient.
The same thing, for a scientist
Cycle 152 concluded NEITHER on IRX1 and named the fix: *"replace the three hand-chosen comparators with every gene in the authors' 1,091-gene table called in at least one AFF1 sample, and place IRX1's patient count in that distribution."* Running it **corrects that cycle and reframes five before it.** 🔴 **First correction: my MEIS1 comparison was an artefact of the instrument.** I scored MEIS1 with **this project's gene-body rule**, which gave 2 samples / 2 patients - the same as IRX1 - and concluded the criteria could not discriminate. In the **authors' own caller**, MEIS1 has **7 samples / 6 patients**. This project's rule recovers only **20.96%** of the authors' calls, and that insensitivity **flattened a 99.9th-percentile gene down to IRX1's level**. The comparison was between two poorly-measured genes, which is the objection the cycle-152 entry recorded in its own counter-evidence and I did not act on [E1]. On the authors' instrument, across the **719** genes called in at least one KMT2A::AFF1 sample [E1]: | gene | samples | patients | percentile | non-AFF1 calls | exclusive | |---|---:|---:|---:|---:|---| | GNAQ | 8 | 7 | 100.0% | 8 | no | | MEIS1 | 7 | **6** | 99.9% | 15 | no | | PROM1 | 7 | 6 | 99.9% | 10 | no | | PAN3 | 7 | 6 | 99.9% | 8 | no | | ELF1 | 7 | 6 | 99.9% | 5 | no | | AFF1 | 6 | 5 | 90.4% | 4 | no | | HOXA9 | 3 | 3 | 75.9% | 16 | no | | **IRX1** | 3 | **2** | **60.5%** | **0** | **YES** | | IRX2 | 1 | 1 | 46.7% | 0 | YES | | EPHA7 | 1 | 1 | 46.7% | 0 | YES | 🟠 **So IRX1 sits below every canonical target on patient support - but none of them is exclusive.** MEIS1 is called in 15 non-AFF1 samples, HOXA9 in 16. They are pan-KMT2A targets with broad support; IRX1 is an exclusive gene with narrow support. **The two are not competing on the same axis**, and cycle 152 compared them as though they were [E1]. 🔴 **Second, and this is the finding that reframes the line: AFF1-exclusivity is the majority category.** Of the 719 genes called in any AFF1 sample, **378 (52.6%) are called in zero non-AFF1 samples.** And within that exclusive set [E2]: | independent patients | genes | share | |---:|---:|---:| | **1** | **273** | **72.2%** | | 2 | 58 | 15.3% | | 3 | 36 | 9.5% | | 4 | 10 | 2.6% | | 5 | 1 | 0.3% | **The median AFF1-exclusive gene rests on ONE patient.** Exclusivity is not a distinction here - it is what a gene gets for being called once, because a single call in an AFF1 sample is trivially zero-in-non-AFF1. **Half the table has the property this project spent five cycles treating as remarkable** [E2]. 🟢 **The one thing that survives, stated precisely.** Within the exclusive set, IRX1's 2 patients put it at the **87.6th percentile** - better supported than 87.6% of the 378. So IRX1 is a well-supported member of a category that is mostly noise. **That is a much weaker claim than "partner-specific address" and a much stronger one than "indistinguishable from MEIS1".** Both of my previous framings were wrong, in opposite directions.
What this does to the exclusive-recurrent framing
🔴 The recurrence requirement was doing all the work and nobody said so. If 72% of exclusive genes have one patient, then demanding >=3 samples is not an additional filter on an interesting set - it is the only filter, and it is selecting against the single-call artefacts that make up most of the category. The permutation nulls in q152/q154/q158 are still valid; what changes is the reading of what they were testing.
🔵 It also explains a result that looked strange at the time. q154 found the singleton class (>=1 of 8, 0 of 18) at 63 observed against a null of 73.9 - below chance. That is exactly what this distribution predicts: singleton exclusivity is so common that the permutation reproduces it easily. The two findings are the same fact seen from two directions, and neither entry noticed the other.
🟠 And the comparator choice in cycle 152 was worse than arbitrary - it was anti-correlated with the property being tested. I picked canonical targets, which are by definition broadly expressed across partners, and then compared them on exclusivity-adjacent support. A fair comparator set was always the exclusive genes, and it was one join away.
🔵 What would still be worth knowing: whether the 58 two-patient exclusive genes have anything in common with IRX1 - lineage, expression, chromatin state. If IRX1 is unremarkable among them the line is finished; if it separates, that is the first positive discrimination in several cycles. That is one clustering away and is not run.
| script | analyses/q169-where-does-irx1-sit-in-the-patient-support-distribution/run.py |
|---|---|
| status | run |
| instrument | the authors' own call table, deliberately, not this project's gene-body rule |
| answers open question from | ana-q167 / cycle-152 king log |
| n genes table | 1091 |
| n called in any aff1 | 719 |
| median patients all | 2 |
| corrects | ana-q167 / cycle 152, which compared IRX1 to MEIS1 using this project's rule and concluded they were indistinguishable; MEIS1 has 6 patients on the authors' caller |
| explains | q154's singleton-class result of 63 observed against a null of 73.9, below chance - singleton exclusivity is common enough that a permutation reproduces it |
patient distribution all — the numbers behind the chart
{
"1": 336,
"2": 99,
"3": 111,
"4": 70,
"5": 34,
"6": 68,
"7": 1
}named genes — the numbers behind the chart
{
"IRX1": {
"samples": 3,
"patients": 2,
"pctile": 60.5,
"non_aff1": 0,
"exclusive": true
},
"MEIS1": {
"samples": 7,
"patients": 6,
"pctile": 99.9,
"non_aff1": 15,
"exclusive": false
},
"PROM1": {
"samples": 7,
"patients": 6,
"pctile": 99.9,
"non_aff1": 10,
"exclusive": false
},
"HOXA9": {
"samples": 3,
"patients": 3,
"pctile": 75.9,
"non_aff1": 16,
"exclusive": false
},
"ELF1": {
"samples": 7,
"patients": 6,
"pctile": 99.9,
"non_aff1": 5,
"exclusive": false
},
"GNAQ": {
"samples": 8,
"patients": 7,
"pctile": 100,
"non_aff1": 8,
"exclusive": false
},
"PAN3": {
"samples": 7,
"patients": 6,
"pctile": 99.9,
"non_aff1": 8,
"exclusive": false
},
"AFF1": {
"samples": 6,
"patients": 5,
"pctile": 90.4,
"non_aff1": 4,
"exclusive": false
},
"IRX2": {
"samples": 1,
"patients": 1,
"pctile": 46.7,
"non_aff1": 0,
"exclusive": true
},
"EPHA7": {
"samples": 1,
"patients": 1,
"pctile": 46.7,
"non_aff1": 0,
"exclusive": true
}
}exclusive set
{
"n": 378,
"share_of_called": 52.6,
"median_patients": 1,
"distribution": {
"1": 273,
"2": 58,
"3": 36,
"4": 10,
"5": 1
},
"one_patient_share": 72.2,
"irx1_percentile": 87.6
}measurements
[
{
"quantity": "genes called in at least one AFF1 sample that are AFF1-exclusive",
"value": 52.6,
"unit": "percent (378 of 719)",
"subject": "zero calls among the 18 non-AFF1 samples, in the authors' own target table",
"universe": "72.2% of those 378 rest on a single independent patient; the median exclusive gene has one"
},
{
"quantity": "IRX1's percentile for independent-patient support among AFF1-exclusive genes",
"value": 87.6,
"unit": "percentile",
"subject": "2 patients against a set whose median is 1; 60.5th percentile among all 719 AFF1-called genes",
"universe": "MEIS1, PROM1, ELF1, GNAQ and PAN3 all sit at 99.9-100th on patient support but none is exclusive"
}
]- Cluster the 58 two-patient exclusive genes against IRX1 on lineage, expression and chromatin state. If IRX1 is unremarkable among them the line is finished; if it separates, that is the first positive discrimination in several cycles. One clustering away, not run.
- Permute the arm assignment to find how much of the 52.6% exclusive rate is structural. With 8 index samples against 18 comparators a single call is far more likely to be exclusive than not, and that ratio was held fixed here.
- Re-examine every comparison in this project that used the gene-body rule to score a gene whose authors'-table value was available. MEIS1 was flattened from 6 patients to 2 by a rule with 20.96% recall, and that instrument has been used for cross-gene comparisons elsewhere.
- Decide what to do with ana-q167. Its arithmetic is correct and its conclusion rests on a comparison this entry refutes; it should carry the correction in its prose, not only an inbound edge.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.journal-article
Scored on Janssens et al.'s own per-sample target table, 719 of the 1,091 genes are called in at least one of the 8 KMT2A::AFF1 samples, and independent-patient support across them has a median of 2 with 46.7% of genes resting on a single patient. IRX1 is called in 3 samples spanning 2 patients (60.5th percentile) with zero non-AFF1 calls; MEIS1 is called in 7 samples spanning 6 patients (99.9th percentile) with 15 non-AFF1 calls; PROM1 7/6 with 10; HOXA9 3/3 with 16; ELF1 7/6, GNAQ 8/7 and PAN3 7/6. This project's own gene-body rule, which recovers 20.96% of the authors' calls, had scored MEIS1 at 2 samples / 2 patients.
What this evidence cannot show
Patient collapse uses this project's three recorded replicate pairs; the authors' table uses hyphenated TB identifiers which were mapped to this project's form, and a mis-mapping would move individual genes. Percentiles are computed over genes called in at least one AFF1 sample, so they describe position within the authors' called set, not within the genome. The 20.96% recall figure is carried from a previous beat and was not re-derived here.
Evidence that argues against this (2)
- Using the authors' caller as the arbiter assumes it is the better instrument, which is argued from its higher call count rather than demonstrated. A more sensitive caller also makes more false positives, and MEIS1 appearing in 15 non-AFF1 samples could be sensitivity or could be noise - in which case this project's sparser rule was not 'flattening' MEIS1 so much as disagreeing with it.
- IRX1 sitting below the canonical targets on patient support is close to tautological given how it was found. It was noticed because it appeared in a small distinctive set of samples; a gene discovered that way cannot also have broad support, so the comparison penalises it for the property that made it visible.
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.journal-article
Of the 719 genes called in at least one KMT2A::AFF1 sample in the authors' table, 378 (52.6%) are called in zero non-AFF1 samples and are therefore AFF1-exclusive by the definition this project has used; within that exclusive set 273 genes (72.2%) rest on a single independent patient, 58 (15.3%) on two, 36 on three, 10 on four and 1 on five, giving a median of one patient. IRX1's two patients place it at the 87.6th percentile within the exclusive set.
What this evidence cannot show
'Exclusive' here means zero calls among the non-AFF1 samples of this deposit, which is 18 samples - it is not a statement about KMT2A::AFF1 specificity in general. The 52.6% figure is a property of the authors' caller and this deposit's composition and would change with a different sensitivity threshold or a different partner mix. Genes called in zero AFF1 samples are excluded from the denominator throughout, so this says nothing about the 372 genes in the table that no AFF1 sample carries.
Evidence that argues against this (2)
- That most exclusive genes are single-patient does not make exclusivity uninformative for the genes that are not. The argument proves that exclusivity ALONE is cheap, which the recurrence requirement already conceded by existing; a reader could reasonably say this formalises a caveat rather than overturning anything.
- The 52.6% figure may be inflated by the arm sizes rather than by biology: with 8 AFF1 samples against 18 comparators, a gene called once has 18 chances to be excluded and only 8 to be included, so the base rate of 'exclusive' depends on a ratio this analysis holds fixed and does not vary. A permutation over arm assignment would settle how much of the 52.6% is structural.
MEIS1 has 6 patients on the authors caller; the indistinguishability was an instrument artefact
explains why singleton exclusivity sits below its null: it is the majority category
- related to →IRX1 — Iroquois homeobox 1, the gene that names the HOXA-negative half of infant t(4;11) ALL
60.5th percentile overall, 87.6th within the exclusive set whose median is one patient
runs the clustering that entry named as the last route to a positive
q169 showed AFF1-exclusivity is the majority category; q171 shows the two-instrument narrowing of it is not discriminating either