The reciprocal-antibody ratio splits the six-gene AFF1-exclusive class in two: H3C15 and SPAG7 sit at or below background, the three most enriched are the three the authors' caller says are not exclusive, and IRX1 is the only member that is both
also called q162 N over C · H3C15 sits at background · SPAG7 is below background · IRX1 is enriched and exclusive · the IgG arm is sparsity here
The dataset has two antibodies against the same protein: one recognises the part kept in the fusion, the other the part lost from it. Comparing them tells you whether a site looks like the fusion or like the normal protein. Of the six genes this project calls fusion-specific, two show no fusion signal at all, and only one is both fusion-enriched and confirmed unique to this partner.
The same thing, for a scientist
q161 found that histone-cluster genes are **31% of every gene-body occupancy call** this project makes. This asks the obvious follow-up with the arms already on disk: are those loci **artefact**, **wild-type MLL1**, or **fusion**? KMT2A-N sees fusion plus wild-type; KMT2A-C sees wild-type only, because the C-terminus is not retained in any KMT2A fusion. 🔴 **The IgG arm cannot answer this, and the reason is sparsity, not signal.** IgG reads **0.000** at histone loci, which looks like a pristine control until you check the boring explanation: histone gene bodies have a **median length of 510 bp against 24,079 bp** for everything else, and IgG libraries run at **0.452x** the KMT2A-N1 depth. The **raw** IgG fragment count is zero at **60.8% of occupied histone loci** against **25.5%** of other occupied loci. **A zero at a short locus in a shallow library is sparsity.** The N/IgG ratio is not reported here; a first pass that did report it produced 3.5e9 from a 1e-9 pseudocount [E1]. q154 had already recorded the IgG arm as uninterpretable under this normalisation. 🟠 **On the axis that does work, histone clusters are indistinguishable from the rest of the occupied set.** Requiring real C-arm signal (C > 0.05), median **N/C = 1.775** for occupied histone loci against **1.769** for occupied non-histone loci, and paired across samples the histone set is lower in 15 of 26, **two-sided sign test p = 0.56** [E1]. **So they are not preferentially wild-type MLL1** - whatever makes them 31% of calls, it is not that. 🔵 **The per-gene result is the finding.** Median N/C across the 8 KMT2A::AFF1 samples, against an occupied-background median of **1.769** [E2]: | gene | median N/C | x background | exclusive in the authors' caller? | |---|---:|---:|---| | **GNAQ** | 16.69 | **9.44x** | **no** - called in 4 non-AFF1 samples | | **PAN3** | 12.69 | **7.17x** | **no** - called in 5 non-AFF1 samples | | **ELF1** | 8.94 | **5.05x** | **no** - called in 3 non-AFF1 samples | | **IRX1** | 4.80 | **2.71x** | **yes** - exactly p279, p318, p454 | | **H3C15** | 1.84 | **1.04x** | not in the authors' table | | **SPAG7** | 0.88 | **0.49x** | not in the authors' table | 🔴 **H3C15 sits exactly at background (1.04x) and SPAG7 sits below it (0.49x).** Neither shows any fusion-arm enrichment. **For H3C15 that is the third independent reason to distrust its membership**, and the three were found in three different beats without meeting: it is a histone-cluster gene (q161), it has the worst margin of the six at **1.01x** (q154), and it has no N/C enrichment (here). **SPAG7 below background is new** - more C-arm than N-arm material relative to the occupied set, which is what wild-type MLL1 looks like [E2]. 🟠 **And the deflating part, which I did not expect: N/C does not track exclusivity - in this set it runs against it.** The three *most* fusion-arm-enriched members are precisely the three the authors' independent caller **refutes** as AFF1-exclusive. **A gene can be strongly fusion-enriched and not partner-specific at all**, which is the distinction this project has repeatedly collapsed. 🟢 **What survives is a clean 2x2 with one cell occupied.** Of the six, **IRX1 is the only member that is both fusion-arm-enriched (2.71x background) and confirmed exclusive by an independent caller.** That is the first positive result for IRX1 in several cycles - and it sits alongside the standing findings that IRX1 is not individually significant (p 0.077-0.176) and cannot appear in the lineage-matched class.
What N/C does and does not mean
🟠 N/C > 1 is not 'fusion'. The KMT2A-N antibody sees the fusion and wild-type MLL1; the C antibody sees wild-type only. So an elevated N/C means more N-terminal than C-terminal material than the occupied background carries - consistent with fusion occupancy, but equally consistent with a truncated or differentially accessible wild-type epitope, or with the two antibodies simply having different efficiencies at that locus. The ratio is a screen, not an attribution.
🔵 Why the background matters more than the absolute value. Every occupied locus sits near N/C 1.77, so the denominator is not 1 and a gene at 1.8 is at background, not at 'slightly enriched'. Reporting raw N/C without that comparator is how H3C15's 1.84 could be read as a positive. I report the fold over the occupied background for exactly that reason.
🔴 Two genes are missing from the authors' comparison entirely. H3C15 and SPAG7 have no row in Janssens' target table, so 'not exclusive' is unmeasured, not refuted - a distinction this corpus has paid for before. Their low N/C is independent evidence, not a second opinion on the same measurement.
🟠 IRX1 loses a sample to the filter. Requiring C > 0.05 leaves IRX1 measurable in 7 of 8 AFF1 samples; the others are 8 of 8. The excluded sample is one where the C arm has essentially no signal, which inflates rather than deflates IRX1's ratio if included, so the filter is conservative for IRX1.
| script | analyses/q162-histone-clusters-artefact-wildtype-or-fusion/run.py |
|---|---|
| addendum script | analyses/q162-histone-clusters-artefact-wildtype-or-fusion/igg_check.py |
| status | run |
| background NC | 1.769 |
| new flag | SPAG7 is BELOW background at 0.49x - more C-arm than N-arm relative to the occupied set |
arms used
[ "KMT2A_N1", "KMT2A_N2", "KMT2A_C1", "KMT2A_C2", "IgG" ]
igg unusable
{
"histone_median_length_bp": 510,
"other_median_length_bp": 24079,
"igg_median_depth": 4570914,
"n1_median_depth": 10102155,
"depth_ratio": 0.452,
"raw_igg_zero_at_occupied_histone": [
525,
863,
"60.8%"
],
"raw_igg_zero_at_occupied_other": [
501,
1965,
"25.5%"
],
"first_pass_error": "a 1e-9 pseudocount turned IgG=0 into N/IgG = 3.5e9; not reported"
}NC histone vs other
{
"histone": 1.775,
"other": 1.769,
"histone_lower_in": 15,
"n_samples": 26,
"sign_test_p": 0.5572,
"verdict": "indistinguishable; histone loci are not preferentially wild-type"
}per gene NC in AFF1 — the numbers behind the chart
{
"GNAQ": {
"median_NC": 16.694,
"fold": 9.44,
"authors_exclusive": false
},
"PAN3": {
"median_NC": 12.685,
"fold": 7.17,
"authors_exclusive": false
},
"ELF1": {
"median_NC": 8.936,
"fold": 5.05,
"authors_exclusive": false
},
"IRX1": {
"median_NC": 4.802,
"fold": 2.71,
"authors_exclusive": true,
"n_samples_with_C": 7
},
"H3C15": {
"median_NC": 1.839,
"fold": 1.04,
"authors_exclusive": "unmeasured"
},
"SPAG7": {
"median_NC": 0.875,
"fold": 0.49,
"authors_exclusive": "unmeasured"
}
}H3C15 third strike
[ "histone-cluster gene (q161)", "worst margin of the six, 1.01x (q154)", "N/C at background, 1.04x (q162)" ]
measurements
[
{
"quantity": "H3C15's fusion-arm enrichment relative to the occupied background",
"value": 1.04,
"unit": "fold (median N/C 1.839 against a background of 1.769)",
"subject": "a member of the six-gene AFF1-exclusive class, measured across 8 KMT2A::AFF1 samples",
"universe": "the other five members span 0.49x (SPAG7) to 9.44x (GNAQ); H3C15 shows no enrichment at all"
},
{
"quantity": "occupied histone-cluster loci where the raw IgG fragment count is zero",
"value": 60.8,
"unit": "percent (525 of 863)",
"subject": "why the IgG control cannot discriminate here - median locus length 510 bp, IgG depth 0.452x the N arm",
"universe": "25.5% (501 of 1,965) at occupied non-histone loci, whose median length is 24,079 bp"
}
]- Re-examine SPAG7. It is below the occupied background on N/C (0.49x), which is what wild-type MLL1 looks like, and it is unmeasured in the authors' table - so it has no independent support at all and is now the weakest member of the class after H3C15.
- Decide whether the occupied background's own N/C of 1.77 is acceptable. In a fusion whose C-terminus is absent, a typical occupied locus carrying more than half as much C-arm as N-arm signal suggests wild-type MLL1 dominates what this rule calls occupancy - which would be a larger finding than anything about individual genes.
- Vary the C > 0.05 threshold. It was chosen to avoid a divide-by-zero and never swept, and this corpus has recorded that a verdict tracking its threshold is a measurement of the ruler.
- Test whether N/C and exclusivity are genuinely anti-correlated on a larger gene set. Six genes cannot establish it, but if it holds it would mean the two axes measure unrelated things and that this project has been conflating them.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.journal-article
The IgG arm cannot discriminate at histone-cluster loci because of sparsity: histone-cluster gene bodies have a median length of 510 bp against 24,079 bp for all other gene bodies, IgG libraries have a median depth of 4,570,914 fragments against 10,102,155 for KMT2A-N1 (ratio 0.452), and the raw IgG fragment count is zero at 525 of 863 occupied histone-cluster loci (60.8%) against 501 of 1,965 occupied other loci (25.5%). On the N/C axis, computed only where the C arm carries signal above 0.05, occupied histone loci give a median N/C of 1.775 against 1.769 for occupied non-histone loci, with the histone set lower in 15 of 26 samples (two-sided sign test p = 0.56).
What this evidence cannot show
The length and depth figures explain why IgG zeros are uninformative but do not prove the histone signal is genuine - a short locus can be both sparsely covered in IgG and artefactually covered in the specific arms. The C > 0.05 threshold is arbitrary and was chosen to avoid the same divide-by-zero that invalidated the N/IgG pass; it was not varied. Histone-cluster membership is a symbol-prefix regex, not a curated locus list. The sign test treats samples as exchangeable and ignores the very different call counts per sample.
Evidence that argues against this (2)
- Showing that histone loci are not distinguishable from the rest of the occupied set on N/C is equally consistent with the whole occupied set being contaminated. If the rule's background is itself largely artefact, then 'histone loci look like the background' is not reassurance about histone loci - it is an indictment of the background, and this analysis cannot tell the two readings apart.
- A median N/C of ~1.77 across ALL occupied loci is itself suspicious. If the occupied set were dominated by genuine fusion sites, a much larger separation from the C arm would be expected in a fusion where the C-terminus is absent. That the typical occupied locus carries more than half as much C-arm as N-arm signal suggests wild-type MLL1 contributes heavily to what this rule calls occupancy.
- E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.journal-article
Median N/C across the 8 KMT2A::AFF1 samples, against an occupied-background median of 1.769, is 16.69 for GNAQ (9.44x), 12.69 for PAN3 (7.17x), 8.94 for ELF1 (5.05x), 4.80 for IRX1 (2.71x), 1.84 for H3C15 (1.04x) and 0.88 for SPAG7 (0.49x); the three most enriched - GNAQ, PAN3 and ELF1 - are precisely the three whose AFF1-exclusivity is refuted by calls in non-AFF1 samples in Janssens' own target table, while IRX1 is the only member of the six that is both above background on N/C and confirmed exclusive by that independent caller.
What this evidence cannot show
n = 8 samples and these are medians of per-sample ratios, so a single library with a weak C arm moves a gene several fold; no confidence intervals are computed. IRX1 is measurable in 7 of 8 samples under the C > 0.05 filter against 8 of 8 for the others. H3C15 and SPAG7 have no row in the authors' table, so their exclusivity is UNMEASURED rather than refuted. The anti-correlation between N/C and exclusivity is observed across six genes, which is far too few to call a relationship.
Evidence that argues against this (2)
- The cleanest reading of the anti-correlation is unflattering to IRX1 rather than flattering: if fusion-arm enrichment and partner-exclusivity point in opposite directions across these six genes, then whatever exclusivity measures may be unrelated to fusion occupancy altogether, and IRX1's position in the one occupied cell of the 2x2 could be where a gene lands by chance when the two axes are independent and the sample is six.
- IRX1's 2.71x is the WEAKEST enrichment among the four genes above background, and it is the one with a missing sample. Calling it 'the only member that is both' rewards it for a mediocre score on one axis and a good score on another; a reader who weighted N/C more heavily would rank it fourth of six.
answers q161s named next step with the C and IgG arms
H3C15 at 1.04x and SPAG7 at 0.49x of background show no fusion-arm enrichment
IRX1 is the only member both above background on N/C and exclusive in the authors caller
independently confirms the IgG arm is unusable, here by locus length and depth sparsity
uses the same reciprocal arm per gene rather than per class
answers the N/C 1.77 question q162 raised, with a set overlap rather than a ratio