PAF1 and MED29 return the identical occupancy call in all 27 GSE252378 samples, because 86.5% of their promoter windows is the same DNA
also called PAF1 MED29 shared promoter · PAF1 promoter attribution
PAF1, a protein this project cares about, has another gene starting 271 letters away facing the other direction. When the fusion-protein signal is measured at each of them across 27 patient samples, the two genes give the same answer every single time - because the measurement is mostly reading the same stretch of DNA twice.
The same thing, for a scientist
`PAF1` (`chr19:39,391,106`, minus strand) and `MED29` (`chr19:39,391,377`, plus strand) are a head-to-head divergent pair **271 bp** apart, so their 2 kb windows share **1,729 of 2,000 bp = 86.5%** [E1]. Both are rows of this project's promoter table and of the 133-library matrix, so the q147 occupancy rule can be applied to each. Applied verbatim — within-sample share ≥ 2.0 in **both** KMT2A-N antibodies **and** above that sample's 95th-percentile width-matched floor — **PAF1 is occupied in 11 of 27 samples, MED29 is occupied in 11 of 27, and the two calls disagree in 0 of 27** [E1]. In **19 of the 27** samples the two shares are identical to three decimal places in at least one KMT2A-N arm, and in **8** they are identical in both. **So the sentence "PAF1's promoter is fusion-occupied in 11 of 27 KMT2A-rearranged samples" is, in this data, exactly and equally the sentence "MED29's promoter is fusion-occupied in 11 of 27"** — and PAF1 is a protein this project has an entry for and a mechanism around, while MED29 is a Mediator subunit it does not. This is the IRX2 problem in a case where **both members happen to be in the table**, which is why it can be demonstrated rather than argued [E1]. It is also a rare piece of good news for the corpus: **no such claim has been written.** Neither PAF1 nor MED29 was in q147's locus list and no promoter-scale occupancy result for either exists in this graph [E2].
The 27 calls
Occupied in both: A73051 (AF10), A82485 (AFDN), A89691 (AFDN), A93949 (AF9), A95941 (AF9), R3595 (EPS15), TB051939 (AFF1), TB920488 (AFF1), TB971992 (ENL), p179 (AFF1), p318 (AFF1) — 11 of 27, and the same 11 for both genes.
Not occupied in both: the other 16, including TB986292 where the KMT2A-N2 share reaches 24.5 at both rows and the N1 arm at 1.83 keeps both below the two-antibody gate.
The pair is not a degenerate case where the two rows are byte-identical: A80956 reads PAF1 2.860 / MED29 3.590 in N1, and A93692 reads 1.582 / 1.598. The 271 bp of unique sequence on each side does carry different coverage. It never carries enough to change a call.
Why this is the demonstration and IRX2 is only the illustration
At IRX2 the neighbour is in the table under its previous symbol C5orf38, so the comparison can be made, and the two rows are identical to four decimal places in 18 of 27 samples with neither ever occupied. That is consistent with the attribution problem but it is also consistent with there simply being no signal at that locus — a pair of zeros agreeing is not evidence of anything.
PAF1/MED29 is the case where the rule fires, eleven times, and fires identically. An agreement among positives is the informative version, and it is the version that shows the measurement cannot name its gene.
The general form of this, measured over every overlapping pair in the table, is in ana-q153-a-promoter-occupancy-call-agrees-with-its-overlapping-neighbour-91-percent-of-the-time.
| script | analyses/q153-can-an-irx2-promoter-measurement-be-attributed/q153_discriminate.py |
|---|---|
| status | run |
| result summary | PAF1 11/27, MED29 11/27, 0 disagreements, 86.5% shared window. A worked demonstration that a 2 kb promoter call cannot name its gene when the window is shared. |
occupied samples
[ "A73051", "A82485", "A89691", "A93949", "A95941", "R3595", "TB051939", "TB920488", "TB971992", "p179", "p318" ]
measurements — the numbers behind the chart
[
{
"quantity": "n_samples_occupied",
"subject": "PAF1 2 kb promoter, KMT2A-N1 and N2, GSE252378",
"value": 11,
"units": "samples",
"n": 27,
"universe": "27 KMT2A-fusion samples, rule >=2.0x within-sample share in BOTH KMT2A-N antibodies AND above the 8-locus 95th-percentile floor",
"null": 2.73,
"note": "null is the expectation for an arbitrary promoter under the same rule, from finding-q151-the-two-antibody-2x-occupancy-rule-calls-a-tenth-of-all-promoters-occupied"
},
{
"quantity": "n_samples_occupied",
"subject": "MED29 2 kb promoter, same libraries and same rule",
"value": 11,
"units": "samples",
"n": 27,
"universe": "identical to the PAF1 universe; the two windows share 1,729 of 2,000 bp",
"null": 2.73,
"note": "the SAME eleven samples; this is the point of the entry"
},
{
"quantity": "n_disagreeing_calls",
"subject": "PAF1 versus MED29 occupancy call, per sample",
"value": 0,
"units": "samples",
"n": 27,
"universe": "27 paired calls; rule of three gives a true disagreement rate below 11.1% with 95% confidence",
"null": null,
"note": "against a null of 95.7% disagreement for random promoter pairs containing a positive"
},
{
"quantity": "promoter_window_overlap_pct",
"subject": "PAF1 window against MED29 window",
"value": 86.45,
"units": "percent of a 2,000 bp window",
"n": 2,
"universe": "hg38 refGene TSSs chr19:39,391,106 (PAF1, minus) and chr19:39,391,377 (MED29, plus), 271 bp apart",
"null": 21.6,
"note": "null = fraction of all 19,404 project windows sharing >=50% with another gene's window"
}
]- PAF1 is a proposed point of intervention in KMT2A-fusion leukaemia. If an occupancy-based argument for targeting it is ever built on a 2 kb promoter readout in this deposit, the argument is 86.5% an argument about a Mediator subunit, and a PAF1-directed therapy designed from it would have been nominated by the wrong gene's DNA.
- Is the eleven-sample PAF1/MED29 signal a real fusion peak or a generic active-promoter signal? The rule calls a median 10.75% of all 19,404 promoters occupied per sample, so 11 of 27 at one locus is well above the 2.73 expectation but the entry makes no claim about which gene's biology it reflects - and cannot.
- Would a 500 bp window separate PAF1 from MED29? At 271 bp apart a 500 bp window would still share 46% and both TSSs would still be inside it, so probably not; the only scale that separates them is base resolution or gene body.
- How many other pairs in the table are BOTH in the matrix, so that this demonstration can be repeated? 1,563 pairs share >=50% of their window; the subset where both members carry signal is the reusable test set and has not been enumerated by locus.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
PAF1 (chr19:39,391,106, minus) and MED29 (chr19:39,391,377, plus) have 2 kb promoter windows sharing 1,729 of 2,000 bp (86.5%), and under the q147 two-antibody occupancy rule applied verbatim to the GSE252378 promoter matrix each is occupied in 11 of 27 samples with zero disagreements between them.
q153 promoter-call discriminability over overlapping window pairs, GSE252378 · Iqbal Lab oncofusion knowledge graph, beat Q153 · this project 2026What this evidence cannot show
Confidence is moderate, not established, for three reasons. The occupancy rule is reimplemented here rather than re-run from q147's script: it reproduces the FLOOR-B definition (95th percentile of ALB, RHO, MYH7, MYOD1, NEUROD1, KRT14, GFAP, INS) and the >=2.0-in-both-N-libraries gate, but it drops q147's depth-grade UNMEASURED branch, which only relabels zeros and cannot create a disagreement between two rows. The input matrix was asserted at load to be RAW mean coverage (column means span 0.9177-18.7951, 0 of 133 within 1% of 1.0) and converted to within-sample shares by dividing by the column mean, exactly as q147 does. n = 27 samples over 24 independent patients, so the calls are not fully independent. There is no spike-in in this deposit, so every value is a within-sample share and none of it separates low abundance from shallow library.
Evidence that argues against this (1)
- Zero disagreements in 27 comparisons does not prove the rule CANNOT separate them; with 11 positives and a 86.5% shared window, the expected number of disagreements is small under any model. The informative comparison is against the null of random promoter pairs, where the same rule disagrees on 95.7% of comparisons that contain a positive - that null, not this pair on its own, is what makes the result interpretable. Iqbal Lab oncofusion knowledge graph et al. 2026
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
No promoter-scale occupancy result for PAF1 or MED29 exists anywhere in this graph: neither symbol appears in q147's locus list (POS_CONTROLS, NEG_CONTROLS, AXIS, IRXF, SECOND, LINEAGE_CTRL, UBIQ, FLAT_CONTROL) and no entry states one.
q147 analyze147.py locus list and a repository-wide check of promoter-scale occupancy claims · Iqbal Lab oncofusion knowledge graph, beats Q147 and Q153 · this project 2026What this evidence cannot show
Bounded by reading q147's hardcoded locus list and by grep over the node files for the two symbols on 2026-09-24. It is a statement about this corpus, not about the literature: PAF1's role with KMT2A fusions is well documented elsewhere, and this claim is only that no PROMOTER-SCALE OCCUPANCY number for it exists here.
Evidence that argues against this (1)
- That no such claim has been written is luck rather than design - PAF1 is a core member of the complex this project reasons about, it is in the promoter matrix, and scoring it is a two-line change. The finding is therefore a warning about the next analysis, not a clean bill of health for the last one. Iqbal Lab oncofusion knowledge graph et al. 2026
PAF1's 2 kb promoter is called occupied in 11 of 27 GSE252378 samples and so is MED29's, in the same 11, from an 86.5% shared window
the single worked pair behind the aggregate 91.3% agreement, and the case where the rule actually fires rather than agreeing on zeros