PLANNED: margin-ranked partner-specificity scan over all 6,049 occupied-anywhere promoters, with a depth control
also called q152 planned margin scan
A plan, not a result. Rank every gene switch by how far its KMT2A::AFF1 signal sits above the best non-AFF1 patient, shuffle the patient labels to get a fair comparison, and check whether the ranking finds known target genes better than simple counting does.
The same thing, for a scientist
Planned follow-up to `hyp-q152-the-separation-margin-not-the-recurrence-count-identifies-a-partner-specific-promoter`. Compute, for each of the 6,049 occupied-anywhere promoters, margin = min over occupied AFF1 samples of the min-of-two-replicate within-sample share, divided by the max over the 19 (or 17 untreated) non-AFF1 samples of the same statistic. Permute the 8-sample index arm 10,000 times holding the share matrix fixed and take the distribution of the top-k margins. Score recovery of this project's q143 published-oncoprotein-target list in the top 100 by margin versus the top 100 by recurrence. Regress margin on comparator-arm mean library coverage from C:/Temp/q144mag/counts and report the Spearman coefficient before any verdict.
What is pre-committed now
The three falsifiers in the hypothesis entry, the 10,000-permutation count, the top-100 comparison set (q143's recovered published targets), the depth covariate (#total_fragments from the 133 per-library count files already on disk) and the Spearman > 0.5 abandonment threshold. Recording them here means the scan cannot be re-specified after its result is seen.
Cost
No download. The share matrix (15 MB) and the per-library fragment counts are already on local disk; the permutation is 10,000 x 6,049 comparisons and runs in under a minute. This is the cheapest remaining test in this line of work.
| script | analyses/q152-the-aff1-only-recurrent-promoter-core/ (not yet written) |
|---|---|
| status | planned |
| result summary | Not yet run. Pre-committed specification and abandonment threshold recorded in this entry and in the parent hypothesis. |
| cost tier | low |
inputs
[ "C:/Temp/q147part/promoter_matrix.tsv", "C:/Temp/q144mag/counts (133 .tsv, #total_fragments)" ]
- Should the comparator arm for the margin be the 19 non-AFF1 samples or the 17 untreated ones? Q152 measured that the choice changes the AFF1-only count from 332 to 342; it will change every margin slightly and the decision should be made before the scan, not after.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
The 19,404 x 133 promoter coverage matrix and the 133 per-library fragment-count files needed for this scan are already present on local disk at C:/Temp/q147part/promoter_matrix.tsv and C:/Temp/q144mag/counts, so the analysis requires no download.
Q152 input inventory · Iqbal Lab oncofusion knowledge graph, beat Q152 · this project 2026What this evidence cannot show
The scratch directories are on the C: drive of one machine and are not versioned; if they are cleared the matrix must be rebuilt from the deposit, which is a bigWig download this project's disk budget currently forbids.
Evidence that argues against this (1)
- Pre-committing a specification does not make the scan informative. If the depth regression fires (Spearman > 0.5) the scan produces nothing, and this project has already established that no coverage-invariant measure exists in this deposit, so the prior on that outcome is not low. Iqbal Lab oncofusion knowledge graph et al. 2026
the pre-committed scan and its three falsifiers