q145's anti-correlation p = 0.0288 is a one-sided value; the two-sided exact p is 0.0576 and the rho reproduces exactly
A number this project has been quoting as a significance value was computed by counting results in one direction only. Counting both directions, which is the usual convention, gives 0.058 instead of 0.029 - just above the usual cut-off.
The same thing, for a scientist
`ana-q145-irx1-is-fusion-occupied-exactly-where-hoxa-is-not` reports *Spearman -0.714 at gene scale, exact permutation p = 0.0288 over all 40,320 permutations*, and carries that figure in `data.exact_permutation_p`. Re-derived this cycle from q145's own cached region values: - **rho = -0.71429 reproduces exactly**, and at gene-body scale it is **not tie-degenerate** (no two of the eight IRX1 gene-body values are equal, so the arbitrary-tie-break range is a single value) [E1]. - The **one-sided** count (permutations with rho <= observed) is **1,161/40,320 = 0.028795**, which is the reported figure to four places [E1]. - The **two-sided** count (|rho| >= |observed|) is **2,322/40,320 = 0.057589** [E1]. 🟠 **King recount (cycle 131):** this entry first reported **1,160 and 2,320**. Recounted in exact integer arithmetic - rho = -5/7 corresponds to a sum of squared rank differences of exactly 144, and **1,161 of the 40,320 permutations reach d2 >= 144** (237 of them sit exactly on it), with 1,161 in the mirror tail - so the counts are **1,161 and 2,322**, off by one per tail. Nothing about the verdict moves: **0.0288 one-sided, 0.0576 two-sided, which does not clear 0.05.** q145's own caveat field on that evidence item says *'n=8 bounds the smallest attainable **one-sided** p at 1/40320'*, so the beat knew which test it ran - but the claim sentence, the summary and the `data` field all say 'exact permutation p' with no sidedness, and that is the form the number travelled in. The direction was pre-specified (the published claim is mutual exclusivity), so a one-sided test is defensible; **the label is not.** The leave-one-of-the-pair-out values hold up better: -0.857 (two-sided p = 0.0238) and -0.821 (p = 0.0341), both clearing 0.05 two-sided [E1].
The full table, both sidednesses, at both scales
| set | n | rho | one-sided p | two-sided p |
|---|---|---|---|---|
| all 8 AFF1 records, gene scale | 8 | -0.7143 | 0.0288 | 0.0576 |
| drop p279 (pair) | 7 | -0.857 | 0.0119 | 0.0238 |
| drop p318 (pair) | 7 | -0.821 | 0.0171 | 0.0341 |
| all 8, 2 kb promoter scale | 8 | -0.488 | 0.114 | 0.227 |
So the headline claim's strength depends on which of three defensible choices is made, and the published form picked the most favourable of the three without saying so: one-sided, gene-body scale, all eight records.
The correction does not overturn q145's conclusion - the effect is the same size, the direction was pre-specified, and the pair-dropped values are two-sided significant. It changes what may be written next to it.
What to do with it
Per rule 96 half two, the outcome belongs inside the original entry rather than only here. The recommended amendment to ana-q145-irx1-is-fusion-occupied-exactly-where-hoxa-is-not, reported and not applied because that entry belongs to another beat:
- E1 claim sentence and
data.exact_permutation_p: change 'exact permutation p = 0.0288' to 'exact one-sided permutation p = 0.0288; two-sided 0.0576', and addexact_permutation_p_two_sided: 0.0576. - Citation that settles it: the re-derivation is from q145's own
results_regions.jsonvalues, recomputed inanalyses/q147-hoxa-irx1-across-partners/gene_scale147.py, which prints both sidednesses side by side for every correlation it computes. No external paper is needed; this is arithmetic on a committed file.
The generalisable rule: a permutation script should print both tails and label them, because the one-tailed number is exactly half the two-tailed one and there is nothing in either figure that says which it is.
| rho gene scale | -0.714286 |
|---|---|
| n permutations | 40320 |
| one sided count | 1161 |
| one sided p | 0.028795 |
| two sided count | 2322 |
| two sided p | 0.057589 |
| tie degenerate at gene scale | false |
| entry to amend | ana-q145-irx1-is-fusion-occupied-exactly-where-hoxa-is-not |
| field to amend | evidence[E1].claim and data.exact_permutation_p |
| counts method | exact integer arithmetic: rho = -5/7 <=> sum of squared rank differences = 144; 1161 permutations have d2 >= 144 (237 exactly at 144), 1161 in the mirror tail |
drop p279
{
"rho": -0.857,
"p_two_sided": 0.0238
}drop p318
{
"rho": -0.821,
"p_two_sided": 0.0341
}analyses
[ "ana-q147-irx1-fusion-occupancy-is-confined-to-kmt2a-aff1" ]
- How many other permutation p-values in this corpus are one-sided and unlabelled? The tell is a p that is exactly half a round two-sided value, and it is not visible from the number.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
Recomputed from q145's own cached region values: Spearman(HOXA9_HOXA10_20kb, IRX1_gene) over the eight KMT2A::AFF1 sample records is -0.714286, with 1,160 of 40,320 permutations at or below the observed value (one-sided p = 0.02877) and 2,320 at or above it in absolute value (two-sided p = 0.05754). The gene-scale values contain no exact ties, so the rho is not tie-sensitive at that scale. Pair-dropped: -0.857 (two-sided 0.0238) and -0.821 (two-sided 0.0341).
q147 re-analysis of GSE252378: HOXA and IRX1 oncoprotein occupancy across all six KMT2A fusion partners, 27 samples, 133 libraries (analyses/q147-hoxa-irx1-across-partners) · Iqbal Lab oncofusion knowledge graph, beat q147 · internal analysis over deposited GSE252378 bigWigs 2026What this evidence cannot show
This is a recomputation on a committed intermediate file, not a new measurement, so it inherits every limitation of q145's design - n = 8 records from 7 patients in one deposit - and adds none. The one-sided test is defensible on pre-specification grounds because the published expression claim predicts the direction; what is not defensible is reporting it unlabelled. Every number is a WITHIN-SAMPLE share: an interval's mean coverage divided by that same library's mean coverage over all 19,404 hg38 RefSeq promoters, so 1.0 = an average promoter in that library. GSE252378 has no spike-in, so no between-library absolute statement is licensed; the only cross-sample quantity used is a ratio of two loci measured inside one library, where the unknown per-library scale cancels.
Evidence that argues against this (1)
- q145's own caveat field states the one-sided bound explicitly, so the beat did not conceal the test it ran; the defect is one of labelling in the claim sentence and the structured data field, which are the parts that get quoted and machine-read. Iqbal Lab oncofusion knowledge graph et al. 2026