POST HOC: against institution-matched, anchored comparators the same-patient pair has the LOWEST excess in all six factorial cells - which leans to q142, but was not pre-registered
When we compare the one patient's two samples only against other children treated at the same hospital, and always keep one of that patient's own samples on the other side of the comparison, the patient's two diseases share the FEWEST target genes of any comparison we can make. This test was not planned in advance, so it is a lead and not a result.
The same thing, for a scientist
The pre-registered test compared the same-patient pair to the distribution over 27 unrelated KMT2A::AFF1 sample pairs and returned NEITHER. A post-hoc refinement, motivated by the measured institution batch effect, restricts comparators to the three unrelated Texas Children's KMT2A::AFF1 samples (p179, p247, p454) and anchors every comparison on p279 or p318, giving six anchored comparisons in which depth, antibody lot and institution are held fixed on one side. On the excess over the activity-stratified null the same-patient pair is exceeded by all six comparators in **all six of six factorial cells** - a fully unanimous direction, with per-cell two-sided sign-test p = 0.031 in the three absolute-rule cells and 0.22-0.69 in the quantile cells. This is q142's direction. It is labelled POST HOC because the anchoring and the institution match were chosen after the pre-registered test had already returned its answer.
The primary cell in full
Excess over the activity-stratified null, U_pair universe, paper calling rule:
same-patient p279 | p318 1.329
Texas anchored p279 | p179 1.532
p318 | p179 1.560
p279 | p247 1.469
p318 | p247 1.406
p279 | p454 1.397
p318 | p454 1.361
Six of six above. Sign test 0/6, two-sided exact p = 0.031. The same in U_27 (0/6) and U_full (0/6); 1-2 of 6 in the quantile cells, still a minority.
Why anchoring matters and why it is not free
Comparing J(p279,p318) to a cohort median asks two questions at once - is this pair special, and are p279 and p318 unusual samples? Anchoring every comparison on p279 or p318 answers only the first, because the nuisance sample is on both sides. That is a genuine design improvement.
It is not free for two reasons. (1) The six anchored comparisons are not independent: they share two samples, so the sign test's nominal p = 0.031 overstates the evidence, and the six cells are not independent of each other either. (2) The choice of Texas-only comparators was made after seeing that the direction flipped between rules, which is exactly the situation in which a post-hoc restriction can be tuned until the preferred answer appears. The restriction is defensible on the independent batch measurement (p = 0.0016), not on its result - but it is still post-hoc and is recorded as such.
What would turn this into a result
A pre-registered replication with an institution-matched, anchored design. The obstacle is sample size: the paper profiles three lineage-switching B-ALLs, only two have a paired AML in the N/C arm, and one of those two (152985) calls zero promoters because it was taken during revumenib. n = 1 for this design in the entire public record. So the honest next step is not a re-analysis but a new deposit, and the experiment entry for it says so.
| script | analyses/q146-one-universe-jaccard/texas146.py |
|---|---|
| status | run |
| result summary | POST HOC. Institution-matched anchored comparison: same-patient excess is lower than all six Texas-matched anchored comparisons in all six factorial cells (0/6 in the three absolute-rule cells, sign p=0.031; 1-2/6 in the quantile cells). Leans to q142's direction. Not pre-registered. |
| prespecified | false |
| same patient X primary | 1.329 |
| sign test p primary | 0.03125 |
| cells unanimous lower | 6 |
inputs
[ "analyses/q146-one-universe-jaccard/results.json" ]
texas anchored X — the numbers behind the chart
{
"p279|p179": 1.532,
"p318|p179": 1.56,
"p279|p247": 1.469,
"p318|p247": 1.406,
"p279|p454": 1.397,
"p318|p454": 1.361
}- If a patient's own two lineages are the LEAST similar pair available, then a salvage target list carried over from diagnosis is worse than a list borrowed from another child at the same centre - the opposite of what personalised profiling assumes.
- Does the third lineage-switching sample the paper mentions have a deposited paired partner under an id this beat did not pair? If so this design has n=2 and a pre-registered test becomes possible without new samples.
- Is the anchored design's unanimity robust to dropping p247, the one pediatric rather than infant comparator? It is (0/4), but at n=4 the sign test floor is 0.125, so the answer is direction-only.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
p279, p318, p179, p247 and p454 are all Texas Children's Hospital KMT2A::AFF1 samples in the same deposit, which is what makes an institution-matched anchored comparison possible at all.
GSE252378: AutoCUT&RUN of KMT2A-rearranged primary leukemia samples (KMT2A N- and C-terminal antibodies, ENL, DOT1L, H3K4me3, IgG) · Janssens DH, et al. · NCBI Gene Expression Omnibus 2024What this evidence cannot show
p247 is described as a PEDIATRIC B-ALL, not an infant one, so one of the three comparators differs in age group; excluding it leaves two comparators and four anchored comparisons, where the direction is also unanimous but the sign test cannot fall below 0.125.
Evidence that argues against this (1)
- A larger institution-matched comparator set would have to come from outside this deposit, and no other deposit carries paired pre- and post-lineage-switch KMT2A oncoprotein profiling. Janssens DH et al. 2024
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The lineage-switching samples in this cohort are a small, specifically-selected subset - three of 36 profiled samples - which is the binding constraint on any design of this kind.
MLL oncoprotein levels influence leukemia lineage identities · Janssens DH, Duran M, Otto DJ, Wu W, Xu Y, Kirkey D, Mullighan CG, Yi JS, Meshinchi S, Sarthy JF, Ahmad K, Henikoff S · Nature Communications 15:9341 2024What this evidence cannot show
The abstract states three lineage-switching samples; the deposit's N/C arm contains two paired sets, and this beat did not establish what happened to the third pair's partner sample.
Evidence that argues against this (1)
- If the third lineage-switching sample's paired AML is deposited under a sample id this beat did not recognise as a pair, the design's n could be 2 rather than 1 - a possibility this beat could not close. Janssens DH et al. 2024
the post-hoc refinement of the headline comparison
the institution match is justified by the measured batch effect, not by its result
q142's direction, unanimous across all six factorial cells against matched comparators
the same six cells, scored against a different comparator set
the post-hoc result leans against this hypothesis and is recorded inside its falsifier
turns the post-hoc institution-matched result into a pre-registered one
the design the post-hoc analysis uses