The excess overlap survives a structure-respecting null in both beats - permutation p = 0.0005 in all six cells - but every published excess figure shrinks 2.4- to 4-fold
The finding that two samples share more target genes than chance is real and holds up under a much tougher test. What does not hold up is how big the effect was said to be: it is two to four times smaller than reported.
The same thing, for a scientist
The concern that motivated this beat was that the published excess-over-chance figures might be entirely an artefact of promoter non-exchangeability. They are not. In all six factorial cells the observed p279/p318 Jaccard exceeds the 2,000-draw activity-stratified null with one-sided permutation p = 0.0005 - the smallest value 2,000 draws can return - and the null's 95th percentile never reaches the observation (e.g. 0.254 against 0.328 in the primary cell). The excess is between 1.31x and 4.46x depending on the rule. What does not survive is the MAGNITUDE as published: q144's 5.8x becomes 1.46x (3.97-fold shrinkage) and q142's unstated 6.44x becomes 2.69x (2.4-fold). The coordinator's prediction that every excess figure in both beats is an upper bound is confirmed and quantified.
Per-cell
| cell | J observed | stratified null median | null p95 | excess | perm p |
|---|---|---|---|---|---|
| U_pair, paper rule | 0.328 | 0.247 | 0.254 | 1.33x | 0.0005 |
| U_27, paper rule | 0.397 | 0.272 | 0.284 | 1.46x | 0.0005 |
| U_full, paper rule | 0.283 | 0.217 | 0.223 | 1.31x | 0.0005 |
| U_pair, quantile rule | 0.196 | 0.053 | 0.066 | 3.67x | 0.0005 |
| U_27, quantile rule | 0.212 | 0.047 | 0.064 | 4.46x | 0.0005 |
| U_full, quantile rule | 0.164 | 0.061 | 0.071 | 2.69x | 0.0005 |
p = 0.0005 is (0+1)/(2000+1): no permutation out of 2,000 reached the observation in any cell. The excess is not in doubt. Only its size, and its meaning, were.
And the excess is not specific to the same patient
The same test run on 27 unrelated KMT2A::AFF1 sample pairs gives excesses of 1.19-1.62x under the paper rule and 2.1-4.5x under the quantile rule, with the same permutation p. Every KMT2A::AFF1 pair in this deposit overlaps more than a structure-respecting null, by about the same amount as one patient's own two lineages do. That is the finding: the excess is a property of the fusion, not of the patient.
Two corrections that run in opposite directions
The structure-respecting null makes every excess figure here SMALLER than published. But the same-epitope antibody negative control - two antibodies against the same end of KMT2A, which must call nothing - calls 2,711 promoters in p279 against 4,347 real N-over-C calls (62%) and 2,173 against 3,539 in p318 (61%). That reagent contribution is shared by both members of every pair and so cannot manufacture a same-patient effect, but it dilutes every call set toward noise and so makes every excess figure a LOWER bound on the biological excess. The true value is bracketed by a correction that shrinks it and a contamination that inflates the denominator, and this beat cannot separate them. This corpus already recorded the negative control independently; this is a replication of it at a different depth-matching, not a new find.
| script | analyses/q146-one-universe-jaccard/run146.py |
|---|---|
| status | run |
| result summary | Excess over the activity-stratified null survives in all six cells at perm p=0.0005, at 1.31-4.46x. q144's published 5.8x shrinks to 1.46x, q142's implied 6.44x to 2.69x. Unrelated KMT2A::AFF1 pairs show the same excess, so it is a fusion-level and not a patient-level property. |
| perm p all cells | 0.0005 |
| negctl p279 worst | 2711 |
| negctl p279 real | 4347 |
| negctl p318 worst | 2173 |
| negctl p318 real | 3539 |
inputs
[ "analyses/q146-one-universe-jaccard/results.json", "analyses/q146-one-universe-jaccard/checks.json" ]
excess range
[ 1.305, 4.461 ]
- A shared target set across patients is the premise of any fusion-directed rather than patient-directed therapy. A 2.4-4x overstatement of how shared it is would inflate confidence in a one-size-fits-all target list.
- The excess is present in every KMT2A::AFF1 pair. Is it present in cross-fusion pairs at the same size? If yes, the shared component is not even fusion-partner-specific and is a generic active-promoter signal.
- Can the reagent contribution be subtracted rather than merely bounded - for instance by requiring a call in both N antibodies independently and in neither C-versus-C contrast?
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
The same-epitope antibody pairs (KMT2A N1 Millipore 05-764 / N2 CST D2M7U; C1 Santa Cruz H10 / C2 Millipore 05-765) are two reagents against the same end of the same protein, so any site the N/C criterion calls between them is a false positive of the criterion.
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
Computed on raw un-thinned libraries in the pair universe, whereas the real N-over-C contrast was depth-matched, so the 62% is an approximate ratio between differently-normalised quantities.
Evidence that argues against this (1)
- The two N antibodies could in principle have genuinely different epitope accessibility across chromatin states, in which case some of those 2,711 sites are real biology about epitope masking rather than reagent noise - which would be a finding, not a control failure. Janssens DH et al. 2021
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The oncoprotein N/C enrichment is genuinely variable between samples and dynamically regulated, which is the paper's central claim and is consistent with a real but modest shared component across patients.
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 paper's variability claim is about the magnitude of enrichment, not about set overlap; reading it as support for a modest shared target set is an inference.
Evidence that argues against this (1)
- If the oncoprotein's target set were strongly fusion-partner-determined, unrelated KMT2A::AFF1 pairs should overlap much more than 1.2-1.6x above a matched null, and they do not - which is a mild argument against the partner-determines-targets reading of the same paper. Janssens DH et al. 2024
the null whose survival is being tested
the survival result behind the headline's perm p
replicated at a different depth matching: 62% and 61% of the real call count
second independent demonstration that the criterion fires without an oncoprotein
every KMT2A::AFF1 pair shows the same excess