q142's exact comparison - p279 against its own switched AML versus p279 against an unrelated infant - replicates in q144's own universe with q144's own rule, and q144 never made it
The earlier analysis that said a patient's two diseases share fewer target genes than two different children was right, and it stays right even when you use the other analysis's methods. The other analysis simply never ran that particular comparison.
The same thing, for a scientist
q142 reported J(p279,p318) = 0.117 against J(p279,p454) = 0.149 - the same patient's two lineages sharing less than p279 shares with an unrelated KMT2A::AFF1 infant at the same institution. That ordering holds in **all six cells** of the {3 universe} x {2 rule} factorial, on the raw Jaccard AND on the excess over the activity-stratified null. It holds in q144's OWN universe with q144's OWN calling rule: 0.397 for the same patient against 0.434 for p279-versus-p454. q144 did not report this because it compared the same-patient pair to a 351-pair cohort median (0.132) rather than to the best-matched unrelated pair. **The two beats never disagreed about the data; they disagreed about what to compare it to.**
The trio, in all six cells
| cell | J(p279,p318) same patient | J(p279,p454) unrelated | J(p318,p454) unrelated |
|---|---|---|---|
| U_pair, paper rule | 0.3281 | 0.3731 | 0.3294 |
| U_27, paper rule (q144's universe and rule) | 0.3969 | 0.4335 | 0.3904 |
| U_full, paper rule | 0.2829 | 0.3481 | 0.2909 |
| U_pair, quantile rule | 0.1960 | 0.2546 | 0.1786 |
| U_27, quantile rule | 0.2115 | 0.3210 | 0.1893 |
| U_full, quantile rule (q142's universe and rule) | 0.1642 | 0.2236 | 0.1706 |
On excess over the matched null the same holds: X(p279,p318) is below X(p279,p454) in 6 of 6 cells (1.329 vs 1.397; 1.459 vs 1.551; 1.305 vs 1.396; 3.668 vs 4.175; 4.461 vs 5.650; 2.689 vs 3.386).
The honest limit: the p318-anchored version is 3 of 6
The unanimity is on the p279-anchored comparison, which is the one q142 made. Anchoring on p318 instead - J(p279,p318) against J(p318,p454) - the same-patient pair is lower in only 3 of 6 cells. So the robust statement is narrow and must be stated narrowly: p279's own switched AML is a worse match for p279 than an unrelated infant is, under every universe and rule tested. The symmetric claim that the pair is the least similar of the three is true in three cells and false in three, and the institution-matched anchored analysis is the stronger form of that argument.
This is also why a single comparator, as in q142, is not enough even when it is right: q142's n = 1 comparator gave the correct direction and could not have known it was correct.
Why the numbers differ from q142's own
q142 reported 0.117 and 0.149; the matching cell here reports 0.164 and 0.224. The difference is the signal type, which is a fourth divergence between the two beats that this beat holds fixed: q142 scored bigWig promoter signal shares, this beat scores fragment counts from the deposited BEDs. The ordering is identical; the absolute values move by ~40%. That is worth recording because it means neither beat's absolute Jaccard is a portable number - only the ordering, against a stated comparator, travels.
| script | analyses/q146-one-universe-jaccard/run146.py |
|---|---|
| status | run |
| result summary | J(p279,p318) < J(p279,p454) in 6 of 6 factorial cells, on raw Jaccard and on excess over the matched null, including in q144's own universe with q144's own rule (0.397 vs 0.434). The p318-anchored version of the same comparison is 3 of 6. q144 compared to a cohort median instead and never ran this. |
| cells p279 anchored same patient lower | 6 |
| cells p318 anchored same patient lower | 3 |
| note on signal type | q142 used bigWig signal shares; this beat uses fragment counts. Ordering identical, absolute values differ ~40%. |
inputs
[ "analyses/q146-one-universe-jaccard/results.json" ]
trio
{
"U_pair|paper": [
0.3281,
0.3731,
0.3294
],
"U_27|paper": [
0.3969,
0.4335,
0.3904
],
"U_full|paper": [
0.2829,
0.3481,
0.2909
],
"U_pair|quantile": [
0.196,
0.2546,
0.1786
],
"U_27|quantile": [
0.2115,
0.321,
0.1893
],
"U_full|quantile": [
0.1642,
0.2236,
0.1706
]
}q142 published values
[ 0.117, 0.149 ]
- Only the ordering travels, not the number. Any downstream use of a stored Jaccard from this corpus as an absolute measure of target-set conservation is unsupported.
- Why is the p318-anchored comparison only 3 of 6? p318 has the fewest measurable promoters of the three (11,729 versus 12,897 and 12,884), so its pair universes are smaller; whether that alone explains the asymmetry has not been tested.
- The absolute Jaccard moves 40% between bigWig signal and fragment counts on identical samples and identical rules. Which of the two is closer to the paper's own pipeline, and does any conclusion in this corpus depend on the absolute value rather than the ordering?
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
p454 is an unrelated KMT2A::AFF1 infant B-ALL from the same institution as p279 and p318, which is what makes it the correct matched comparator rather than an arbitrary one.
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
Same fusion, same institution, same lineage at sampling, comparable depth (19.2M/19.5M per arm versus p279's 21.3M/16.7M) - but a different child of unknown age, and one comparator is one comparator.
Evidence that argues against this (1)
- p454's libraries are slightly more balanced between the N and C arms than p279's, and the calling rule's binomial null share is TN/(TN+TC), so a more balanced pair of libraries could call sites at marginally different rates for purely technical reasons. Janssens DH et al. 2024
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The publication frames p279 and p318 as one patient's B-ALL and the AML it switched into, and identifies target sites with the most dramatic differences between them - i.e. it treats the two as differing, not as conserved.
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 does not report a quantitative overlap statistic between the two, nor a comparison of that overlap to an unrelated pair, so it neither supports nor contradicts the ordering directly.
Evidence that argues against this (1)
- The paper's emphasis on B-ALL-specific and GMP-like target sets could be read as implying a large conserved core with two differing tails, which a low Jaccard against an unrelated comparator does not obviously support. Janssens DH et al. 2024
q142's p279-anchored ordering replicates in 6 of 6 cells including q144's own universe and rule
q144's own universe and rule give 0.397 same-patient against 0.434 for the matched unrelated pair
the like-for-like comparison behind the headline
the same six cells scored on q142's own comparator