Depth matching changes this pair's call counts by under one percent, so library depth is not what separated the two beats - a candidate explanation cleared
We suspected that differences in how deeply each sample was sequenced might explain the disagreement. It does not: equalising them changes the numbers by less than one percent for this pair.
The same thing, for a scientist
An absolute calling rule with a binomial p-value threshold calls more sites in a deeper library, so depth was a live candidate for the divergence between q142 and q144 - and q144's own entry records that its amplitude statistic passed its pre-registered depth check only just. Tested directly: binomially thinning both p279 and p318 to the pair's minimum per-arm fragment total moves p279's call count from 2,045 to 2,045 and p318's from 1,894 to 1,883 in q144's universe - a 0.6% change - because the two libraries are already closely matched (p279 21.3M/16.7M N/C fragments, p318 22.9M/17.1M). Depth matching is retained in every cell of this beat because it is cheap and the rule demands it in general, but **for this pair it is not the explanation**, and a beat looking for the cause of the divergence should stop looking here.
The measurement
| p279 | p318 | p454 | |
|---|---|---|---|
| N-arm fragments (N1+N2) | 21,336,275 | 22,878,088 | 19,188,453 |
| C-arm fragments (C1+C2) | 16,729,371 | 17,053,864 | 19,453,899 |
| measurable promoters (>=20 fragments) | 12,897 | 11,729 | 12,884 |
| calls in U_27, unmatched (q144) | 2,045 | 1,894 | 2,000 |
| calls in U_27, depth-matched (this beat) | 2,045 | 1,883 | 1,987 |
Depth matching is pair-specific, so a sample's matched count depends on which pair it is in: p279 calls 2,045 when matched against p318 and 2,041 when matched against p454, and the 1,987 above is p454's count inside the p279/p454 pair. That 4-call spread is the size of the whole effect.
These three are among the deepest libraries in the deposit and are within 20% of each other on both arms. Depth matching bites hard elsewhere - it is why 30 pairs are withheld and why the St Jude samples behave differently - but not here.
What this clears and what it leaves
Cleared: depth, for this pair. Still standing as explanations of the divergence: the calling rule (measured, and it is the one that flips the direction), the comparator (measured, and it is confounded with institution), the universe (measured, and it moves the raw Jaccard without moving the direction), and the signal type - bigWig shares versus fragment counts - which moves the absolute Jaccard by ~40% and the ordering not at all.
Recording a cleared candidate matters because the alternative is that a future beat re-runs the same check. It also matters because depth was the most plausible-sounding explanation and it is wrong, which is a reason to distrust plausibility here.
| script | analyses/q146-one-universe-jaccard/run146.py |
|---|---|
| status | run |
| result summary | Within-pair depth matching moves p279's call count by 0 and p318's by 11 of 1,894 (0.6%) in q144's universe. Depth is cleared as an explanation of the q142/q144 divergence for this pair; the rule, the comparator, the universe and the signal type all remain. |
| PCT change p318 | 0.58 |
inputs
[ "analyses/q146-one-universe-jaccard/results.json" ]
depth p279
[ 21336275, 16729371 ]
depth p318
[ 22878088, 17053864 ]
depth p454
[ 19188453, 19453899 ]
calls unmatched
{
"p279": 2045,
"p318": 1894
}calls depth matched — the numbers behind the chart
{
"p279_vs_p318": 2045,
"p318_vs_p279": 1883,
"p279_vs_p454": 2041,
"p454_vs_p279": 1987
}- A cleared confound is only cleared for the samples it was tested on. Applying this negative result to the shallow libraries in the same deposit would be exactly the kind of transfer this corpus has been wrong about before.
- Depth matching is cleared for the deep Texas samples. Is it the whole story for the St Jude samples, whose lower excess over the matched null could be depth rather than institution? Thinning the Texas samples to St Jude depths would separate the two.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
Per-arm fragment totals and per-sample measurable-promoter counts for every sample in the KMT2A N/C arm are computable directly from the deposited hg38 fragment BEDs, and p279, p318 and p454 are within 20% of each other on both arms.
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
Fragment totals are genome-wide counts from the deposited BEDs, not the sequencer's raw read counts, so they already reflect whatever filtering the depositors applied.
Evidence that argues against this (1)
- A library can be deep and still uninformative if its fragments are non-specific; the same-epitope antibody control calls ~60% as many promoters as the real contrast in these very samples, which is a quality problem depth cannot fix. Janssens DH et al. 2024
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The oncoprotein score's amplitude in this deposit does depend on sequencing depth in general, which is why the check was worth running even though it came back negative for this pair.
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 depth dependence was measured by an earlier beat of this corpus on this deposit rather than reported in the publication; the publication's own normalisation is a within-sample score.
Evidence that argues against this (1)
- If the depth dependence is concentrated in the shallow libraries, a check on three of the deepest samples is the least likely place to find it, so this negative result is weak evidence about the general phenomenon. Janssens DH et al. 2024
a candidate explanation of the divergence, cleared
depth dependence exists in the deposit but is negligible for this pair
the depth matching applied in every cell of the headline