Pooling the two AF4 antibodies, as the published analysis does, drops AFF1's fusion-site enrichment from 1.8-4.6x the null to 0.8-2.1x - below the H3K4me3 negative control in all three samples
The published study combined two antibodies for the same protein into one signal. Doing that makes the protein look even less like a marker of the fusion than the better of the two antibodies does on its own.
The same thing, for a scientist
**The decision.** The paper's methods state that reads from the two AF4 antibodies were pooled for downstream analysis [E2]. This beat carried them separately, because they are different reagents in different species against different epitopes, and pooling is only safe once concordance is shown. **What pooling costs, measured** [E1]. Each track's own top 971 promoters, counted into the N/C-defined fusion set, against a random null of 48.6: | | p279 | p318 | p454 | |---|---:|---:|---:| | AFF1-1 alone (mouse mAb) | 112 (**2.3x** null) | 89 (**1.8x**) | 223 (**4.6x**) | | AFF1-2 alone (rabbit pAb) | 19 (0.4x) | 21 (0.4x) | 46 (0.9x) | | **AFF1 pooled** | 50 (**1.0x**) | 41 (**0.8x**) | 100 (**2.1x**) | | H3K4me3, the negative control | 101 (2.1x) | 122 (2.5x) | 129 (2.6x) | and as a share ratio over the wild-type set: AFF1-1 1.08 / 1.03 / 1.45, AFF1-2 0.67 / 0.56 / 0.76, **pooled 0.85 / 0.77 / 1.06**. **Pooled AFF1 is below the H3K4me3 negative control in all three samples. The better single antibody is at or above it in two.** **And the published conclusion survives anyway, which is the part that matters.** The paper's sentence is that AF4 cannot distinguish oncoprotein-target from wild-type sites in patient samples [E2]. Even with the pooling removed and the stronger reagent used alone, AFF1-1 reaches 1.03-1.45 against ENL's 1.90-2.78 and clears the negative control in one state of three. **So the conclusion is not an artefact of pooling** - it is if anything made to look stronger than the reagent-resolved data support. This entry corrects the method, not the finding.
The general form
Averaging two measurements of the same quantity reduces variance. Averaging two measurements of DIFFERENT quantities produces a third quantity that is neither, and its enrichment can be lower than either input's when the two are enriched in disjoint places - which is what happens here, because AFF1-2's top promoters fall in the fusion set at 0.4-0.9x the null, i.e. actively away from it. The tell that pooling is unsafe is available before pooling and costs one correlation: rho 0.712-0.784 between the two reagents in patient samples.
| script | analyses/q142-aff1-in-primary-infants/extra.py |
|---|---|
| status | run |
| result summary | Pooled AFF1 0.8-2.1x null vs AFF1-1 alone 1.8-4.6x and H3K4me3 2.1-2.6x; the published conclusion survives reagent resolution. |
- How many chromatin results in this corpus rest on a track that pooled two antibodies without a reported concordance statistic?
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
In GSE252378 primary samples, pooling the two AFF1 antibodies gives a top-promoter overlap with the N/C-defined fusion set of 0.8-2.1x the random null, below the 2.1-2.6x of the H3K4me3 negative control and below the 1.8-4.6x of the mouse monoclonal alone.
GSE252378 - AutoCUT&RUN and AutoCUT&Tag profiles of 36 MLL-rearranged leukemia samples (171 GSMs, hg38) · Janssens DH, Henikoff S (Fred Hutchinson Cancer Center / Van Andel Institute) · NCBI Gene Expression Omnibus 2024What this evidence cannot show
Pooling here is an equal-weight average of the two within-sample promoter-share vectors, which is not identical to pooling raw reads as the authors did - read pooling weights by library depth. The direction of the effect is robust to that difference but the exact values would differ.
Evidence that argues against this (1)
- Read-level pooling would let the deeper library dominate, so if AFF1-1 is much deeper the authors' pooled track may behave more like AFF1-1 than like the equal-weight average computed here. Library depths were not extracted, so this entry cannot exclude that its pooled numbers are pessimistic relative to the published ones. Janssens DH et al. 2024
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The published methods pooled reads from the two AF4 antibodies for downstream analysis, and the paper concludes AF4 co-occupancy cannot validate oncoprotein-target calls.
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
Read from the methods and results of PMC11522475.
Evidence that argues against this (1)
- Pooling two antibodies is a defensible way to raise coverage in precious primary material, and the paper's conclusion is the conservative one either way; criticising the method here does not identify any error in the paper's stated result. Janssens DH et al. 2024
the concordance statistic that makes pooling unsafe
the published conclusion survives reagent resolution
pooled AFF1 track behaviour