The only second mark with a drug arm in SEM is H3K27ac, and its replicates agree at r = 0.21-0.29 against H3K4me3's 0.78-0.83 - so GSE159608's SEM EPZ-5676 experiment has one usable readout, not two
There is a second chemical mark measured under one of the two drugs, which would have been a useful cross-check. Its repeat measurements barely agree with each other, so it cannot check anything. Recording this matters because the dataset looks from its labels as though it has two readouts.
The same thing, for a scientist
Genome-wide 5 kb Pearson between the SEM H3K27ac libraries of GSE159608: DMSO Rep1 vs Rep2 **r=0.206**, EPZ-5676 Rep1 vs Rep2 **r=0.293**, DMSO Rep1 vs EPZ Rep1 **r=0.223**. For comparison, the H3K4me3 libraries in the same experiment, same submitter, same 72-hour exposure, agree at **r=0.789 (DMSO), 0.827 (EPZ)** and the untreated cross-protocol H3K4me3 pair at **0.901**. The H3K27ac libraries still nearest-neighbour correctly - every one's closest partner is another H3K27ac library - so they are correctly labelled; they are simply not reproducible enough to support a drug comparison.
Correct label, unusable data - these are different failures
This deposit's labels are right (see the antibody self-sorting entry). That says nothing about whether a library is good. The H3K27ac libraries pass the identity check and fail the reproducibility check, and a beat that ran only the identity check would have concluded the arm was fine. Any inventory of 'what readouts exist under a menin or DOT1L inhibitor' should carry a reproducibility column, not just an antibody column. 1
What it costs
The EPZ-5676 arm of this experiment therefore rests on H3K4me3 alone, and the VTP-50469 arm on H3K4me3 alone with no second mark at all (see the bounded-absence entry). The whole menin-inhibitor chromatin result in the infant-derived index-fusion line reduces to two H3K4me3 libraries, both of which have a 2.5-fold lower fraction of reads in peaks than their controls. That is the real state of the evidence and it is much thinner than the deposit's sample count suggests.
| script | analyses/sem-arm-identity-from-signal/run.py |
|---|---|
| status | run |
| verdict | correctly labelled, not reproducible enough to carry a drug comparison |
h3k27ac
{
"DMSO_rep1_vs_rep2": 0.206,
"EPZ_rep1_vs_rep2": 0.293,
"DMSO1_vs_EPZ1": 0.223,
"libraries": [
"GSM5139705",
"GSM5139706",
"GSM5139707",
"GSM5139708"
]
}h3k4me3 for comparison — the numbers behind the chart
{
"DMSO_rep1_vs_rep2": 0.7892,
"EPZ_rep1_vs_rep2": 0.8272,
"untreated_cross_protocol": 0.901
}- An acetylation readout is the natural pharmacodynamic marker for a chromatin drug, and this entry says the one that exists here cannot carry that role. Anyone building an H3K27ac biomarker from this deposit should check replicate concordance first.
- Is the low H3K27ac reproducibility specific to SEM in this series, or does it hold across the 419 libraries? If it is general, several published panels rest on it.
- At 500 bp bins restricted to called enhancers, do these libraries recover? That would separate 'coarse binning penalises a narrow mark' from 'these libraries are weak'.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
SEM H3K27ac replicate pairs in GSE159608 correlate at r=0.206 (DMSO) and r=0.293 (EPZ-5676) over 5 kb bins, against 0.789 and 0.827 for the H3K4me3 pairs in the same experiment.
sem-arm-identity-from-signal: run.py, results.json nearest_neighbour_pearson (this repository) · Q140-sem-menin-arm · Oncofusion knowledge graph 2026What this evidence cannot show
5 kb bins are coarse for H3K27ac, which marks narrow enhancers; a finer binning would raise the numbers somewhat for all libraries, but the comparison with H3K4me3 in the same matrix at the same bin size is internally fair. Depths are comparable (1.95M-2.66M fragments).
Evidence that argues against this (1)
- H3K27ac genuinely has a broader, lower-amplitude distribution than H3K4me3, so some of this gap is expected biology rather than library quality. The gap is large enough (0.21 vs 0.79) that it is unlikely to be entirely that, but this entry cannot apportion it. Janssens DH et al. 2021
Correct labels and poor reproducibility are separate properties; H3K27ac passes the first and fails the second
The SEM H3K27ac replicate pairs agree at r=0.21-0.29 against H3K4me3's 0.79-0.83
The only second mark under a drug cannot carry a conclusion, which narrows the deposit further