Oncofusion

start / graph / ana-q141-the-whole-deposit-is-hg38-even-where-the-filename-does-not-say-so

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q141-menin-at-fusion-targets

The whole deposit is hg38 even where the filename does not say so, and the guard that checked it was worth running

Only some of the files in this dataset are labelled with the version of the human genome they were mapped to. Mixing versions would have silently produced a wrong answer, so the version was checked directly from the coordinates - and all of them turn out to be the same, newer version.

The same thing, for a scientist

A pre-registered hazard that did **not** materialise, recorded because a clean guard is a result. Of the deposited MACS peak files, only some carry `hg38` in the filename (CDK9 `200211`, H3K4me3 `200721`, H3K27ac `201007`, HA `200908`, MLL1c `210928`); the batches `190814`, `1909x2`, `1912x2` and `191121` carry no build tag. Intersecting an hg19 HA peak set with an hg38 target channel would have produced a spurious non-overlap and therefore a spurious 'menin is not at fusion sites'. **Test, from coordinates rather than from filenames:** the maximum peak end per chromosome against hg19 and hg38 chromosome lengths, with chr17 decisive (hg19 81,195,210 vs hg38 83,257,441). **Every control-arm file tested exceeds the hg19 chr17 length** - HA 1909x2 83,081,307; HA 190814 82,520,736; HA 1912x2 82,520,599; MEN1 1912x2 82,459,220; DOT1L 1909x2 83,081,524; H3K79me2 190814 83,088,758; and the filename-tagged hg38 HA 200908 83,081,062. The corroborating check is peak overlap: HA 1909x2 shares 58-80% of its peaks with the filename-tagged hg38 HA file, which is impossible across builds at 1.4 kb resolution.

Go deeper
Why this is recorded rather than dropped

A guard that fires is a finding and a guard that does not fire is also a finding, because the next agent to touch these files needs to know the check was run and what it returned. Without this entry, the next beat re-derives the hazard from the filenames and either wastes a day or, worse, splits the analysis by filename and loses power for no reason.

The one case the test cannot call

The two HA dVHL files return 'undetermined by maximum', because after degradation they retain so few peaks that none reaches chr17q (chr17 max 26,951,666 in batch 1909x2). They are from the same sequencing batches as their controls, so they inherit the call, but the direct test does not cover them.

Facts and figures
scriptanalyses/menin-retention-by-ha-occupancy/analyze_peaks3.py
statusrun
result summaryAll control-arm peak files are hg38, including the untagged batches 190814, 1909x2, 1912x2 and 191121. The declared genome-build hazard did not materialise. The two degraded HA arms are undetermined by the coordinate test and inherit their batch's call.
peak width ratio noteNot applicable - this is a coordinate-range test, not a retention measure.
inputs
[
  "GSE173599 MACS peak tables"
]
How this could bite us back
  • A build mismatch in a pharmacodynamic pipeline would report zero target engagement for a working drug, and nothing in the output would say so; this is the same class of silent failure as an insensitive endpoint.
Still open
  1. Does GEO's sample-level metadata state the build for the untagged batches? The series record does not, and the peak-file headers do not.
Every source behind this entry (3)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    Every deposited control-arm MACS peak file tested has a peak ending beyond the hg19 chr17 length of 81,195,210, including files whose names carry no build tag.

    What this evidence cannot show

    A coordinate-range test proves hg38 only in one direction: exceeding the hg19 length rules out hg19, while failing to exceed it is uninformative, which is why the degraded HA arms cannot be called this way.

    Evidence that argues against this (1)
    • A file could in principle be mapped to a third assembly (T2T-CHM13) whose chr17 is longer still; the peak-overlap corroboration with a filename-tagged hg38 file rules that out here. Olsen SN et al. 2022
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    HA 1909x2 shares 58-80% of its peaks with the filename-tagged hg38 HA 200908 file, which cannot happen across genome builds at these peak widths.

    What this evidence cannot show

    The asymmetry (58% one way, 80% the other) reflects the different peak counts (12,017 vs 11,044), not a build difference.

    Evidence that argues against this (1)
    • High peak overlap between two libraries of the same antibody in the same cell line is expected on any shared assembly, so this check confirms a shared build rather than identifying which build it is. Olsen SN et al. 2022
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    The MACS command lines deposited in the peak-file headers name the BAM directories, which distinguish the sequencing batches but not the genome build.

    What this evidence cannot show

    The headers are the authors' own command lines; the effective genome size (2.70e9) is identical for hg19 and hg38 in MACS and carries no information.

    Evidence that argues against this (1)
    • The BAM directory names distinguish only the sequencing run, and the 2021 re-processing that added hg38 to some filenames may have re-processed earlier runs without renaming them, which would explain the inconsistent tags without any build mixture. Olsen SN et al. 2022
Connected to