Oncofusion

start / graph / ana-q141-the-authors-signal-floor-removes-ninety-two-percent-of-non-occupied-genes-from-the-menin-sheet

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

The authors' signal floor removes 92% of non-occupied genes from the menin table, and all of the occupied ones survive

The published table of menin measurements does not contain every gene - the authors kept only genes with enough signal. All the fusion's target genes make the cut; nine out of ten of the others do not. So 'menin did not change elsewhere' means 'at the other places where menin was detectable', not 'everywhere in the genome'.

The same thing, for a scientist

A filter must report what it removed. The deposited channel sheets are floored on the authors' stated criteria - *'2 fold enrichment over input'* and *'reads > 200 in either DMSO or VHL'* - written in the `Menin TSS 3h` sheet itself. Against a 19,168-gene HA universe: MENIN keeps 6,036 genes, DOT1L 4,561, CDK9 4,756, AFF4 3,138, H3K79me2 5,795, Pol II TSS 13,118, H3K4me3 13,810, MLL1 C-terminal 1,249. **Stratum survival is strongly asymmetric.** For MENIN, **100%** of the authors' 200 target genes survive but only **8.3%** (717 of 8,672) of the bottom half of the HA ranking does. For MLL1 C-terminal it is 34.5% and 4.6%; for DOT1L 89.0% and 11.0%; for AFF4 75.0% and 6.0%. **Consequence, stated plainly:** the 'no change at non-occupied genes' figure of 0.974 is computed on menin sites that happen to lack a fusion, not on the rest of the genome. That is the right comparator - menin loss cannot be measured where there is no menin - but it is not the comparator the phrase suggests.

Go deeper
Does the filter bias the result?

The filter is an OR across arms (>200 reads in DMSO or dTAG-VHL), so it retains genes that gained signal as well as genes that lost it, which is the non-biasing form. The residual concern is that it is applied on raw counts and the arms differ in depth. The direction of any residual bias is toward retaining genes with high signal in at least one arm, which inflates the apparent stability of the surviving low-HA genes - i.e. it runs against the finding of a stratum gap, not in favour of it.

The channel where this nearly disqualifies the comparison

MLL1 C-terminal keeps only 69 of 200 target genes. Its negative stratum gap is therefore an estimate from 69 occupied and 401 non-occupied genes, and it is used in this beat only as a qualitative negative control, never as a magnitude.

Facts and figures
scriptanalyses/menin-retention-by-ha-occupancy/analyze_tss.py
statusrun
result summaryPer-channel sheet sizes against a 19,168-gene universe, and per-stratum survival percentages, are tabulated in results.json under primary_result_3h_HK_normalised_TSS_ratio.channels[*].strata[*].pct_of_stratum_surviving_sheet_filter. MENIN: 100% of occupied, 8.3% of non-occupied.
peak width ratio noteSignal instrument; no peak-width ratio applies.
inputs
[
  "GSE173599_TSS_gene_body_signal_Figure3.xlsx",
  "GSE173599_ChIPseq_TSS_Figure3_revisions.xlsx",
  "GSE173599_TSS_signal_HA_ChIPs.xlsx"
]
How this could bite us back
  • A pharmacodynamic assay that only reports genes above a signal floor will report a drug as having no off-target chromatin effect simply because the off-target sites were filtered out before the comparison.
Still open
  1. Would recomputing from the raw reads without the authors' floor change the non-occupied MENIN figure? The BAMs are deposited (23 GB) and were not downloaded.
Every source behind this entry (3)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    The deposited MENIN sheet states its own filter - 2-fold enrichment over input and reads greater than 200 in either arm - and retains 6,036 of a 19,168-gene universe.

    What this evidence cannot show

    The filter text is written in the sheet by the authors; the exact implementation (which input library, whether per-arm or pooled) is not documented.

    Evidence that argues against this (1)
    • A 2-fold-over-input and 200-read floor is a conventional and defensible choice for ChIP-seq at this depth; describing what it removed is not a criticism of the authors' analysis, and an unfiltered table would carry more noise than signal at the removed genes. Olsen SN et al. 2022
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    100% of the authors' 200 target genes survive into the MENIN sheet but only 8.3% of the bottom half of the HA-occupancy ranking does.

    What this evidence cannot show

    The asymmetry is expected - fusion target genes are high-signal - but it changes what the non-occupied comparator is, and that must be stated wherever the 0.974 figure is quoted.

    Evidence that argues against this (1)
    • If menin were lost at low-signal non-occupied genes and those genes were removed by the floor, the analysis would miss it; the floor is an OR across arms, which limits but does not eliminate that possibility. Olsen SN et al. 2022
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    The MLL1 C-terminal sheet retains only 69 of the 200 target genes, so that negative control rests on a small occupied arm.

    What this evidence cannot show

    1,249 genes total; sequenced in its own batch (210928).

    Evidence that argues against this (1)
    • The MLL1 C-terminal libraries were sequenced last, in 2021, and are shallower than the rest; the low survival may reflect depth rather than a biological absence of wild-type MLL1 at those genes. Olsen SN et al. 2022
Connected to