Oncofusion

start / graph / absence-q141-no-ha-library-exists-in-the-aff4-batch-so-aff4-cannot-be-partitioned-on-peaks

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q141-menin-at-fusion-targets

No HA library exists in the AFF4 sequencing batch, so AFF4 is excluded from the peak partition by name

One protein, AFF4, could not be checked by the peak-counting method, because the experiment that would tell us where the fusion was sitting was never run in the same sequencing batch. This is written down rather than quietly left out.

The same thing, for a scientist

A named exclusion, recorded because a dropped case that is not named is indistinguishable from a case that did not exist. AFF4 was sequenced in batch **191121** (GSM5271643 DMSO, GSM5271644 dTAG-VHL). **No HA library was sequenced in that batch**, so no batch-matched fusion-occupancy peak set exists and the peak-based occupancy partition cannot be run for AFF4. AFF4's peak-level figures that do not need a partition are reported: 1,784 control peaks, 36.5% summit-containment retention, treated/control median peak width ratio **1.05** (below the 1.3 trigger). AFF4 is retained in full in the **signal** analysis, where matching is by gene symbol and no batch matching is needed: 0.821 at the authors' target genes vs 0.998 at non-occupied genes, raw stratum gap 17.7 points on these medians - 17.3 on the marginally different gene set the matching script uses, which requires membership in the HA ranking - rising to 30.4 under signal-decile matching (6 deciles). So AFF4 is not missing from the beat - only from one of its two instruments.

Go deeper
Why a cross-batch HA file was not substituted

HA peak files exist in batches 1909x2, 190814, 1912x2 and 200908, and any of them could have been intersected with the AFF4 peaks. That would have introduced a batch confound into the stratum definition - different fragment sizes, different depths, different peak-width distributions - while looking exactly like a clean result. The exclusion is the conservative choice and it is stated rather than performed silently.

What AFF4's signal figure says on its own

AFF4 barely moves at fusion-occupied promoters (0.821) while CDK9, its partner in the super elongation complex, moves to 0.687 and Pol II pS2 to 0.775. Either the SEC partially disassembles with AFF4 left behind, or one of the two antibodies is reporting poorly. The deposit cannot separate these.

Facts and figures
scriptanalyses/menin-retention-by-ha-occupancy/analyze_peaks.py
statusrun
result summaryAFF4 excluded from the peak-based occupancy partition: no HA library exists in batch 191121. Retained in full in the signal analysis (0.821 occupied vs 0.998 non-occupied). Peak figures without a partition: 36.5% retention, width ratio 1.05.
peak width ratio noteAFF4 1.05, below the 1.3 pre-registered trigger.
absence classnever-measured
absence class basisNo HA library was sequenced in batch 191121, so no batch-matched fusion-occupancy peak set was ever generated.
inputs
[
  "GSE173599 MACS peak tables",
  "GSE173599_TSS_gene_body_signal_Figure3.xlsx"
]
How this could bite us back
  • If AFF4 remains at fusion target promoters after the fusion is gone, the scaffold for the elongation complex is still in place and the programme can be re-established by anything that re-supplies a recruiting factor - which is a compensation route no menin-directed therapy addresses.
Still open
  1. Does AFF4 leave fusion-occupied sites at the peak level when a cross-batch HA definition is used, and does the answer differ from the signal analysis? Running it as a labelled sensitivity analysis would cost nothing and would bound the batch confound.
  2. Why does AFF4 move so much less than CDK9 when both are super elongation complex subunits?
Every source behind this entry (3)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    AFF4 ChIP libraries in GSE173599 are GSM5271643 and GSM5271644, both from sequencing batch 191121, and no HA library appears in that batch.

    What this evidence cannot show

    Established by reading the full sample list; an HA library sequenced in that batch but not deposited would not be visible.

    Evidence that argues against this (1)
    • An HA peak set from a different batch could be substituted and would very likely give a usable partition, since all files are hg38 and HA occupancy is a stable property of the line; refusing the substitution is a choice, and a reader may reasonably judge the batch confound smaller than the lost information. Olsen SN et al. 2022
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    AFF4's peak-level figures without a partition: 1,784 control peaks, 36.5% summit retention, treated/control median peak width ratio 1.05.

    What this evidence cannot show

    n=1 per arm; the AFF4 libraries are among the shallower in the deposit (1,784 and 2,376 peaks).

    Evidence that argues against this (1)
    • The authors describe AFF4 as specifically lost at promoters of MLL::AF9 target genes, which a 36.5% unpartitioned retention figure neither supports nor contradicts - so the peak figure reported here should not be read as evidence against their claim. Olsen SN et al. 2022
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    AFF4 in the signal analysis gives 0.821 at target genes and 0.998 at non-occupied genes, a 17.7-point raw gap that rises to 30.4 under signal-decile matching.

    What this evidence cannot show

    3,138 genes in the sheet; 75.0% of the occupied stratum and 6.0% of the non-occupied stratum survive the authors' floor. Housekeeping-normalised, n=1.

    Evidence that argues against this (1)
    • The authors report AFF4 as specifically lost at promoters of MLL::AF9 target genes, a stronger statement than a 0.821 median supports; their figure is a boxplot and the tail may carry the effect they describe. Olsen SN et al. 2022
Connected to