The specificity control that should have clinched this beat inverts instead: the C-terminal MLL1 epitope the fusion does NOT contain loses MORE than the N-terminal epitope it does contain (68.8% vs 78.4% retained), and because the deposit records different cell-line names for these four samples than for every other one, the arm is reported as uninterpretable rather than interpreted
also called the MLL1 epitope control inverted · MLL1c falls further than MLL1 · the arm I could not use · 68.8 against 78.4
We had what looked like a perfect internal check - one antibody sees both the cancer fusion and the normal version of the protein, another sees only the normal version, and destroying the fusion should have moved the first while leaving the second alone. It did the opposite, and rather than invent a story for that we are recording that the check failed and why we cannot trust it.
The same thing, for a scientist
**The design, which is genuinely elegant.** MLL-AF9 keeps the MLL1 **N-terminus** and replaces the C-terminus with AF9. So: - **MLL1 N-terminal antibody** (Bethyl A300-086A) reads **fusion + wild-type MLL1**. The authors confirm this usage: *"To define MLL::AF9 targets in MOLM13 cells an N-terminal MLL1 antibody was used, which detects both wild-type MLL1 as well as MLL::AF9"* [E1]. - **MLL1 C-terminal antibody** ("MLL1c", Active Motif 61296) reads **wild-type MLL1 only** — the fusion does not contain that epitope. **Pre-registered expectation (falsifier 3): the N-terminal arm falls, MLL1c holds. If MLL1c falls as far as HA, the scoring approach is refuted.** **What happened, 3 h after degradation:** | arm | DMSO peaks | dVHL peaks | summit retention | bin retention | |---|---:|---:|---:|---:| | MLL1 **N-term** (fusion + WT) | 887 | 986 | **78.4%** | 98.4% | | MLL1c **C-term** (WT only) | 3,217 | 514 | **68.8%** | 62.6% | **Falsifier 3 does not fire** — MLL1c at 68.8% is nowhere near HA's 2.5-7.1%. **But the control inverts**: the arm that should have been unmoved moved more, under both measures. **Why I do not interpret it, and this is the point of the entry.** Three independent problems, any one of which is disqualifying: 1. **THE CELL LINE IS AMBIGUOUS IN THE DEPOSIT ITSELF.** The GEO sample-level `source_name` for these four samples reads **HCB2, HCB3, HCB4 and HCB5** — one per sample, in sequence — while the *same records'* `treatment_protocol` says the cells are **HCB1**, and every other sample in the series has a `source_name` that is simply a copy of its own title. See `concept-q139-a-deposit-can-contradict-itself-about-which-cells-an-experiment-used`. 2. **Different antibodies, different vendors, different batches.** Bethyl vs Active Motif, sequencing batches 190814 vs 210928. A cross-antibody retention comparison is the least defensible comparison in this beat. 3. **The MLL1c DMSO library is the shallowest in the deposit** at **3.77 M fragments**, against **7.81 M** for its own dVHL arm. A control arm at half the depth of its treated arm calls fewer, noisier peaks and inflates apparent loss — though note this runs **X-SAFE**: the treated arm being deeper means depth cannot manufacture the loss, so the 68.8% is not a depth artefact in the usual direction. **The N-term arm has its own problem**: only **887 peaks in DMSO**, the smallest reference set of any arm, with a median peak width of 2,852 bp. A top decile of 88 peaks is not a stable basis for anything. **So the honest output is: the control was well designed, it inverted, and the deposit does not support reading the inversion either way.**
What would make this arm usable
Three things, none of them expensive:
- Resolve the cell line. One email to the authors, or a sample-level metadata correction in GEO.
- Score the two epitopes against a shared reference. Both arms' retention is currently measured against its own DMSO peak set, which differs in size by 3.6-fold. Scoring both against the HA peak set from the same cells would ask the question the control was designed to ask — does the fusion-containing epitope leave the fusion's sites while the wild-type-only epitope stays?
- Read the wild-type MLL1 question somewhere else. The graph already carries
ana-q120-a-menin-inhibitor-evicts-wild-type-dominated-mll1-sites-harder-than-the-tagged-fusions-own-sites, which reaches a related conclusion under a drug rather than a degron. If wild-type-dominated MLL1 sites really are more labile than the fusion's own, this inversion may be a real and reproducible phenomenon rather than an artefact — which is an argument for resolving it, not for discarding it.
| script | analyses/what-leaves-when-the-fusion-is-degraded/run.py |
|---|---|
| status | run |
| result summary | Pre-registered falsifier 3 does not fire, but the specificity control inverts: MLL1c (wild-type only) 68.8% retained vs MLL1 N-term (fusion + wild type) 78.4%. Reported as uninterpretable on three grounds - unresolved cell line in the deposit, cross-antibody and cross-batch comparison, and an 887-peak reference set in the N-term arm. |
| preregistration | analyses/what-leaves-when-the-fusion-is-degraded/PREREGISTRATION.md, falsifier 3 |
inputs
[ "GSE173599 GSM5696327/GSM5696328 (MLL1c)", "GSE173599 GSM5696329/GSM5696330 (MLL1 N-term)" ]
- Reporting an arm as uninterpretable is the correct call here but it is also the easy one, and it should not become a habit for results that do not fit. The three disqualifying reasons are named specifically so that a later reader can check whether each actually holds rather than accepting the refusal.
- Scored against the HA peak set from the same cells rather than against each epitope's own DMSO, does the N-terminal arm lose the fusion's sites while MLL1c holds? That is the question the control was designed to answer and it was never asked in this form.
- Is wild-type-dominated MLL1 genuinely more labile than the fusion's own occupancy? Two entries in this graph now point that way from independent experiments, and neither was looking for it.
- What are HCB2 through HCB5? If they are independent cord-blood transformants rather than a metadata artefact, this deposit contains cross-line replication that nobody has used - which would be a considerably more valuable finding than the epitope control.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The N-terminal MLL1 antibody detects both wild-type MLL1 and MLL::AF9, which is the basis for the epitope-specificity design; the C-terminal epitope is not present in the fusion.
MLL::AF9 degradation induces rapid changes in transcriptional elongation and subsequent loss of an active chromatin landscape (STAR Methods). · Olsen SN, Godfrey L, Healy JP, Choi YA, Kai Y, Hatton C, Perner F, Haarer EL, Nabet B, Yuan GC, Armstrong SA · Molecular Cell 2022What this evidence cannot show
The quoted sentence concerns MOLM13 rather than HCB1, and the authors do not state anywhere in the text that MLL1c reads wild-type MLL1 exclusively - that inference follows from the fusion's domain structure and from the antibody being raised against the C-terminus, which the GEO sample characteristics record as 'antibody: MLL1 (c terminus), Active Motif Cat# 61296'. The inference is sound but it is an inference.
Evidence that argues against this (1)
- An N-terminal MLL1 antibody in these cells reads TWO wild-type alleles plus a transduced transgene, so wild-type signal may dominate its peak set. If so, the N-term arm is not meaningfully a fusion-containing channel, and the expected direction of the control was wrong to begin with - which would make the 'inversion' no inversion at all. NCBI Gene Expression Omnibus 2022
- E2weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.dataset
Three hours after fusion degradation the C-terminal MLL1 epitope retains 68.8% of its strongest peak calls and the N-terminal epitope 78.4%, inverting the expected direction of the specificity control.
GSE173599, samples GSM5696327 / GSM5696328 (MLL1c, C terminus) and GSM5696329 / GSM5696330 (MLL1, N terminus) · Olsen SN, Armstrong SA, et al. · NCBI Gene Expression Omnibus 2022What this evidence cannot show
CONFIDENCE IS DELIBERATELY WEAK AND THE RESULT IS REPORTED AS UNINTERPRETABLE. The four samples' GEO source_name fields read HCB2/HCB3/HCB4/HCB5 while their treatment_protocol says HCB1, so the cell line is unresolved. The two epitopes use different antibodies from different vendors in different sequencing batches. The N-term DMSO arm has only 887 peaks, the smallest reference set in the deposit, so its top decile is 88 peaks. The MLL1c DMSO library is the shallowest in the deposit at 3.77 M fragments against its treated arm's 7.81 M. n = 1 throughout.
Evidence that argues against this (1)
- This graph already holds ana-q120-a-menin-inhibitor-evicts-wild-type-dominated-mll1-sites-harder-than-the-tagged-fusions-own-sites, which reports the same counter-intuitive direction under a drug. Two independent observations of the same inversion is a reason to investigate rather than dismiss, and this entry's refusal to interpret should not be read as a claim that the inversion is false. Oncofusion knowledge graph 2026
An attempted wild-type-MLL1 specificity control using a C-terminal epitope the fusion does not contain. It inverted - the wild-type-only arm lost MORE (68.8%) than the fusion-containing arm (78.4%) - and is reported as uninterpretable, not interpreted.
The first of three reasons this arm is not interpreted: the deposit names HCB2-HCB5 in source_name and HCB1 in treatment_protocol for these four samples, and the filenames carry neither answer.
SECOND INDEPENDENT SIGHTING OF THE SAME INVERSION: that entry found wild-type-dominated MLL1 sites more labile than the tagged fusion's own under a drug; this found the wild-type-only epitope more labile under a degron. Neither was looking for it, which is a reason to investigate rather than dismiss.