start / graph / ana-q143-menin-enl-and-dot1l-are-cut-and-tag-and-the-kmt2a-antibodies-are-cut-and-run-three-weeks-apart

AnalysisstrongQ143-menin-under-drug-in-infants

Menin, ENL and DOT1L are CUT&Tag and the four KMT2A antibodies are CUT&RUN, run three weeks apart - so every menin-versus-fusion comparison in this pair crosses both assay and batch

The experiment that asks 'is menin sitting where the fusion is' compares two measurements made with different chemistries on different days. That does not make the comparison wrong, but it means a difference between menin and the fusion could be a difference between two methods.

The same thing, for a scientist

Within each of 148752 and 152985 the ten libraries split into two protocols and two sequencing batches: **AutoCUT&Tag, batch `240423`** for menin (Bethyl A300-115A), ENL (CST D9M4B) and DOT1L (CST D4O2T), two replicates each; and **AutoCUT&RUN, batch `240515`, 500K cells** for KMT2A-N1 (Millipore 05-764), N2 (CST D2M7U), C1 (Santa Cruz H10) and C2 (Millipore 05-765). The library sizes differ accordingly - the CUT&RUN libraries carry 14-18 million placed fragments and the CUT&Tag libraries 1.3-5.9 million. The promoter Spearman matrix over 19,404 genes does **not** cluster by assay: in 148752 menin correlates with KMT2A-N2 at 0.763 and with its own replicate at 0.69, so target identity dominates protocol. That is the pre-registered gate for this confound and it passes - but it passes in 148752 only.

Go deeper

The split, from the sample records

Every library's protocol is stated in !Sample_extract_protocol_ch1 and its batch in the parenthesised suffix of !Sample_title 1:

channel	protocol	batch	antibody
menin R1, R2	AutoCUT&Tag	240423	Bethyl A300-115A
ENL R1, R2	AutoCUT&Tag	240423	CST D9M4B
DOT1L R1, R2	AutoCUT&Tag	240423	CST D4O2T
KMT2A-N1	AutoCUT&RUN	240515	Millipore 05-764 (mouse)
KMT2A-N2	AutoCUT&RUN	240515	CST D2M7U (rabbit)
KMT2A-C1	AutoCUT&RUN	240515	Santa Cruz H10 (mouse)
KMT2A-C2	AutoCUT&RUN	240515	Millipore 05-765 (mouse)

The pre-registered gate, and what it says

The gate was: if the promoter correlation matrix clusters by ASSAY rather than by TARGET, the co-localisation question cannot be asked in this deposit. It does not, in 148752 2:

the two N-terminal reagents agree at 0.909 and the two C-terminal reagents at 0.935;
menin R1 (CUT&Tag) against KMT2A-N2 (CUT&RUN) is 0.763, and menin R2 against the same channel is 0.770, higher than menin against its own CUT&Tag replicate (0.691);
DOT1L, also CUT&Tag, agrees with its own replicate at only 0.311 and with KMT2A-N2 at 0.481.

So the matrix is organised by how much real signal a library has, not by protocol. In 152985 it is organised by nothing: see ana-q143-the-152985-cut-and-tag-replicates-agree-at-0-11-to-0-25.

What the split still costs

Two things survive the gate. First, the two protocols differ in what they call background, and there is no IgG for either. Second, no menin-versus-KMT2A ratio in this pair is a same-tube measurement - they are different aliquots of a banked primary sample processed three weeks apart, so a difference in cell viability or nuclear integrity between the two thaws is fully confounded with any menin-versus-fusion difference. Where this beat reports menin against KMT2A-N it reports both as enrichment over a common reference rather than as a ratio to each other.

Facts and figures

gate	does the promoter correlation matrix cluster by assay rather than target?
gate result 148752	no - organised by target and depth
residual confound	no menin-vs-KMT2A number in this pair is a same-tube measurement
status	run
script	analyses/menin-under-revumenib-in-an-infant-patient-pair/run.py

key correlations 148752

protocols

{
  "AutoCUTnTag": [
    "MENIN",
    "ENL",
    "DOT1L"
  ],
  "AutoCUTnRUN": [
    "KMT2A_N1",
    "KMT2A_N2",
    "KMT2A_C1",
    "KMT2A_C2"
  ]
}

batches

{
  "AutoCUTnTag": "240423",
  "AutoCUTnRUN": "240515"
}

placed fragments millions

{
  "CUTnRUN_range": [
    14.2,
    17.5
  ],
  "CUTnTag_range": [
    1.26,
    5.93
  ]
}

key correlations 148752 — the numbers behind the chart

{
  "N1_vs_N2": 0.909,
  "C1_vs_C2": 0.935,
  "MENIN_R1_vs_R2": 0.691,
  "MENIN_R1_vs_N2": 0.7628,
  "MENIN_R2_vs_N2": 0.7703,
  "DOT1L_R1_vs_N2": 0.4811,
  "DOT1L_R1_vs_R2": 0.311
}

How this could bite us back

A cross-protocol occupancy ratio quoted as a biological fact will be quoted again without its protocol footnote, and this pair is the only infant material of its kind, so it will be quoted.
If a follow-up experiment is designed to 'reproduce' this in the index fusion, it should run menin and KMT2A in the SAME protocol on the SAME aliquot, which this deposit does not do.

Still open

Is there any target measured by BOTH AutoCUT&RUN and AutoCUT&Tag in this deposit, which would give a direct protocol-transfer estimate? None was found for these two samples.
Do the two thaws three weeks apart differ in viability, and did the submitters record it? Nothing in GEO says.

Every source behind this entry (2)

E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
The ten libraries of each sample split into AutoCUT&Tag (menin, ENL, DOT1L; batch 240423) and AutoCUT&RUN (four KMT2A antibodies; batch 240515).
GSE252378_family.soft.gz - the per-sample SOFT records for GSE252378, read raw with line numbers · Janssens DH et al. · NCBI GEO FTP 2024
What this evidence cannot show
Protocol read from `!Sample_extract_protocol_ch1`, batch from the title suffix. The batch code is a submitter convention, not a GEO field, so it is a strong inference rather than a stated fact.
Evidence that argues against this (1)
- The batch code is a submitter naming convention embedded in the sample title, not a GEO field, so the inference that two libraries were sequenced together is strong but not stated; a submitter could reuse a date prefix across runs. Janssens DH et al. 2024
E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Over 19,404 promoters in 148752 the correlation matrix is organised by target and library depth rather than by protocol: menin-vs-KMT2A-N2 across protocols is 0.763, above menin's own within-protocol replicate agreement of 0.691.
analyses/menin-under-revumenib-in-an-infant-patient-pair - pre-registered re-analysis of GSE252378 samples 148752 and 152985 · Q143-menin-under-drug-in-infants (Oncofusion Knowledge Graph) · this repository 2026
What this evidence cannot show
Spearman on CPM over a fixed filter-free promoter set; n = 1 sample. A single high cross-protocol correlation does not prove protocol-independence, it only fails to show protocol dependence.
Evidence that argues against this (1)
- Menin and KMT2A physically occupy the same promoters, so a high menin-vs-KMT2A correlation is the expected biology and cannot distinguish 'the protocols agree' from 'the proteins co-localise'. The clean version of this control would be the same target in both protocols, which this deposit does not contain for any target. Kaya-Okur HS et al. 2019

Connected to

depends on →148752 and 152985 are one patient before and during revumenib, not two treated patients - so the drug-free arm this corpus said did not exist is sample 148752
the assay split is a property of the pair
related to →Neither infant sample has an IgG control, and neither of their two sequencing batches contains one - so every number from this pair is confounded with chromatin accessibility
the two structural limits of this deposit: no IgG, and a cross-protocol comparison
← depends onIn drug-naive primary infant KMT2A::MLLT1 B-ALL, menin is as enriched as the fusion N-terminus at the top fusion promoters (2.73x vs 3.12x over an accessibility reference) while the wild-type C-terminal reporter sits at background (1.10x)
the menin-versus-N comparison crosses assay and batch