Oncofusion

start / graph / ana-the-rs4-11-h3k79me2-table-in-gse149183-is-two-thirds-molm13

Analysisestablished confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.Q138-absolute-h3k79me2

The deposited RS4;11 H3K79me2 table in GSE149183 is two-thirds MOLM13: 38,580 of its 57,794 rows are byte-for-byte identical to the MOLM13 file, so the index fusion is measurable at only 19,214 TSS entries

also called the RS4;11 sitepro file is contaminated · 38,580 of 57,794 rows are MOLM13 · only 19,214 genuine RS4;11 rows

A public data file named for the leukaemia cell line this project cares about most turns out to contain another cell line's numbers in two-thirds of its rows - the same values, digit for digit. Anyone who used the whole file, as this project did in an earlier cycle, was mostly measuring the wrong cells.

The same thing, for a scientist

**Statement.** `GSE149183_RS4_11_H3K79me2_TSS-ChIPseq_signal_+_dm6_spike-in_1000u_3000d_res100_sitepro.txt.gz` has 57,794 data rows but only **45,169 distinct `Gene_TSS` keys**; 12,625 keys appear exactly twice with different values. The cause is not duplication: **rows 19,215-57,794 (38,580 rows, 66.8%) are byte-for-byte identical in all nine H3K79me2 columns to the MOLM13 file**, and the contaminated block is perfectly contiguous. Only the first 19,214 rows carry values that differ from MOLM13 and can therefore be RS4;11. **The defect is in the deposit as served today, not in our copy.** Both files were re-downloaded from the NCBI FTP site on 2026-09-23; the md5s match the local copies exactly (`da6d9a4accb9666e0aec3f026e97efba` for RS4;11, `d41582b3755bd3a81c1e985be1ab0f09` for MOLM13). The MOLM13 file itself is clean: 57,794 rows, 57,794 distinct keys. **The dm6 spike-in table in the same sheet is NOT affected.** It sits in the right-hand columns, carries `RS411_ChIP_H3K79me2_CST_*` sample names, and is RS4;11's own. The earlier entry built on those ratios alone is unharmed by this; the entry built on the per-gene columns is not. **Consequence.** Every RS4;11 per-gene statement from this deposit is a statement about 19,214 TSS entries, 33.2% of the annotation, and **that subset is not a random sample**: its median DMSO baseline is 107.6 against 53.6 for the full annotation, so it is enriched about two-fold for signal-bearing TSSs.

Go deeper
How it was detected, and why an offset was not trusted

The contamination is re-detected from the data on every run rather than hard-coded: a row is called contaminated if its Gene_TSS key is present in the MOLM13 file and all nine of its signal values equal MOLM13's exactly. That test found 38,580 rows, and the first of them is at index 19,214 with no non-contaminated row after it - a perfectly contiguous block. The first row of the sorted block, DCLK3_36781352, is the single row in that region that is not identical to MOLM13, which is why the retained count is 19,214 and not 19,213.

The block is also sorted alphabetically from DCLK3_36781352 to the end, while the first 19,214 rows are unsorted - the signature of a paste over a sorted table. 1

What this invalidates in this graph

analyses/what-a-share-does-to-a-known-collapse/run.py (cycle 110) loaded all 57,794 rows of the RS4;11 file and reported column sums, stratified share ratios and stratified 'absolute' ratios for RS4;11 from them. Two-thirds of the rows behind every RS4;11 number in that analysis are MOLM13. The MOLM13 arm of that analysis is unaffected by this defect (though see the sibling entry on scaling, which affects both arms).

The numbers are reproduced exactly in this beat's results.json under E_correction_to_cycle_110, so a reader can see the old and new tables side by side. 1

What is NOT claimed

This does not say the paper's conclusions are wrong. The paper's figures were made from the primary alignments, not from this summary sheet, and the 19,214 genuine rows behave exactly as an RS4;11 H3K79me2 experiment should. It says that this one supplementary summary table is mis-assembled, and that anyone reusing it must drop the duplicated block. It is a data-reuse hazard, not a challenge to 2.

Facts and figures
scriptanalyses/absolute-h3k79me2-in-the-index-fusion/run.py
statusrun
result summary38,580 of 57,794 rows (66.8%) of the RS4;11 file are identical to MOLM13 in all nine signal columns; 19,214 rows retained as genuine RS4;11; retained subset median DMSO baseline 107.6 vs 53.6 for the full annotation.
rows total57794
rows contaminated38580
rows retained19214
PCT contaminated66.79
distinct keys in rs411 file45169
keys appearing twice12625
contaminated block starts at row index19214
md5 rs4 11da6d9a4accb9666e0aec3f026e97efba
md5 molm13d41582b3755bd3a81c1e985be1ab0f09
md5 verified on2026-09-23
inputs
[
  "GSE149183_RS4_11_H3K79me2_TSS-ChIPseq_signal_+_dm6_spike-in_1000u_3000d_res100_sitepro.txt.gz",
  "GSE149183_MOLM13_H3K79me2_TSS-ChIPseq_signal_+_dm6_spike-in_1000u_3000d_res100_sitepro.txt.gz"
]
How this could bite us back
  • This is a reuse hazard, not an intervention claim, but it bites an intervention indirectly: a pharmacodynamic threshold for a DOT1L inhibitor derived from this table for the index fusion would be derived mostly from an AML line.
Still open
  1. Do the per-sample .tdf tracks (GSM4491635-GSM4491643) reproduce the 19,214 genuine rows, which would confirm the retained block is RS4;11 rather than a third source?
  2. How many other published reanalyses have used this file whole? A citation trace on GSE149183 would bound the spread of the defect.
  3. Should the corresponding author be notified? The defect is cheap to fix and the corrected table would restore 38,580 RS4;11 measurements that currently do not exist in public form.
Every source behind this entry (3)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    The series-level supplementary file named for RS4;11 in GSE149183 contains MOLM13 values, byte-for-byte, in 38,580 of its 57,794 data rows, as served by NCBI on 2026-09-23.

    GSE149183 series supplementary files: MOLM13 and RS4_11 H3K79me2 TSS-ChIPseq signal + dm6 spike-in sitepro tables · Perner F, Gadrey JY, Xiong Y, et al. · NCBI Gene Expression Omnibus 2020
    What this evidence cannot show

    This is a direct property of the file, re-downloaded and md5-checked on 2026-09-23, not an inference. If NCBI later replaces the file the md5 will differ and this entry must be re-run - that falsifier is written into the beat's PREREGISTRATION.md.

    Evidence that argues against this (1)
    • The alternative reading - that the MOLM13 file is the contaminated one and the RS4;11 file is correct - is excluded by the MOLM13 file being internally clean (57,794 rows, 57,794 distinct keys) while the RS4;11 file carries 12,625 doubled keys. The contamination is one-directional. NCBI GEO FTP 2020
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    GSE149183 is the ChIP-seq dataset of Perner et al., Blood 2020, which measured H3K79me2 at days 3, 6 and 9 after DOT1L-inhibitor treatment in MOLM13 and RS4;11 with Drosophila spike-in chromatin.

    Novel inhibitors of the histone methyltransferase DOT1L show potent antileukemic activity in patient-derived xenografts. · Perner F, Gadrey JY, Xiong Y, Hatton C, Eschle BK, Weiss A, et al. · Blood 2020
    PMID 32575123DOI
    What this evidence cannot show

    PMID and DOI fetched from NCBI eutils on 2026-09-23, not recalled. The GEO series overall design string was read from the GSE149183 SOFT family file, also fetched on 2026-09-23.

    Evidence that argues against this (1)
    • No published correction or erratum to this paper was found, and none would be expected - the defect is in a convenience summary table, not in the analysis the paper reports. NCBI GEO 2020
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    RS4;11 (CVCL_0093) carries an in-frame KMT2A::AFF1 fusion and is a B-ALL line with t(v;11q23.3); MOLM-13 (CVCL_2119) carries KMT2A::MLLT3 and is an acute monocytic leukaemia line.

    What this evidence cannot show

    Fetched from the Cellosaurus REST API on 2026-09-23. The fusion line in CVCL_0093 reads 'Gene fusion; HGNC:7135 AFF1 + HGNC:7132 KMT2A; Name(s)=KMT2A-AFF1, MLL-AFF1, ALL-1/AF4; Note=In frame', citing PMID 1423625, 8358709 and 32721124.

    Evidence that argues against this (1)
    • The two lines differ in lineage, donor, sex and age as well as in fusion partner, so no comparison between them isolates the partner. That bound is written into every cross-line statement in this beat. Cellosaurus, SIB 2026
Connected to