Before quoting anything, this beat re-derived six figures that two earlier beats had stored - three human, three mouse - from the deposited files end to end, and all six came back identical, which is the control that licenses every cross-beat comparison in the entry above it
also called the reproduction check · six stored numbers re-derived
Numbers in this project are often quoted from earlier work rather than recalculated. Before comparing anything, this beat recalculated six of them from the original public files, and every one matched.
The same thing, for a scientist
This beat's two headline results are both *comparisons against a sibling beat's stored measurement*. That is only safe if the pipeline is deterministic and the stored figures are not drifting, so six figures were re-derived from the deposited files before any comparison was made [E1]: | quantity | stored by | stored value | re-derived here | |---|---|---:|---:| | MOLM13 MLL1-N held-set share ratio | Q118 | 0.9438 | **0.9438** | | MOLM13 MLL1-N per-bin median share ratio | Q118 | 0.9991 | **0.9991** | | MOLM13 MLL1-N fraction still in own top 0.1% | Q118 | 0.6645 | **0.6645** | | murine FLAG held-set share ratio (post-blacklist) | Q117 | 0.1793 | **0.1793** | | murine FLAG_DMSO track total, coverage-bases | Q117 | 528,768,301 | **528,768,301** | | murine FLAG_VTP track total, coverage-bases | Q117 | 532,157,522 | **532,157,522** | Plus two coordinate checks: the mm10 `Myb` and `Meis1` refGene windows were **re-fetched live from the UCSC API in this beat** and matched Q117's stored windows to the base pair (chr10:21,119,930-21,162,984 and chr11:18,875,428-19,020,969), and the locus-level FLAG ratios came back 0.3623 and 0.1171 against stored 0.362 and 0.117. **Build verification was re-done from scratch for all twelve files** rather than inherited: each file's own per-chromosome declared lengths were compared against live hg19/hg38 or mm10/mm39 sizes, and every file called its expected build. A file closer to the wrong build on any chromosome would have aborted. **Why this is an entry and not a footnote.** The project's standing failure mode is a stored figure quoted as live. The opposite discipline - re-deriving before quoting - produced nothing surprising here, and *that is the result*: a clean reproduction is what makes the 0.209 H3K79me2 divergence and the 2.23-fold FLAG divergence readable as differences between systems rather than between runs.
What did NOT reproduce, and it is a correction rather than a drift
One stored mouse-side number is superseded rather than reproduced. Q118's cross-species locus table normalises the murine FLAG ratios by Gapdh at 0.508, describing it in its own text as 'two genes and the best available mouse-side normaliser', and Q117's memo had named a proper floor as its open gap.
Applying Q118's own aggregate-floor definition to the murine data gives FLAG 0.9295 over 505,306 bins and MLL1 0.9165 over 491,926 bins. The Gapdh stopgap is 1.83-fold too low, so every mouse-side floor-normalised figure in that table is inflated 1.83x. That is not a reproduction failure - both numbers are correct computations of different things - and it is reported for the coordinator in ana-q120-in-the-same-mouse-nuclei-mll1-diverges-from-flag-at-myb-and-agrees-at-meis1.
| script | analyses/q120-h3k79me2-second-line-and-the-mll1-epitope/arm1-molm13-ordering.py and arm2-mouse-epitope.py |
|---|---|
| status | run |
| result summary | Six stored figures re-derived, six matched; two stored coordinates re-fetched live and matched; twelve genome builds verified from file extents. One stored mouse-side floor (Gapdh 0.508) is superseded by an aggregate floor of 0.9295, a 1.83-fold correction reported separately. |
| n figures rederived | 6 |
| n matched | 6 |
| genome build | hg19 for ten human files and mm10 for two mouse files, each verified from its own per-chromosome declared lengths against live reference sizes |
| shown for fusion | instrument control; applies to KMT2A::MLLT3, KMT2A::AFF1 and MLL-AF9 arms alike |
| absence class | measured-not-reported |
| absence class note | Reproduction checks are generated by every re-analysis and almost never written down - the number was there to be recorded and nobody records it. The deposited accessions that make it checkable are GSE127507 and GSE260742. |
| absence class assigned | 2026-09-23 (Q120) |
- A perfect reproduction across three beats is also the signature of three beats sharing one instrument. The check licenses cross-beat comparison and says nothing about whether the instrument is right.
- Would a different binning tool - bigWigAverageOverBed, or deeptools on re-aligned reads - give the same numbers? Everything in this beat and its two predecessors runs through one Java TDFReader loop, so a systematic error in that loop would reproduce perfectly and invisibly.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Six figures stored by earlier beats were re-derived from the deposited files and matched exactly: MOLM13 MLL1-N held-set share ratio 0.9438, per-bin median 0.9991 and fraction still in own top 0.6645 (Q118); murine FLAG held-set share ratio 0.1793 and FLAG_DMSO/FLAG_VTP track totals 528,768,301 and 532,157,522 coverage-bases (Q117). The mm10 Myb and Meis1 refGene windows were re-fetched live from UCSC and matched Q117's stored windows to the base pair, and the locus FLAG ratios came back 0.3623 and 0.1171 against stored 0.362 and 0.117. All twelve files' genome builds were verified from their own per-chromosome extents.
Q120 re-analysis of GSE127507 MOLM13 arm; analyses/q120-h3k79me2-second-line-and-the-mll1-epitope/arm1-molm13-ordering.py and results-arm1.json, run 2026-09-23 · Iqbal Lab oncofusion knowledge graph 2026What this evidence cannot show
Four of the six are MOLM13 MLL1-N figures computed by a script derived from Q118's own, on the same machine, from the same deposited files - so this demonstrates determinism and provenance integrity, NOT methodological independence. The two murine track totals are a stronger check because they were computed by Q117's binner in a separate session and only read here. A genuinely independent check would use a different binning tool.
Evidence that argues against this (1)
the control that licenses reading the cross-line deltas as differences between systems rather than between runs
Q117's stored FLAG values and track totals reproduced exactly, so the FLAG side of the epitope comparison is not a re-run artefact
the same check re-run from scratch on ten more human files and two mouse files, all calling their expected build