Oncofusion

start / graph / ana-h3k79me2-has-two-replicate-channels-too-and-unlike-h2ak119ub-they-never-materially-disagree

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.

Answering a question three cycles filed and none ran: H3K79me2 also has two channels in GSE260742 - rep1 and rep2, 199 numeric leaves each - so cycle 104 quoted one of two available numbers without saying so. Unlike H2AK119ub they never materially disagree, 0 of 18 comparable ratios, and H3K27me3 and H3K4me3 were never scored under any second channel at all

also called the other marks check · H3K79me2 has a rep1 and a rep2 · 0 of 18 material disagreements · processing states are not replicates

One chemical mark in a public dataset was known to exist in two differently processed versions that sometimes disagree. Nobody had checked whether the other three marks in the same dataset have the same problem. One of them does have two versions - but they are two repeats of the experiment rather than two processing recipes, and they never disagree by enough to matter. The remaining two marks were only ever computed one way, so the check cannot say whether a second version exists, only that none was used.

The same thing, for a scientist

**Statement.** Cycles 104, 105 and 106 each filed the same open question - *do the other three marks in GSE260742 have two channels like H2AK119ub?* - and none ran it. H2AK119ub was checked twice only because that is where somebody happened to notice a disagreement, which is not a reason. **Run now across 276,074 numeric leaves in 169 analysis result files:** | mark | PAIRED | rep1 | rep2 | comparable ratios | sign disagreement | **material** | |---|---:|---:|---:|---:|---:|---:| | **H2AK119ub** | 416 | 308 | 403 | 68 | 51.5% | **2** | | **H3K79me2** | 0 | **199** | **199** | 18 | 11.1% | **0** | | H3K27me3 | 0 | 0 | 0 | — | — | — | | H3K4me3 | 0 | 0 | 0 | — | — | — | **`S2` fires: H2AK119ub is not the only mark with two channels.** **H3K79me2 has a rep1 and a rep2**, 199 numeric leaves each, and this project has been quoting one of them. Cycle 104's H3K79me2 panel - 0.8065 at the peaks, 0.8383 at D10 - **was one of two available numbers, and the entry did not say so.** **But the two cases are not the same kind of thing, and the distinction is the point.** H2AK119ub's `PAIRED` and `REP2` are **processing states** - `.filt.bam.tdf` normalised against `.filt.bam.tdf` inputs, versus bare `.tdf`. H3K79me2's `rep1` and `rep2` are **replicates**. Two runs of the same protocol are expected to agree; two different pipelines over the same reads are not the same claim. **And the second channel changes nothing. 0 of 18 comparable H3K79me2 ratios disagree materially**, against H2AK119ub's 2 of 68. The 11.1% sign flips are all inside the 10% floor. **So cycle 104 did under-report, and the number it reported survives the check** - the right correction is a provenance line, not a retraction. **H3K27me3 and H3K4me3 carry no channel token at all**, and that is a **weak negative** by this analysis's own `U1`: it shows no second channel was ever **computed**, never that none **exists** in the deposit.

Go deeper
The rule I wrote was too broad, and it hid the only positive result

The first version of this script excluded REP1 for every mark, on the strength of cycle 101: GSM8123407-412, titled H2AK119ub rep1 in GEO, are FLAG ChIPs, and Q125's criterion I4 dropped them.

That is a fact about six files, not about the string rep1. Applied blanket, it deleted H3K79me2's rep1 channel - 199 leaves - and the script reported H3K79me2 as single-channel. The first run returned S1, the reassuring answer, and it was an artefact of my own exclusion. Scoping the exclusion to H2AK119ub inverted the verdict to S2.

This is the second over-broad rule in two cycles: cycle 106's other beat excluded 1,054 of 1,290 fingerprints with a collision filter and reported 0 entries where 13 existed. Both filters failed in the same direction - they suppressed the finding and returned the comfortable answer.

Facts and figures
questiondo H3K27me3, H3K4me3 and H3K79me2 also have two channels in GSE260742?
answered afterfiled identically in cycles 104, 105 and 106 and never run
results files walked169
numeric leaves276074
verdictS2 - H2AK119ub is not the only mark with two channels, but H3K79me2's are replicates rather than processing states and they never materially disagree
what cycle 104 should have saidits H3K79me2 numbers - 0.8065 at the peaks, 0.8383 at D10 - are the value of one of two available channels; the other agrees, but the entry did not say a choice had been made
the negative is weakH3K27me3 and H3K4me3 carry no channel token, which shows no second channel was ever COMPUTED, not that none EXISTS
my own defectthe first version excluded REP1 for every mark on the strength of a fact about six H2AK119ub files, which deleted H3K79me2's 199-leaf rep1 channel and returned S1 - the reassuring answer - as an artefact of the exclusion
absence classmeasured-not-curated
bite backThis entry tells future beats that H3K79me2 is safe to quote from one channel because its two agree across 18 ratios. That licence rests on a small, non-independent denominator and a borrowed floor. If a later beat computes H3K79me2 at statistics outside this panel and the channels diverge there, this entry will have been the document that said not to bother checking.
channels by mark
H3K79me20.000H3K79me2: 0H3K27me30.000H3K27me3: 0H3K4me30.000H3K4me3: 0H2AK119ub416.0H2AK119ub: 416
channels by mark — the numbers behind the chart
{
  "H2AK119ub": {
    "PAIRED": 416,
    "rep1": 308,
    "rep2": 403
  },
  "H3K79me2": {
    "PAIRED": 0,
    "rep1": 199,
    "rep2": 199
  },
  "H3K27me3": {
    "PAIRED": 0,
    "rep1": 0,
    "rep2": 0
  },
  "H3K4me3": {
    "PAIRED": 0,
    "rep1": 0,
    "rep2": 0
  }
}
comparisons
{
  "H2AK119ub_PAIRED_vs_REP2": {
    "comparable": 68,
    "sign_disagreement_pct": 51.5,
    "material": 2
  },
  "H3K79me2_rep1_vs_rep2": {
    "comparable": 18,
    "sign_disagreement_pct": 11.1,
    "material": 0
  }
}
Still open
  1. Read the GEO processing notes for the H3K79me2 files. This entry infers 'replicates' from the rep1/rep2 naming, and cycle 101 is the standing proof that a GEO title can be wrong about exactly this.
  2. Do H3K27me3 and H3K4me3 have a second channel in the DEPOSIT that this project never computed? The file list answers it and nobody has looked; the negative here is weak by construction.
  3. Add a provenance line to cycle 104's four-mark panel recording that its H3K79me2 numbers are one of two available channels and that the other agrees.
  4. Re-score H3K79me2's 18 comparisons against a floor measured from a same-condition replicate rather than the borrowed 10%. GSE299100 supplies one.
Every source behind this entry (1)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.analysis

    Across 169 analysis results.json files and 276,074 numeric leaves, H2AK119ub carries numeric leaves under three channel tokens (PAIRED 416, rep1 308, rep2 403) and H3K79me2 under two (rep1 199, rep2 199); H3K27me3 and H3K4me3 carry none. Comparing each mark's two channels at matched ratio statistics: H2AK119ub PAIRED vs REP2 gives 68 comparable ratios, 51.5% sign disagreement and 2 material disagreements; H3K79me2 rep1 vs rep2 gives 18 comparable ratios, 11.1% sign disagreement and 0 material. Materiality requires both sides outside the 10% floor.

    What this evidence cannot show

    THE NEGATIVE FOR H3K27me3 AND H3K4me3 IS WEAK AND MUST NOT BE READ AS A CLEAN BILL. Absence of a channel token in this project's result files shows only that no second channel was ever COMPUTED; the deposit may hold one that nobody processed. Marks are matched by this project's own spellings (k27, k4, k79, k119), so a mark recorded some other way is invisible. H3K79me2's 18 comparable ratios come from a single beat's panel and are not independent of one another. The 10% materiality floor is BORROWED from input pairs bounding genotype variation and is used here for replicate and processing variation, which are different quantities - at a 3% floor some of H3K79me2's two sign flips could become material. Calling rep1/rep2 'replicates' for H3K79me2 is inferred from the naming and from cycle 101 having had to establish the opposite for H2AK119ub specifically; the GEO records for the H3K79me2 files were not re-read in this beat.

    Evidence that argues against this (3)
    • The headline result is a null and nulls of this shape are cheap to produce. Eighteen comparable ratios is a small denominator, drawn from one panel in one beat; a mark with a genuine replicate problem confined to statistics this project has not yet computed would look exactly like this. The correct reading is 'no disagreement has been demonstrated', not 'the channels agree'.
    • The distinction this entry leans on - processing states versus replicates - is inferred from token names, not verified against the deposit. If H3K79me2's rep1 and rep2 also differ in processing, the comparison is not the reassuring like-for-like one claimed here, and if H2AK119ub's PAIRED is itself a replicate-pair average then the two cases are closer than this entry says. Nobody has read the GEO processing notes for either.
    • This beat exists because the question was filed three times without being run, and it took under an hour once the matching code existed from another beat. That is a reason to be sceptical of the open-questions list as a work queue: the item was cheap the whole time and sat there because nothing in the process distinguishes a cheap open question from an expensive one.
Connected to