Oncofusion

start / graph / ana-q125-two-replicates-of-one-culture-disagree-in-sign-and-span-four-times-the-corpus-indifference-threshold

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q125-the-h2ak119ub-arm-of-the-pcgf1-knockout

The first replicated share ratio this corpus has ever had: two title-declared replicates of ONE murine culture span 0.2623 on the same quantity - 4.5x the 0.0584 between-culture spread this corpus uses as its indifference threshold - and they disagree in SIGN about whether the PCGF1 knockout blunts the DOT1L-inhibitor response

also called the replicate spread nobody had · 0.0584 is too small · an average across a sign disagreement

Almost every chromatin number in this project comes from a single sample, so nobody knew how much such a number wobbles. Here the same measurement exists twice, and the two copies disagree by more than four times the amount this project treats as meaningful - and on one question they point in opposite directions.

The same thing, for a scientist

**The H2AK119ub NTvKO design is the first arm in this project with more than one track per condition per genotype**, so it is the first chance to put a spread on a within-sample share ratio. After criterion I4 disqualified the `rep1` convention (its own TDF header names a FLAG ChIP), two conventions remain - `paired` (`.filt.bam.tdf`, filename family `F3_NT`/`F3_C1`) and `rep2` (bare `.tdf`, family `NT`/`CLONE1`) - so what follows is honestly **a difference between two tracks, not a spread**, and criterion S3 was recorded UNSCOREABLE for that reason [E1]. | quantity at `S_FLAG`, murine MLL-AF9 | `paired` | `rep2` | difference | |---|---:|---:|---:| | `R_H2AK119ub(VTP/DMSO)`, control `NT` | 1.1694 | 0.9071 | **0.2623** | | `R_H2AK119ub(SGC/DMSO)`, control `NT` | 1.4932 | 1.3855 | 0.1077 | | `R_H2AK119ub(SGC/DMSO)`, PCGF1-KO `C1` | 1.3037 | 1.5298 | 0.2261 | **Q122's 0.0584 - the measured between-culture absolute difference this corpus adopted as its 'PCGF1-independent' and 'replicates agree' threshold - is smaller than all three.** Q124's offered criterion H4 asked for a range below 0.0584 across conventions; the range is 0.2623 [E1]. **And the sign disagreement is the sharper result.** Does the knockout blunt the SGC0946-induced H2AK119ub gain? `paired` says yes by **+0.1895**; `rep2` says the knockout's gain is LARGER by **0.1443**. The mean over the two conventions is **+0.0227** and scores 'DOES NOT FIRE' - *an average across a sign disagreement*, and it would have been this beat's headline had the conventions not existed [E1]. **At the most-evicted quartile the two conventions agree**: NT - C1 = **+0.4104** (`paired`) and **+0.1463** (`rep2`), both blunting. **So the disagreement is a property of the aggregate statistic, not of the biology** - which is the same conclusion the gradient entry reaches by a different route [E2].

Go deeper
What this number is and is not, because it is confounded three ways

It is an upper bound on technical noise and a lower bound on nothing. The two conventions differ in at least three respects at once and the design cannot separate them:

  1. Processing state. paired files are .filt.bam.tdf; rep2 files are bare .tdf. Q117 measured that this series' processing states differ in blacklist load by an order of magnitude in the trimmed_* family, though both families here are ChIP_* and carry 1.4-1.9%.
  2. Filename family. F3_NT/F3_C1 against bare NT/CLONE1 - the distinction Q122 section 2a and Q124 section 2b established marks different cultures elsewhere in this same series.
  3. What 'replicate' means. Nothing in the 42 NTvKO records says whether these are technical repeats, biological repeats or related cultures, and antibod occurs zero times in the whole dump.

So 0.2623 is not a technical standard deviation and must not be quoted as one. What it licenses is narrower and still useful: no single-track share ratio in this corpus should be trusted to better than a couple of tenths at this set, and that includes Q122's 0.0584, which was itself computed from two single tracks and is the threshold most of this corpus's genotype and cross-partner verdicts are scored against.

The consequence for the thresholds this corpus uses

Two thresholds are in play across Q118, Q120, Q122, Q124 and this beat:

  • 0.0584, Q122's between-culture absolute difference at this set, used as the 'they agree' / 'independent of' bound;
  • 0.2089, Q120's RS4;11-vs-MOLM13 cross-partner difference, used as the 'meaningful effect' bound.

The measurement here puts a two-track difference of 0.2623 inside ONE culture, which exceeds BOTH. That does not invalidate the earlier verdicts - they were scored on the quantities available - but it means an UNDECIDED verdict in the 0.0584-0.2089 band is, at this set and in this channel, indistinguishable from track-to-track variation. The band should be read as 'below the resolution of a single track' rather than as an intermediate effect size.

Facts and figures
scriptanalyses/q125-the-h2ak119ub-arm-of-the-pcgf1-knockout/deviation2_quartiles_and_conventions.py
statusrun
result summaryTwo conventions of one culture differ by 0.2623 on R_H2AK119ub(VTP) at S_FLAG against a 0.0584 corpus threshold, and disagree in sign on the SGC genotype contrast at the aggregate (+0.1895 vs -0.1443) while agreeing at the most-evicted quartile (+0.4104 vs +0.1463).
q124 offered H4 verdictREPORTED AS A DIFFERENCE, NOT SCORED - fewer than three usable conventions
absence classmeasured-not-analysed
absence class noteReplicate tracks for a chromatin mark in this series were deposited in 2024 and public from 2025-11-23 and had never been binned, so this corpus has quoted single-track share ratios for five cycles with no measurement of their reproducibility. GSE260742's H2AK119ub NTvKO arm is the only deposit identified that supplies one. What remains NEVER-MEASURED is what 'paired', 'rep1' and 'rep2' mean: no record in the series says whether they are technical or biological repeats.
absence class assigned2026-09-23 (Q125)
measured ranges at S FLAG
{
  "R_VTP_control": [
    1.1694,
    0.9071
  ],
  "range": 0.2623,
  "R_SGC_control": [
    1.4932,
    1.3855
  ],
  "R_SGC_knockout": [
    1.3037,
    1.5298
  ]
}
How this could bite us back
  • If a single-track share ratio at this set has a resolution of roughly a quarter, then any pharmacodynamic chromatin marker built on one ChIP per patient sample has the same resolution, and a 20-30% drug effect is inside it. The clinically useful version of any readout in this corpus needs replicates or a spike-in, and this series has neither for most of its arms.
Still open
  1. Should the 0.0584 threshold be retired in favour of a within-arm resolution estimate? Every verdict in the 0.0584-0.2089 band across Q118, Q120, Q122, Q124 and this beat would need re-reading as 'below resolution' rather than 'intermediate'.
  2. Is the H3K79me2 arm's single-track variability the same? The H3K79me2 NTvKO design has one convention, so the only way to bound it is a second series - and none has been identified.
Every source behind this entry (2)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    In GSE260742's H2AK119ub NTvKO arm, two title-declared replicate conventions of the non-targeting control culture give R(VTP/DMSO) at the FLAG-defined top-0.1%% set of 1.1694 (paired, GSM8123403/406) and 0.9071 (rep2, GSM8123420/422), a difference of 0.2623; R(SGC/DMSO) 1.4932 and 1.3855 in the control and 1.3037 and 1.5298 in the PCGF1-knockout, so the control-minus-knockout SGC difference is +0.1895 in one convention and -0.1443 in the other.

    What this evidence cannot show

    NO SPIKE-IN NORMALISATION IS POSSIBLE. Q117 verified zero hits for the entire spike-in vocabulary across all 419,824 characters of GSE260742's 158 sample records and no scale factor is deposited, so every cross-condition number here is a WITHIN-SAMPLE SHARE, which cancels the unknown per-sample scale factor exactly and IS BLIND TO A UNIFORM CHANGE BY CONSTRUCTION. Nothing here measures any mark's abundance; everything measures its distribution. EXPOSURE ASYMMETRY: DMSO 72 h and SGC0946 5 uM 72 h are matched, VTP-50469 500 nM is 48 h, so every VTP ratio is a LOWER BOUND on the 72-hour effect - identical in both genotypes, so the genotype DIFFERENCE is unaffected. REAGENTS: H2AK119ub is Ubiquityl-Histone H2A (Lys119) (D27C4) XP Rabbit mAb, NEB 8240S, and it CANNOT distinguish PRC1.1 from canonical PRC1 - RING1A/B is the ligase in both and PCGF1 is one of six PCGF paralogs - nor deposition from loss of a deubiquitinase. The H3K27me3 and H3K4me3 reagents are NOT RESOLVED and are recorded unresolved rather than guessed; `antibod` occurs zero times in the 158 GEO records. S_FLAG is defined by DYKDDDDK Tag (D6W5B) Rabbit mAb, NEB 14793S, on a RETROVIRALLY OVER-EXPRESSED tagged MLLAF9 construct, so it is the over-expressed molecule's sites and not an endogenous fusion's. THE KNOCKOUT is one CRISPR guide, one clone C1, no second clone, no rescue and NO deposited PCGF1 protein or transcript measurement. SYSTEM: murine MLL-AF9 (KMT2A::MLLT3) myeloid blood cells, NOT the infant B-lineage KMT2A::AFF1 disease this graph is about. TWO conventions is a DIFFERENCE and not a spread, and it is reported as one throughout: no standard deviation, no standard error, no confidence interval is computed from n = 2. The third title-declared convention (rep1, GSM8123407-412) was disqualified by criterion I4 because its own TDF-internal trackNames names a FLAG ChIP, so this comparison could have had three arms and has two. The two conventions differ in processing state and filename family as well as in replicate identity and the design cannot separate those.

    Evidence that argues against this (2)
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    At the most-evicted eviction quartile the two conventions agree in sign about the knockout: the SGC0946-induced H2AK119ub gain is 1.9183 in the control against 1.5079 in the knockout in the paired convention and 1.6978 against 1.5515 in rep2, so the control exceeds the knockout by +0.4104 and +0.1463 respectively.

    What this evidence cannot show

    The quartile stratification is Q117's and was recorded in that beat as its own DEVIATION 3, so it was not pre-registered there either. Applying a threshold to the quartile-level genotype difference is POST HOC in this beat and is labelled so. n = 1 per convention per cell.

    Evidence that argues against this (1)
Connected to