Oncofusion

start / graph / exp-the-free-prc1-1-intersection-is-a-cluster-job-not-a-desk-job

Experimentestablished confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.

Half the 'free' PRC1.1 intersection really was free: Zhou's locus set is a publisher table, Neville's does not exist, and only that half costs 25 GB

also called PRC1.1 locus intersection · GSE260742 · GSE249076 · free experiment costing

CORRECTED 2026-09-20: two papers disagree about where a protein complex acts after a menin inhibitor, and this project spent three cycles saying the disagreement could be settled free by comparing the location lists each published. One of the two did publish its list at the journal and the other did not, so half the comparison costs nothing and the other half still needs a real pipeline run on 25 gigabytes.

The same thing, for a scientist

**CORRECTED 2026-09-20, same day.** The publisher supplements this entry named as unchecked have now been checked. **Zhou's locus set IS published** - supplemental Table 4, 9,292 genes, 356 KB, free. Neville's is not. So **one half of the intersection costs nothing and the other half is still the cluster job described below.** See [[ana-the-locus-list-existed-at-the-publisher-not-in-geo]]. Q52 proposed intersecting the deposited locus sets of PMID 41634341 (Neville) and PMID 39158067 (Zhou) as a **free first step** deciding whether they disagree about PRC1.1's position, and I carried that into three cycle logs and the ledger. **The data are public; the locus sets are not in GEO - but Zhou's is at the publisher.** | series | study | content | |---|---|---| | GSE260742 | Neville | ChIP-seq, 158 samples, **`.tdf` coverage tracks** + 21.4 GB RAW tar | | GSE249076 | Zhou | ChIP-seq, 12 samples, **`.bigWig` coverage tracks** + 3.3 GB RAW tar | **Coverage is continuous signal, not a locus set.** No `.bed`, no `.narrowPeak`, no gene list in either. The intersection requires peak calling first.

Go deeper
What the deposits actually contain

Both studies deposited properly and both are retrievable - the problem is the form, not the availability.

  • GSE260742 (Neville, ChIP-seq, 158 samples): .tdf files, the IGV coverage format, 120-270 MB each, named for the antibody and condition - ChIP_K119UB_K562_WT_DMSO, ChIP_K79_HUDEP_SGC. Plus GSE260742_RAW.tar at 21.4 GB.
  • GSE249076 (Zhou, ChIP-seq, 12 samples): .bw bigWig tracks, 180-390 MB each - A2_LD_119ub.bw, A2_PD_MENIN.bw. Plus GSE249076_RAW.tar at 3.3 GB.
  • The two RNA-seq series, GSE260456 and GSE249078, expose no processed matrix on their accession pages either.

So the comparison Q52 described cannot be done FROM GEO, because neither study deposited a list there. It can be done halfway from the publishers: Zhou supplemental Table 4 is a 9,292-row gene-level list, and Neville has no equivalent in GEO or at the journal. Deriving one means calling peaks against input, which is an alignment-and-peak-calling pipeline with its own parameter choices, and those choices are exactly what would determine whether the two sets appear to overlap.

What it actually costs, and where it should run

Disk. GSE260742_RAW.tar is 21.4 GB; this machine's working drive has 20 GB free. The larger archive does not fit, before counting the space peak calling needs for intermediates.

The cloud drive does not solve it. This box has multiple terabytes of network-backed storage available, and the project's own standing guidance is that it is for cold archival data and explicitly not for heavy or random I/O during compute. Staging a peak-calling run against it would be slower than not doing it.

So this is a cluster job. The project has SSH access to a Slurm GPU/CPU cluster, which is the right venue: download both RAW archives there, align, call peaks against the matched inputs with one parameter set applied identically to both studies, and intersect. That last point is the methodological crux - using each study's own published parameters would reproduce their conclusions rather than test them.

A cheaper partial test exists. The coverage tracks alone can be compared at named loci without peak calling: take the genes each paper names in its text, extract signal at those coordinates from both studies' bigWigs, and ask whether the two agree where they overlap. That is a few hundred megabytes rather than 25 GB, it does not decide the general question, and it would establish whether the disagreement survives at the loci both papers actually discuss.

Facts and figures
geo series
{
  "neville_chip": "GSE260742",
  "neville_rna": "GSE260456",
  "zhou_chip": "GSE249076",
  "zhou_rna": "GSE249078"
}
raw archive bytes
{
  "GSE260742_RAW.tar": 21417431040,
  "GSE249076_RAW.tar": 3323627520
}
How this could bite us back
  • This corrects a recommendation this project made three times, including in the coverage ledger, that the PRC1.1 disagreement could be settled for free with no bench work. It cannot, as described. The question remains worth answering and the route is a cluster job with one parameter set applied identically to both datasets - using each study's own parameters would reproduce their conclusions rather than test them.
  • And this entry was itself corrected within a day of being written, by checking the gap its own caveat named. The claim in E1 was scoped to GEO and survived; the summaries were not scoped and did not. **Scope the summary as tightly as the claim** - the summary is what gets read.
Still open
  1. ANSWERED 2026-09-20: Zhou's publisher supplemental Table 4 is the gene list; Neville's supplement has none. The remaining cost is entirely on the Neville side.
  2. Does a per-GSM supplementary file carry processed calls where the series listing does not?
  3. Would the coverage-track comparison at named loci - a few hundred megabytes - be enough to show whether the disagreement survives where both papers actually look?
Every source behind this entry (2)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    The GEO deposits supporting both PRC1.1 studies contain coverage tracks rather than locus calls: GSE260742 provides .tdf files across 158 ChIP-seq samples with a 21.4 GB raw archive, and GSE249076 provides .bigWig files across 12 samples with a 3.3 GB raw archive, with no BED, narrowPeak or gene-list file in either, so an intersection of deposited locus sets cannot be performed as described.

    What this evidence cannot show

    Retrieved from the NCBI GEO FTP filelists on 2026-09-20. **Established for what the series-level supplementary listings contain.** Two limits: per-SAMPLE supplementary files were not enumerated exhaustively, so a processed peak file attached to an individual GSM rather than the series would have been missed; and **the papers' own supplementary tables, which are on the publisher's site rather than in GEO, were not checked** and are the likeliest place a locus list would live. Both are cheap to resolve and would weaken this finding rather than overturn its cost conclusion.

    Evidence that argues against this (1)
    • CONFIRMED 2026-09-20: this note was right. Zhou's publisher supplement carries exactly the gene list GEO does not - supplemental Table 4, 9,292 genes in three menin/H2AK119ub groups. Neville's supplement carries MAGeCK CRISPR output and source data only. So this entry establishes that the GEO route is expensive, not that the question is, and for Zhou's half it is not expensive at all. Neville D et al. 2026
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.analysis

    The local working disk has approximately 20 GB free, which is less than the 21.4 GB raw archive of the larger of the two required GEO series, so that archive cannot be staged locally before any processing space is accounted for.

    What this evidence cannot show

    Measured 2026-09-20. **This is a statement about one machine on one day** and not a general property of the experiment - a cluster or a larger disk removes it entirely. It is recorded because the recommendation it corrects was framed as costing nothing, and a step that cannot be run where it was proposed is not free.

    Evidence that argues against this (1)
    • Only the larger archive exceeds local disk; the 3.3 GB Zhou archive fits comfortably, and the tracks could be processed one series at a time or streamed without full staging. The disk limit is an inconvenience rather than a barrier, and the peak-calling pipeline is the real cost. Zhou X et al. 2024
Connected to