Oncofusion

start / graph / exp-q180-reanalyse-pxd009338-for-peptide-level-evidence-under-the-rank-cut

Experimentmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q180-the-af4-specific-interactors

Reopen PXD009338 at peptide level — the search results are already deposited, so no re-search is needed, but the deposit carries no sample-to-file mapping and that is the real cost

also called PXD009338 reanalysis · peptide-level evidence for the AF4-specific tail

The raw data behind the only MLL-fusion interaction map is public, and so are the search results, so checking how solid each protein identification is does not need the analysis to be redone from scratch. The catch is that the deposited files are not labelled with which fusion each one came from, and working that out is the actual work.

The same thing, for a scientist

A previous beat flagged re-searching PXD009338 below the rank cut as a cheap open gap. **Checked against the live PRIDE file listing on 2026-09-24, it is cheaper than that in one way and harder in another** [E1]. Cheaper: the deposit carries **12 SEARCH files** — Mascot `.dat.gz` and Phenyx `.pidres.xml.gz`, 26.8 MB to 1.29 GB, ~4.7 GB total — alongside 44 RAW and 44 MGF, so **peptide-level identifications are already there and no re-search is required** for the first three questions below. Harder: **there is no SDRF and no sample-to-file mapping** — the `v3/projects/PXD009338/files/sdrf` endpoint returns 404, and the run names (`M397-D01-O139-C1655-P4258-1`) encode plate and well positions with no bait annotation [E2]. **Which of the seven baits each file belongs to is not recoverable from the deposit, and that, not compute, is the cost.** Total deposit 33.18 GB; the SEARCH subset alone is ~4.7 GB and fits a bin-and-delete pattern comfortably.

Go deeper
Four questions, in cost order

Q1. What peptides produced the six candidates' spectra? Parse the 12 SEARCH files for peptide-spectrum matches assigned to O14715 (RGPD8), P11310 (ACADM), P15927 (RPA2), Q9UJU2 (LEF1), P00367 (GLUD1) and Q8N8E3 (CEP112). Report sequence, score, charge and rank per PSM. Answers the RGPD8 question outright — if none of the three RGPD8 spectra maps to one of its four discriminating tryptic peptides, that entry's decline becomes a refutation.

Q2. Was CTNNB1 detected below the rank cut? Same parse, accession P35222. Decisive for the LEF1 row either way.

Q3. Were RPA1 (P27694) and RPA3 (P35244) detected below the cut? Same parse. Their presence would convert the RPA2 row from an incoherent single subunit into a complex-level observation.

Q4. Do the sixteen mitochondrial tail members appear across several baits' raw data? This one NEEDS the bait mapping and is therefore the expensive question, not the cheap one it looks like.

Q1-Q3 are bait-agnostic: they ask whether a peptide exists anywhere in the deposit, so they can be answered without knowing which file is which bait. Q4 cannot. That asymmetry is the single most useful thing this entry records.

How to get the bait mapping if Q4 is wanted

Three routes, none free:

  1. Ask the depositors. Grebien's group (then CeMM, now Vetmeduni Vienna) deposited it; a sample-to-file table almost certainly exists locally. This is the cheapest route and it is an email, not an analysis.
  2. Infer from the bait's own peptides. Each run should be dominated by KMT2A N-terminal peptides plus the 3' moiety of its bait — EPS15, AFF1, MLLT3, CREBBP, SH3GL1, MLLT1 or GAS7. Those seven proteins are the label, and they are recovered at 18-257 spectra, so the assignment should be unambiguous from the deposited search results alone. This is the route to take and it is a by-product of Q1's parse rather than extra work.
  3. Cross-reference the run identifiers against GSE110521, the paper's GEO deposition. Unlikely to help — different assay, different naming — and not checked.

Route 2 makes Q4 tractable and should be attempted first; if it works, the bait mapping is recovered for free and the whole set of four questions is answerable from ~4.7 GB of deposited search results with no re-search and no wet work.

Facts and figures
designDownload the 12 SEARCH files of PXD009338 (~4.7 GB, bin-and-delete so peak disk stays near the largest single file at 1.29 GB). Parse peptide-spectrum matches. (a) Extract every PSM assigned to O14715, P11310, P15927, Q9UJU2, P00367, Q8N8E3, P35222 (CTNNB1), P27694 (RPA1) and P35244 (RPA3), with sequence, engine score, charge and whether the peptide is unique to the protein. (b) Recover the file-to-bait mapping by finding, per run, which of EPS15/AFF1/MLLT3/CREBBP/SH3GL1/MLLT1/GAS7 dominates. (c) With the mapping in hand, ask whether the sixteen mitochondrial tail members appear across several baits' runs.
readoutPer candidate: number of distinct peptides, number of PSMs, whether any peptide is unique to the protein against the current human proteome, best engine score. For RGPD8 specifically: whether any PSM maps to one of its four discriminating tryptic peptides (GSGTGVAGASDTTIKPNAENTGPTLEWDNYDLR, EKPLENDTGLQAQDIR, LAAVAQDEEENPSR, IAVAVLEEITR).
controlsSpike control as rule 67 requires: the parse must recover MEN1 (O00255) at high PSM count in every bait run and KMT2A (Q03164) in all of them, since the published table gives MEN1 26-103 spectra per bait and KMT2A 117-366. If the parser cannot find MEN1 and KMT2A, it is broken and no result from it may be reported. Negative control: DDX5 and DDX17, which the authors used as background normalisers and which should be present in every run including controls.
cost tierlow
timelineOne to two days of compute and parsing for questions 1-3; add a day for the bait-mapping recovery if question 4 is wanted. No wet work, no new samples, no approvals.
peak disk gb1.3
total transfer gb4.7
alternative routeRe-search the 44 deposited .mgf.gz peak lists with a current engine (e.g. MSFragger or Comet plus Percolator) against a current human proteome with shared-peptide-aware protein inference. More work, but it replaces a 2012-era SwissProt and a legacy FDR model, and it is the only route that can properly resolve the RGPD paralogue family. Peak disk for the MGF set is larger and was not costed.
shown for fusionThe deposit is MLL-AF4 among seven baits, in a Jurkat T-ALL host. This experiment does not fix the lineage problem and cannot - it makes the existing measurement legible, not more relevant.
epigenetic readoutnone - a proteomic reanalysis.
absence classmeasured-not-analysed
absence class noteEvery question here is answerable from data that has been public since 2018-05-25 and has never been reopened by anyone, including the depositors.
blockers
[
  "Mascot .dat and Phenyx .pidres.xml are legacy formats; Phenyx .pidres in particular has essentially no current tooling and may need a bespoke XML parser.",
  "No sample-to-file mapping in the deposit (SDRF 404). Recoverable in principle from the bait moieties, untested.",
  "12 SEARCH files against 44 RAW files - the deposited search results do not cover every run and the covered subset was not determined."
]
what each answer would change
{
  "RGPD8 PSMs do not map to a discriminating peptide": "interface-kmt2a-rgpd8-declined-ninety-six-percent-shared-tryptic-peptides moves from DECLINED to REFUTED",
  "RGPD8 PSMs do map to discriminating peptides": "that decline must be re-reasoned; the identification would be sound and only the biology unproven",
  "CTNNB1 present below the cut": "revives the beta-catenin bridge for interface-kmt2a-lef1-declined-a-t-lineage-factor-in-a-t-lineage-host",
  "CTNNB1 genuinely absent from the raw data": "closes the beta-catenin bridge",
  "RPA1 and RPA3 present below the cut": "converts the RPA2 row into a complex-level observation and reopens it",
  "the mitochondrial sixteen appear across baits": "confirms a preparation signature and lets ~16 tail members be declined in one pass"
}
How this could bite us back
  • If this reanalysis comes back clean - every candidate resting on several unique well-scored peptides - then six declines in this beat rest on the lineage and compartment arguments alone, which are weaker. The experiment is as likely to embarrass this beat as to confirm it, and that is the reason to run it.
  • A reanalysis that only looks for the six candidates will find whatever is there for the six and say nothing about the other 104. Whoever runs it should parse the whole prey set, not the shortlist, or the next beat repeats the download.
Still open
  1. Does the ProteomeXchange XML manifest or the FTP directory README carry the sample-to-file mapping? Neither was fetched, and either would remove the main blocker at zero cost.
  2. Which subset of the 44 runs do the 12 SEARCH files actually cover? If they cover only some baits, questions 1-3 become bait-dependent after all and the bait-mapping problem returns.
  3. Is re-searching the MGF peak lists with a modern engine actually more expensive than parsing the legacy search results? It would give a better FDR model, a current proteome and shared-peptide-aware inference - the last of which is the only way to settle RGPD8 properly.
Every source behind this entry (2)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    PXD009338 contains 100 files totalling 33.18 GB: 44 RAW (.raw), 44 PEAK (.mgf.gz) and 12 SEARCH files (Mascot .dat.gz and Phenyx .pidres.xml.gz, 26.8 MB to 1.29 GB, ~4.7 GB in total) — so peptide-level identifications are already deposited and a re-search of the raw data is not required to read them.

    What this evidence cannot show

    12 SEARCH files against 44 RAW files means the deposited search results do NOT cover every run one-to-one; several are named '.merged', suggesting they aggregate fractions, and the coverage of runs by search results was not worked out. So 'the identifications are already there' is true for whatever subset those 12 files cover, and that subset was not determined. No file was downloaded; this is a listing only.

    Evidence that argues against this (1)
    • Mascot .dat and Phenyx .pidres.xml are both legacy formats from 2012-2013-era software, and parsing them reliably is not trivial - the pyteomics Mascot .dat reader exists but Phenyx .pidres is essentially unsupported in current tooling. So 'no re-search required' understates the effort: reading these files may be comparable in cost to re-searching the MGF peak lists with a modern engine, which would additionally give a better FDR model and shared-peptide-aware protein inference. A beat taking this on should cost both routes before choosing. ProteomeXchange / PRIDE, EMBL-EBI 2018
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    PXD009338 carries no SDRF and no sample-to-file mapping — the PRIDE v3 sdrf endpoint returns HTTP 404 and the run names encode only plate and well positions (e.g. M397-D01-O139-C1655-P4258-1) with no bait annotation — so which of the seven MLL-fusion baits each file belongs to cannot be read from the deposit.

    What this evidence cannot show

    A 404 on an SDRF endpoint is the expected response for a 2018 deposit - SDRF-Proteomics postdates it - so this is not a defect of the deposit and should not be read as one. Other routes to the mapping were not exhausted: the ProteomeXchange XML manifest, the PX submission summary file, and any README inside the FTP directory were not fetched, and any of them could carry the assignment. The claim is that the mapping is not in the PRIDE REST metadata, not that it is unavailable anywhere.

    Evidence that argues against this (1)
    • The mapping may be recoverable without asking anyone: each run should be dominated by its own bait's 3' moiety, and those seven proteins (EPS15, AFF1, MLLT3, CREBBP, SH3GL1, MLLT1, GAS7) are recovered at 18 to 257 spectra in the published table, so the label is inside the data. If that works, this evidence item overstates the obstacle - the mapping costs one extra pass over search results that are being parsed anyway. It has not been tried. This repository 2026
Connected to