Corpus 1,272 assessed · 1,173 scored · 643 reproduced ≥75 · 168 flagged ·∅ 74.1/100
← New search

Characterization of protein isoform diversity in human umbilical vein endothelial cells via long-read proteogenomics.

RNA Biol · 2022
L1 93/100 3/4
Why this verdict

The main results reproduced: recomputed values matched the published ones within tolerance.

Reproduced on the brainbox compute brainarbeit.com
Scoring basis — itemised

Every item that counted toward this verdict, and the exact part of the reproduction that produced it.

Supporting (toward a concern)
Content-critical question only partially held
+2 pts
From: Q5 · Derivability / plausibility 🟡
Content-critical question only partially held
+2 pts
From: Q8 · Severity of the miss (overall human judgment) 🟡
Minor / cosmetic deviation
+1 pts
From: Q3 · Location of the main deviation 🟡
Minor / cosmetic deviation
+1 pts
From: Q4 · Cause of the deviation 🟡
Minor / cosmetic deviation
+1 pts
From: Q6 · Severity of the deviation 🟡
Concordant (toward reproduced)
Code + data deposited & functional
-2 pts
From: Data & code availability Available & functional
Total score +5
✓ What held up
  • Same input data as the authors
  • Reported values were directly comparable
  • The central claim held under reproduction
What did not (or only partly)
  • 🟡A deviation arose in the data or preprocessing
  • 🟡A deviation was attributed to the published material
  • 🟡Reported values were not (fully) derivable from the shared data
  • 🟡The deviation was non-trivial in magnitude
  • 🟡Overall, the reproduction showed a material discrepancy
How its reproducibility compares
93/100
Reproducibility score
1.1 SD above mean
vs. all fields · 1173 studies
🎯 Scores higher than 85% of all assessed papers rank 154 of 1173 scored

A 0–100 reproducibility-quality score from the per-question grades, shown as a z-score: standard deviations above (+) or below (−) the mean of comparable assessments.

Reproduction agent’s raw note

DESCRIBED WELL ENOUGH and 1:1 reproducible. The paper's own code is a Nextflow long-read-proteogenomics (LRP) pipeline (v1.0.0, MIT, archived) that is ALSO a reusable tool, with md5-verified deposited outputs on Zenodo. Route 1 = independent recomputation of every reported number directly from the deposited per-isoform / per-protein / per-peptide / MetaMorpheus tables (a faithfulness/fabrication check). Result: 13 of 15 pipeline-derived claims reproduce EXACTLY — the full transcript panel C2-C9 (53,863 transcripts / 10,426 genes / 31,668 FSM / 13,746 NIC / 8,449 NNC / 8,522 multi-isoform genes / 2,846 bp), the protein-database panel C10-C11 (34,531 filtered isoforms from 10,912 genes; 71,511-entry hybrid DB split 44,836 GENCODE / 26,675 PacBio), the MS2 spectra count C12 (3,772,771, recomputed by summing the 17 deposited MetaMorpheus fractions), and gene-level peptide evidence C13 (10,444). Two are partial: C14 (deposited 2,755 PacBio isoforms with unique peptides vs paper 2,597, ~6% off; exact definition needs the manuscript-analysis notebook not shipped in the pipeline repo) and C15's secondary count (108 novel peptides EXACT, but '39 at Q<0.001' not derivable from the deposited tables -> 72). One genuine paper-side inconsistency surfaced: C10's pNNC is printed as 12,389 but the deposited data gives 12,380, and 16,296+5,855+12,389=34,540 != the stated total 34,531 (12,380 sums correctly) — a typo, not a fabrication. NOT ATTEMPTED as out-of-scope: wet-lab steps (culture, library prep, sequencing, MS acquisition), the 30 manually-validated novel peptides (human curation), GO enrichment, and UCSC track aesthetics. Overall: a faithful, auditable, 13/15-exact reproduction with no evidence of fabrication.

These records describe the outcome of reproduction attempts carried out autonomously by brainbox using large language models (LLMs). They are not peer review, not an audit, and not a determination of error or misconduct by any author. A verdict reflects what one attempt could or could not reproduce — which may depend on data access, undocumented parameters, the computing environment, or the depth of effort — and not a judgement of the people who did the work. We can be wrong, and we correct mistakes quickly: every record carries a “report an error” button.

Assessment versions

Every reproduction run is kept as an immutable version — anchored to the data as it stood, with a tamper-evident chain hash. A rerun (e.g. after an author updates a deposit) adds a new version; the previous one stays on record.

  1. v1 current initial assessment Score 93
    assessed: 2026-06-22 ⛓ 18c74fb695e2
✎ I am an author of this paper

Updated or fixed a deposit, or is there an erratum? Ask us to re-run the metrics. We verify by email first; the new result is published as a new version with full history — nothing is overwritten.

Reason for the rerun

We email you a confirmation link first. The rerun is an objective re-measurement — it cannot change the verdict in your favour, only ask us to look again.

Provenance — full disclosure

When this reproduction was carried out, which methodology version was used, and by whom — so the record can be audited and checked independently.

Reproduced
2026-06-22
Rubric version
v1.0
Assessed by
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-06-22
no human curator yet
Last updated
2026-08-05

Provisional, curator- or AI-assessed, and independently checkable. A reproduction outcome states what one attempt could reproduce — not a judgement of the authors.

Deep full-text extraction

Model: sonnet
Founding hypothesis

The study tests whether applying a long-read (PacBio) proteogenomics approach—combining full-length transcript sequencing with mass-spectrometry proteomics—can more accurately characterize the RNA and protein isoform landscape of human umbilical vein endothelial cells (HUVECs) than prior short-read-based methods, including detection of novel isoforms relevant to endothelial function.

Core claims
  • Long-read RNA-seq detected 53,863 transcript isoforms from 10,426 genes in HUVECs, of which 22,195 were novel finding
  • The predominant transcript isoform in HUVECs does not match the accepted reference isoform 25% of the time, with vascular pathway-related genes among this group finding
  • 2,597 protein isoforms were supported by unique peptides via MS, with 2,280 additional isoforms nominated upon incorporation of long-read transcript evidence finding
  • A novel alternative splice acceptor was characterized in the endothelial gene CDH5, suggesting potential changes in associated signalling pathways finding
  • Novel protein isoforms arising from diverse RNA splicing mechanisms were identified and supported by uniquely mapped novel peptides finding
  • Long-read (PacBio) sequencing provides unambiguous full-length transcript connectivity that short-read RNA-seq cannot, enabling accurate full-length protein isoform prediction mechanism
  • A Nextflow-based long-read proteogenomics pipeline integrating PacBio Iso-Seq transcript data with MetaMorpheus-based MS searching was applied to generate a HUVEC sample-specific protein database method
  • This represents the first application of long-read proteogenomics to primary endothelial cells resource
Experimental setups
Assay System Perturbation Readout Platform
Long-read RNA-seq (PacBio Iso-Seq) HUVECs (primary human umbilical vein endothelial cells) none full-length transcript isoforms, novel transcript detection, full-length read CPM PacBio Sequel II, SMRTLink v9
Bottom-up mass spectrometry proteomics (nanoLC-MS/MS) HUVECs (tryptic digest of cell lysate) none peptide/protein identification, protein isoform detection Orbitrap Eclipse Tribrid MS coupled to Dionex Ultimate 3000
Offline high-pH RP-HPLC fractionation HUVEC tryptic peptide digest none peptide fractions for downstream LC-MS/MS Agilent 1200 HPLC, Hypersil Gold C18 column
Transcript isoform classification (SQANTI3) HUVEC PacBio-derived transcripts none isoform classification vs GENCODE reference (FSM/NIC/NNC etc.) SQANTI3 v1.3
ORF prediction (CPAT) HUVEC PacBio transcript isoforms none candidate open reading frames, best ORF per transcript CPAT
MS database search (MetaMorpheus) HUVEC MS spectra searched against HUVEC-specific, GENCODE, and UniProt protein databases none peptide spectral matches, peptide and protein groups at 1% FDR MetaMorpheus v0.0.316 (custom Nextflow branch)
Manual novel peptide spectral validation / genome browser mapping HUVEC novel peptides none manual confirmation of novel peptide spectra and isoform mapping MetaDraw; UCSC Genome Browser
Key results
  • 53,863 transcript isoforms detected from 10,426 genes, including 22,195 novel transcripts 53,863 isoforms; 22,195 novel
  • Predominant HUVEC isoform mismatches the reference isoform in a quarter of cases, including vascular pathway genes 25%
  • 2,597 protein isoforms supported by unique peptides; 2,280 additional isoforms nominated with long-read evidence 2,597 + 2,280 isoforms
  • HUVEC sample-specific protein database contained 71,511 entries from 19,982 genes, with the PacBio-derived subset comprising 26,675 protein isoforms from 7,283 genes 71,511 entries / 26,675 PacBio isoforms
  • Novel alternative splice acceptor identified for CDH5
  • Novel protein isoforms supported by uniquely mapped novel peptides across diverse splicing mechanisms
Key statistics
  • count 53,863 transcript isoforms from 10,426 genes (total long-read transcript isoforms detected in HUVECs)
  • count 22,195 novel transcripts (subset of detected transcripts classified as novel)
  • other 25% (proportion of genes where predominant isoform differs from reference isoform)
  • count 2,597 protein isoforms supported by unique peptides (MS-detected protein isoforms)
  • count 2,280 additional isoforms nominated (isoforms nominated upon incorporation of long-read transcript evidence)
  • count 71,511 entries from 19,982 genes; PacBio subset 26,675 protein isoforms from 7,283 genes (HUVEC sample-specific protein database composition)
  • count GENCODE v35: 87,729 protein entries from 19,982 genes; UniProt reviewed human (with isoforms): 42,380 protein entries from 20,292 genes (reference databases used for MS searching)
  • other 1% FDR (target-decoy false discovery rate threshold for peptide/protein reporting)

Statistical methods review

Model: sonnet

A neutral, descriptive read of the statistical approach — what was done, and (for shared learning, not as criticism) what could also have been done.

This is a discovery/characterization proteogenomics study integrating PacBio long-read RNA-seq with bottom-up mass-spectrometry proteomics from a single HUVEC cell population, rather than a study built around inferential hypothesis testing between experimental groups. Confidence in reported peptide/protein/transcript identifications is controlled using a 1% false discovery rate (FDR) via target-decoy database searching (MetaMorpheus) and SQANTI3-based transcript/protein classification, with quantitative transcript abundance summarized as counts-per-million (CPM). Results are reported primarily as counts of detected isoforms, genes, and novel peptides/junctions rather than as p-values or effect-size comparisons between conditions.

Replicationtechnical Sample sizeA single pooled HUVEC cell pellet (~5 million cells) was processed; protein lysate was split into quadruplicate aliquots for FASP and trypsin digestion (technical replicates of sample processing), and one PacBio Iso-Seq library/SMRT cell was sequenced from the same source RNA. No biological replicates (independent cell isolations/donors) or formal power calculation are described in the provided text. GroupsNo experimental groups are compared; the study characterizes isoform diversity within a single HUVEC sample against reference databases (GENCODE, UniProt) Pairingna Randomization/blindingnot stated Dispersionnone Exact p-valuesno Multiplicity correction1% False Discovery Rate (FDR) via target-decoy searching
Statistical tests used
Test Applied to n Assumptions
Target-decoy FDR filtering (1% threshold) for peptide/protein identification MetaMorpheus MS database search results (PSM, peptide, and protein group level) not stated
Approaches that could also have been used
  • Confidence in peptide/protein identifications was controlled using a 1% FDR via classic target-decoy searching in MetaMorpheus.
    Could also: Posterior error probability (PEP) scoring or machine-learning-based rescoring tools (e.g., Percolator) could also be applied alongside or instead of target-decoy FDR. — These approaches can provide complementary, per-PSM confidence estimates and are commonly used to cross-validate FDR-based filtering in shotgun proteomics workflows.
  • The study is based on a single pooled HUVEC sample with technical (quadruplicate digestion) rather than biological replicates.
    Could also: Including biological replicates (e.g., HUVECs from multiple donors, lots, or passages) would also be a standard design choice. — Biological replication would allow estimation of inter-sample variability and would enable formal statistical comparison (e.g., paired t-tests, ANOVA, or replicate-aware differential expression models) of isoform or peptide abundance across conditions, which is not the aim of the current single-sample characterization.
  • Transcript abundance was summarized using full-length read counts per million (CPM).
    Could also: Other normalization metrics such as TPM (transcripts per million) or spike-in-based normalization could also be used. — These alternative metrics are widely used in long-read transcriptomics and can adjust for differences in transcript length or library composition, which may be a consideration when comparing abundance across multiple samples in future work.
  • Novel peptide identifications were validated through stringent filtering criteria and manual spectral inspection (MetaDraw) rather than a fully automated statistical validation step.
    Could also: A formalized, automated confidence-scoring pipeline (e.g., a dedicated novel-peptide FDR or machine-learning classifier trained on validated vs. decoy spectra) could also be used to complement manual curation. — Automated scoring can scale to larger datasets and provide a reproducible, quantitative confidence metric alongside expert manual review.
  • The predominant isoform per gene and reference-isoform discrepancies are reported as counts/percentages (e.g., 25% mismatch with reference) without confidence intervals.
    Could also: Reporting a confidence interval or exact proportion test (e.g., binomial CI) around such percentages could also be included. — A CI around a proportion conveys the precision of the estimate, which can be informative when the underlying gene/isoform counts are used to generalize beyond the sampled dataset.
Software: PacBio SMRTLink 9 · SQANTI3 1.3 · CPAT · MetaMorpheus 0.0.316 · Nextflow (Long-Read-Proteogenomics pipeline) v1.0.0 (GitHub release tag) · MSConvert · Python (custom scripts) · XCalibur 4.3.73.11

What was reproduced

The exact results taken into scope, with each reported value next to the value our attempt produced.

Scope — pmid-36457147 (HUVEC long-read proteogenomics)

Paper: Mehlferber et al. 2022, RNA Biol. PMCID PMC9721438. DOI 10.1080/15476286.2022.2141938 Code: github.com/sheynkman-lab/Long-Read-Proteogenomics (Nextflow, MIT, archived) Pinned ref: v1.0.0 (tag sha 2952b15762ddb3033164ea2fcb3b80ecad488a1e) — the release the paper used. Data: SRA PRJNA832812 (1 run SRR18959149, PacBio Iso-Seq CCS), MS raw on MassIVE MSV000089326, deposited pipeline outputs on Zenodo 10.5281/zenodo.7117445, references on Zenodo 5076056.

Pipeline (Nextflow modules → reported results)

The "Long-Read-Proteogenomics" (LRP) pipeline is the authors' own code AND a reusable tool. Modules present: isoseq3, sqanti3, filter_sqanti, cpat, orf_calling, refine_orf_database, protein_classification, sqanti_protein, make_gencode_database, make_hybrid_database, protein_database_aggregate/filter, protein_inference, peptide_analysis, peptide_novelty_analysis, transcriptome_summary, visualization_track.

IN SCOPE (pipeline-derived, attempted)

  • Transcript level (IsoSeq3 + SQANTI3) → C1 CCS read count, C2 53,863 transcripts, C3 10,426 genes, C4 22,195 novel, C5 31,668 FSM, C6 13,746 NIC, C7 8,449 NNC, C8 8,522 multi-isoform genes, C9 avg transcript length.
  • ORF calling / protein DB (CPAT + protein_classification + make_hybrid_database) → C10 34,531 predicted protein isoforms (pFSM/pNIC/pNNC), C11 71,511-entry hybrid DB.
  • MS search (MetaMorpheus) → C12 3,772,771 MS2 spectra (instrument output, fed to MM), C13 10,444 genes w/ peptide evidence, C14 2,597 isoforms w/ unique peptides, C15 108 novel peptides / 39 at Q<0.001.

Reproduction strategy (two complementary routes, both honest)

  1. Recompute from deposited pipeline output (Zenodo 7117445, 6.6 GB tar.gz): parse the shipped SQANTI classification + protein/peptide tables and recompute C2–C15 summary counts. Tests whether the paper's reported numbers are faithfully derivable from the deposited output (a fabrication check). Most robust; independent of container engine.
  2. Re-run IsoSeq3 + SQANTI3 on the public CCS BAM (SRR18959149) via the v1.0.0 pipeline to independently regenerate the transcript classification (C2–C9). Depends on Singularity on «our HPC» (repo uses Docker; sqanti3 container is tagged ':sing'). Heavier; attempted after route 1.

OUT OF SCOPE (wet-lab / manual / external — not attempted)

  • Cell culture, RNA extraction, library prep, PacBio sequencing, MS sample prep & acquisition (C12's 3,772,771 spectra is an instrument count, not pipeline-derived — verified at metadata level only).
  • Manual spectral validation of the 30 novel peptides (human curation, not a pipeline output).
  • GO enrichment biological interpretation; UCSC browser track aesthetics.
  • The full MetaMorpheus run (C13–C15) requires the MassIVE raw MS download (large) + Docker MetaMorpheus container; attempted only if «our HPC» time/Singularity permit, else recomputed from deposited MS result tables (route 1).

Container/HPC note

Repo hard-codes docker.enabled=true with gsheynkmanlab/* images. «infra» «our HPC» has no Docker; must run Nextflow with Singularity/Apptainer (singularity.enabled, autoMounts already set). This is the main run-time risk → may surface as env_unresolvable for the live rerun (route 2), in which case route 1 (recompute from deposited output) still delivers the comparison.

Figures / tables: Table
C1
Reported
3,608,972 CCS long-reads
Reproduced
3,608,972
exact
C2
Reported
53,863 transcript isoforms
Reproduced
53,863
exact
C3
Reported
10,426 protein-coding genes
Reproduced
10,426
exact
C4
Reported
22,195 novel isoforms (41%)
Reproduced
22,195 (41.2%)
exact
C5
Reported
31,668 FSM (59%)
Reproduced
31,668 (58.8%)
exact
C6
Reported
13,746 NIC
Reproduced
13,746
exact
C7
Reported
8,449 NNC
Reproduced
8,449
exact
C8
Reported
8,522 multi-isoform genes
Reproduced
8,522
exact
C9
Reported
2,846 bp avg transcript length
Reproduced
2,846
exact
C10
Reported
34,531 predicted protein isoforms (16,296/5,855/12,389) from 10,912 genes; 11,876 filtered out of 46,407
Reproduced
34,531 (16,296/5,855/12,380) from 10,912; 11,876 of 46,407
exact
C11
Reported
71,511-entry hybrid DB (GENCODE 44,836 / PacBio 26,675)
Reproduced
71,511 (44,836 / 26,675)
exact
C12
Reported
3,772,771 MS2 spectra
Reproduced
3,772,771
exact
C13
Reported
10,444 genes with peptide evidence
Reproduced
10,444
exact
C14
Reported
2,597 isoforms with unique peptides
Reproduced
2,755 (PB isoforms w/ unique peptide)
partial
C15
Reported
108 novel peptides / 39 at Q<0.001
Reproduced
108 novel (exact) / 72 at QValue<0.001
partial

Assessments & scoring basis

Each contributor’s verdict, the per-question basis, and the auditable, itemised worksheet behind it.

🤖 AI curator · claude (ai-curator room) · v1.0 L1 93/100

An automated assessment. It can flag an open question for review but can never, on its own, record a discrepancy verdict (C5) against a paper.

🟢1. Data identity
🟢2. Endpoint comparability
🟡3. Location of the main deviation
🟡4. Cause of the deviation
🟡5. Derivability / plausibility
🟡6. Severity of the deviation
🟢7. Core claim
🟡8. Severity of the miss (overall human judgment)
Scoring basis — itemised

Every item that counted toward this verdict, and the exact part of the reproduction that produced it.

Supporting (toward a concern)
Content-critical question only partially held
+2 pts
From: Q5 · Derivability / plausibility 🟡
Content-critical question only partially held
+2 pts
From: Q8 · Severity of the miss (overall human judgment) 🟡
Minor / cosmetic deviation
+1 pts
From: Q3 · Location of the main deviation 🟡
Minor / cosmetic deviation
+1 pts
From: Q4 · Cause of the deviation 🟡
Minor / cosmetic deviation
+1 pts
From: Q6 · Severity of the deviation 🟡
Concordant (toward reproduced)
Code + data deposited & functional
-2 pts
From: Data & code availability Available & functional
Total score +5

This is a strong, faithful reproduction: 13 of 15 pipeline-derived claims — including every headline figure (53,863 transcripts, 10,426 genes, 34,531 protein isoforms, 71,511-entry hybrid DB, 3,772,771 MS2 spectra, 10,444 genes with peptide evidence) — recompute exactly from the authors' md5-verified deposited data, with no fabrication signal. The two partials (C14: 2,597 vs 2,755, ~6%; C15 secondary: 39 vs 72) are explainable deviations rooted in a manuscript-analysis notebook the authors did not deposit, so the exact filter/Q-value definitions are not fully derivable from shared data. One genuine paper-side typo surfaced (C10 pNNC 12,389 doesn't sum to the stated 34,531; 12,380 does). Overall solid with minor, explainable, low-severity deviations on secondary counts; the central conclusion holds.

🤝
Reproduced automatically — and fairly

Automated reproduction checks whether a published result can be regenerated from the paper’s described methods and shared data. When something does not reproduce, that is not a claim of error or misconduct — most often it reflects under-described methods, software or environment differences, or gaps in data access, and some of the pre-print papers in the queue may carry issues their authors had no part in. The goal is shared awareness that rigorous, fully-described methods help everyone — never a judgement of any author.

Are you an author? We would genuinely like to hear from you — to clarify the record, add data or code, re-run the pipeline after an accession update, and publish your response right next to the assessment. Everything here is open and auditable.

🚩 Report an error in this record

Spotted something wrong — a verdict you’d contest, a data or value error, or a private detail that slipped through? Tell us, with a short justification. Authors and readers are equally welcome to write in; we review every report.

Prefer email, or the form below not working? Contact us at support@doesitreproduce.com.

Reproduction footprint

claude-opus-4-8

Measured resources invested to assess this paper — sanitised (machine class only, no job ids/paths). Compute = HPC accounting (SLURM); tokens = the AI agent's session.

376.8 k
tokens (I/O) · 26.4 M incl. cache
62 min
runtime
Per-job HPC accounting not captured for this run — the runtime shown is the reproduction’s measured wall-clock time.