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

TRAPID 2.0: a web application for taxonomic and functional analysis of de novo transcriptomes.

Nucleic Acids Res · 2021
L1 60/100 3/4
Why this verdict

The main results reproduced, with only marginal, non-material deviations.

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: Q7 · Core claim 🟡
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 🟡
Minor / cosmetic deviation
+1 pts
From: Q2 · Endpoint comparability 🟡
Concordant (toward reproduced)
Code + data deposited & functional
-2 pts
From: Data & code availability Available & functional
Total score +8
✓ What held up
  • Nothing in this column.
What did not (or only partly)
  • 🟡Could not use the authors’ exact input data
  • 🟡Reported values were only indirectly comparable
  • 🟡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
  • 🟡The central claim did not (fully) hold under reproduction
  • 🟡Overall, the reproduction showed a material discrepancy
How its reproducibility compares
60/100
Reproducibility score
0.8 SD below mean
vs. all fields · 1173 studies
🎯 Scores higher than 21% of all assessed papers rank 918 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

TRAPID 2.0 is a web-application tool paper; its headline annotation numbers (C3-C7) and performance metric (C8) are produced only by the live web server / authors' cluster and are out of scope as standalone compute by design. The reproducible upstream pipeline ran end-to-end on «our HPC» (Trimmomatic 0.36 + MIRA 4.0.2, «job», mira_exit=0): C1 read count reproduces within ~3% (raw deposit 559,011, post-trim 542,211 vs reported 526,527 - same data, underspecified trim), and C2a MIRA assembly reproduces closely (56,143 contigs / N50 393 bp vs 53,569 / 391 bp). C2b (contigs+singletons = 143,308) MISMATCHES: MIRA's default est/accurate run reported 0 singletons and discarded 292,309 unassembled reads as debris, whereas the paper retained 89,739 singletons; the singleton-retention parameter is not documented, so 143,308 is not reproducible from the methods. No fabrication concern - every in-scope value is derivable from the shipped data, deltas trace to underspecified parameters and assembler non-determinism. NOT attempted: the full TRAPID self-host (CakePHP/MySQL/SGE + DIAMOND/Kaiju/Infernal/eggNOG-mapper + pico-PLAZA 3.0 / NCBI-nr, hundreds of GB), MMETSP 26-sample validation, and ciliate genetic-code analysis (all web-server driven).

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 42
    assessed: 2026-06-18 ⛓ 19de7ec472d2
✎ 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
2026-06-25
Rubric version
v1.0
Assessed by
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-06-19
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 authors aimed to develop TRAPID 2.0, an improved web application, to overcome the computational challenges of processing fragmented and contaminant-containing de novo assembled (meta)transcriptomes, providing fast structural, functional, and taxonomic annotation together with downstream comparative analyses.

Core claims
  • TRAPID 2.0 is a web application performing global characterization of de novo transcriptomes via structural, functional, and taxonomic annotation in an initial processing phase, followed by an exploratory phase of downstream analyses. resource
  • TRAPID 2.0 supports analysis of assembled (meta)transcriptomes with mixed prokaryotic and/or eukaryotic composition due to improved processing speed, updated reference databases, and integrated protein-based taxonomic classification. method
  • Core gene family completeness assessment allows estimation of gene space completeness along a user-defined evolutionary lineage. method
  • Using the ciliate nuclear genetic code instead of the standard genetic code for ORF prediction improves the 'best hit recovery ratio' for ciliate transcriptomes with stop codon reassignment. finding
  • Splitting the Kaiju taxonomic classification index for memory efficiency results in performance nearly identical to a regular (non-split) Kaiju MEM run. finding
  • Functional annotation transfer from gene families requires the annotation label to be present in at least 50% of GF members (majority vote), validated by benchmarking on five model species. method
  • Application of TRAPID 2.0 to diatom-dominated Antarctic metatranscriptomes reveals functional variation between phytoplankton communities from different sampling sites. finding
  • TRAPID 2.0 offers unique features relative to comparable transcriptome annotation tools. finding
Experimental setups
Assay System Perturbation Readout Platform
Taxonomic classification (protein-based k-mer classification) 42 MMETSP re-assembled unicellular eukaryote transcriptomes none taxonomic assignment of transcripts Kaiju
Similarity search / gene family assignment / functional annotation Query transcriptomes vs. reference proteomes (PLAZA 4.5, pico-PLAZA 3.0, PLAZA diatoms 1.0, eggNOG 4.5) none top protein hit, gene family assignment, GO/KO terms, protein domains DIAMOND; eggNOG-mapper v1
Non-coding RNA identification Query transcript sequences none RNA family/clan assignment and GO term annotation Infernal (cmsearch) with Rfam 14.1 models
Ribosomal RNA taxonomic classification Putative SSU/LSU rRNA transcripts none taxonomic classification of rRNA sequences SINA v1.5.0 with SILVA 132
ORF prediction / best hit recovery ratio comparison 16 ciliate MMETSP samples (825,773 sequences; 257,454 with similarity hits) genetic code choice (standard translation table 1 vs ciliate nuclear translation table 6) ratio of predicted ORF length to best similarity search hit length TRAPID 2.0 pipeline / eggNOG 4.5 reference
Multiple sequence alignment Two MMETSP0018 (Uronema sp. Bbcil) transcripts assigned to eggNOG orthologous group 0IF5I plus 4 Alveolata reference sequences genetic code used for translation (standard vs ciliate) MSA visual comparison of translated sequences MAFFT; MSA R package (TEXshade)
Metatranscriptome assembly and expression quantification Diatom-rich community metatranscriptomes, 3 western Antarctic Peninsula sampling sites (SRX727358, SRX727361, SRX727362) sampling site (environmental) assembled contigs, TPM expression values, expressed transcript subsets Trimmomatic, FastQC, MIRA v4.0.2, BWA mem v0.7.17, FeatureCounts (subread v1.6.2)
Subset functional enrichment analysis Diatom (Bacillariophyta)-assigned transcript subsets expressed per sampling site sampling site comparison enriched functional annotation labels (hypergeometric test, BH-corrected q-value) TRAPID 2.0 with pico-PLAZA 3.0 reference
Key results
  • Splitting Kaiju's reference index for memory efficiency yields nearly identical classification performance to a non-split run.
  • Using the ciliate nuclear genetic code improved the best hit recovery ratio compared to the standard genetic code for ciliate transcriptomes.
  • Global metatranscriptome assembly from 526,527 combined reads produced 53,569 contigs (N50 = 391 bp); 208,592 reads unused by MIRA, of which 89,739 sequences >200 bp were retained as bona fide transcripts. N50=391 bp
  • Final combined metatranscriptome dataset used for analysis contained 143,308 sequences with N50 of 320 bp. N50=320 bp
  • Functional enrichment analysis of diatom-specific transcript subsets revealed functional variation between the three phytoplankton communities across sampling sites.
Key statistics
  • count 825,773 sequences (total transcripts across 16 ciliate MMETSP samples processed)
  • count 257,454 transcripts (transcripts with protein similarity search hits used for best hit recovery ratio computation)
  • other ≥90% of species in clade (threshold for a gene family to be considered 'core' for completeness analysis)
  • other ≥50% of GF members (majority vote) (threshold for transferring functional annotation labels from gene families)
  • count 526,527 reads (total retained reads combined from 3 Antarctic sampling site libraries for assembly)
  • count 53,569 contigs, N50 391 bp (MIRA global metatranscriptome assembly output)
  • count 143,308 sequences, N50 320 bp (final metatranscriptome dataset (contigs + retained singletons) used for downstream analysis)
  • other TPM ≥ 2 (threshold to consider a transcript expressed in a given sampling site library)

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.

TRAPID 2.0 is a bioinformatics software paper presenting a web application for de novo transcriptome annotation and exploration. Statistical methods appear in three contexts: (i) hypergeometric tests with Benjamini–Hochberg FDR correction for subset functional enrichment analysis; (ii) bootstrap-based branch support estimation in phylogenetic tree inference; and (iii) a descriptive ratio metric (best hit recovery ratio: predicted ORF length / best similarity-search hit length) used to compare the impact of standard versus ciliate nuclear genetic codes on ORF prediction across 257,454 sequences. Assembly quality and expression levels are reported as N50 lengths and TPM values, respectively, with no inferential summary statistics accompanied by dispersion measures in the main text.

Replicationmixed Sample size42 MMETSP samples retrieved for overview; 16 ciliate MMETSP samples used for genetic-code evaluation; 3 Antarctic sampling sites for diatom metatranscriptome demonstration; 5 model species for functional annotation benchmarks (details deferred to Supplementary Note S2) GroupsStandard vs. ciliate nuclear genetic code for ORF prediction; diatom-specific transcript subsets from three western Antarctic Peninsula sampling sites; split Kaiju index vs. regular Kaiju MEM run (Supplementary Note S1) Pairingmixed Randomization/blindingnot stated Dispersionnone Exact p-valuesno Effect sizesno Confidence intervalsno Multiplicity correctionBenjamini–Hochberg (BH) FDR correction
Statistical tests used
Test Applied to n Assumptions
Hypergeometric distribution test with Benjamini–Hochberg q-value correction Subset functional enrichment analysis: functional annotation labels (e.g. GO terms) in a user-defined transcript subset tested against the whole-experiment background not stated in main text (whole experiment used as background; subset sizes not given) not stated
Ultrafast bootstrap approximation (UFBoot), 1000 replicates Branch support estimation in IQ-TREE maximum-likelihood phylogenetic trees 1000 replicates not stated
Standard non-parametric bootstrap, 100 replicates Branch support estimation in FastTree2, PhyML, and RAxML phylogenetic trees 100 replicates not stated
Best hit recovery ratio — descriptive metric, no formal significance test stated Evaluation of standard (NCBI translation table 1) vs. ciliate nuclear (table 6) genetic code effect on ORF prediction quality, across 16 ciliate MMETSP samples 257,454 transcript sequences with protein similarity search hits (from 825,773 total sequences across 16 samples) na
TPM threshold (TPM ≥ 2) for binary expression calling — not an inferential test Definition of expressed transcript subsets per diatom sampling site 143,308 input sequences; 526,527 reads mapped across 3 libraries na
Approaches that could also have been used
  • The best hit recovery ratio was compared descriptively between standard and ciliate nuclear genetic codes across 257,454 sequences, with no formal inferential test reported
    Could also: A paired non-parametric test (e.g., Wilcoxon signed-rank test on per-transcript ratio differences) or a permutation test could also quantify this comparison — A formal test would yield a p-value and a standardized effect size, allowing readers to gauge whether the improvement attributable to the alternative genetic code exceeds chance variation across the large sequence set
  • Subset functional enrichment used the hypergeometric distribution with a user-supplied background (all transcripts in the experiment)
    Could also: Fisher's exact test (one-tailed; mathematically equivalent to the hypergeometric test) or rank-based gene set enrichment analysis (GSEA) using a ranked TPM list could also be applied — Fisher's exact test is widely familiar and gives identical results; GSEA avoids the need to set a binary expression threshold by using the full ranked list, which can improve sensitivity for moderate but consistent enrichment signals
  • A fixed TPM threshold of ≥ 2 was used to dichotomize transcripts as expressed or unexpressed per sampling site
    Could also: Count-based differential expression methods (e.g., DESeq2 or edgeR) could also model expression differences across sampling sites with statistical testing — Count-based methods explicitly model overdispersion in RNA-Seq count data and provide FDR-controlled p-values for differential expression between sites, whereas a fixed TPM threshold treats all above-threshold transcripts equivalently regardless of count-level uncertainty
  • Branch support in IQ-TREE was estimated using UFBoot (1000 replicates); other programs used standard non-parametric bootstrap (100 replicates)
    Could also: Approximate likelihood-ratio tests (aLRT) or Bayesian posterior probabilities (e.g., via MrBayes) could also estimate branch support — aLRT is computationally cheaper than bootstrap resampling; Bayesian posterior probabilities have a direct probabilistic interpretation and can be preferable when bootstrap values are known to be overconfident in certain tree topologies
  • Core gene family status was defined by a fixed 90% species-representation threshold within a clade, with a weighted completeness score computed as a ratio
    Could also: A sensitivity analysis across a range of thresholds (e.g., 80%, 95%) or the widely used BUSCO framework for universal single-copy ortholog completeness assessment could also be applied — Threshold sensitivity analyses would reveal how completeness scores change with stringency; BUSCO provides a community-standard benchmark that facilitates direct cross-study comparisons
  • Functional annotation labels were transferred to a transcript from its gene family using a 50% majority-vote threshold across GF members
    Could also: Weighted voting (weights proportional to sequence similarity or phylogenetic distance) or probabilistic annotation transfer (e.g., PANNZER2-style scoring) could also be applied — Weighting by sequence similarity would give higher confidence to annotation labels supported by closely related family members, which may improve precision in large and functionally heterogeneous gene families
Software: Kaiju · DIAMOND · eggNOG-mapper 1 · Infernal · SINA 1.5.0 · MAFFT · MUSCLE · FastTree2 · IQ-TREE with ModelFinder · PhyML · RAxML · R / MSA package · Trimmomatic 0.36 · FastQC 0.11.2 · MIRA 4.0.2 · BWA mem 0.7.17 · FeatureCounts (subread package) 1.6.2 · Krona · CakePHP / MySQL

What was reproduced

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

C1
Reported
526,527 combined input reads
Reproduced
raw 559,011; post-Trimmomatic-0.36 542,211
within tolerance
C2a
Reported
53,569 contigs, N50 391 bp (MIRA initial)
Reproduced
56,143 contigs, N50 393 bp
within tolerance
C2b
Reported
143,308 sequences, N50 320 bp (contigs+singletons)
Reproduced
56,143 sequences, N50 393 bp; MIRA emitted 0 singletons (292,309 reads dropped as debris)
did not match
C3-C7
Reported
TRAPID web annotation stats (gene families/RNA/taxonomy/GO/InterPro)
Reproduced
OUT_OF_SCOPE: web-server-only (pico-PLAZA 3.0 / NCBI-nr self-host), non-reproducible as standalone CLI by design
m.public.grade.out-of-scope
C8
Reported
119,699 seqs in <5h
Reproduced
OUT_OF_SCOPE: server hardware performance metric
m.public.grade.out-of-scope

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 60/100

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: Q7 · Core claim 🟡
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 🟡
Minor / cosmetic deviation
+1 pts
From: Q2 · Endpoint comparability 🟡
Concordant (toward reproduced)
Code + data deposited & functional
-2 pts
From: Data & code availability Available & functional
Total score +8

TRAPID 2.0 is a web-application descriptor paper; its reproducibility is limited by design, not by an authors' defect or fabrication. The one number we could check (C1: 526,527 reported vs 559,011 raw, +6.2%) deviates moderately with a benign explanation (raw vs Trimmomatic-trimmed). The headline case-study figures (C2–C7: assembly size, gene-family/taxonomy/GO percentages) are not derivable from shared data because the MIRA assembly was never deposited and annotation is web-server-only and non-deterministic. Net: solid-but-incomplete partial reproduction, deviations explainable and on the data-availability/our-methodology side — no significance flip or fabrication concern.

🤝
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.

122.2 k
tokens (I/O) · 7.5 M incl. cache
24 min
runtime
Per-job HPC accounting not captured for this run — the runtime shown is the reproduction’s measured wall-clock time.