Gene Expression Analysis Platform (GEAP): A highly customizable, fast, versatile and ready-to-use microarray analysis platform.
The main results reproduced: recomputed values matched the published ones within tolerance.
Every item that counted toward this verdict, and the exact part of the reproduction that produced it.
- ✓Same input data as the authors
- ✓No relevant deviation in data/preprocessing
- ✓Reported values are derivable from the shared data
- ✓Any deviation was negligible
- 🟡Reported values were only indirectly comparable
- 🟡A deviation was attributed to the published material
- 🟡The central claim did not (fully) hold under reproduction
- 🟡Overall, the reproduction showed a material discrepancy
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
Tool/software paper (GEAP microarray platform). The GEAP GUI is a C#/.NET-Mono binary NOT in the GitHub repo; the public repo rgeap is the R backend wrapping standard Bioconductor packages. The paper's validation (Figure 3, 15 datasets) is entirely qualitative and prints NO numeric pipeline result, so the only clean 1:1 numeric claim for the registry-pinned dataset GSE9466 is its dual-channel classification. We reproduced (C1, EXACT) that GSE9466 is genuinely two-color, confirmed both from GEO metadata (channel_count=2, GPL6068 Agilent G4140A) and by reading the raw Agilent two-color files into an RGList with separate R/G channels on «our HPC». We then ran GEAP's documented default limma two-color DE pipeline (parameters extracted verbatim from rgeap source: normexp/saddle background correction, quantile between-array normalization, dye-swap-corrected M-values, lmFit/eBayes, BH, lfc=log2(1.2)) on all 84 GSE9466 arrays (C2, PARTIAL): the pipeline runs end-to-end and produces sensible, internally-consistent per-metal DE counts (4540-6647 DE probes/contrast; As_EC50 strongest) and scatter/volcano plots that qualitatively match Figure 3A. Because the paper supplies no reference number, C2's honest ceiling is partial; we did not manufacture a numeric match. Out of scope (dropped, recorded): the GUI executable and its performance benchmarks (env_unresolvable: binary not shipped / non_pipeline), and the other 8 GSE (80/20, same pipeline class). No fabrication signal. Overall: PARTIAL reproduction (one claim exact, the in-scope pipeline reproduced as an internally-consistent artifact).
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.
-
v1 current initial assessment Score 83assessed: 2026-06-14 ⛓ 52b166ec59bf
✎ 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.
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
- 2026-06-14
- Rubric version
- v1.0
- Assessed by
-
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-06-15no 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: sonnetMicroarray data analysis remains inaccessible to non-programming biomedical researchers because existing manufacturer software and R-based approaches are either too limited in file/format and statistical support or require programming skills; the authors propose that combining a GUI front-end with an R back-end can deliver a flexible, fast, and ready-to-use platform for both advanced and non-programming users.
- ★ GEAP is a GUI-based microarray analysis platform combining a C# front-end with an R (RTerm) back-end via the rgeap package, enabling analysis independent of manufacturer/platform. resource
- ★ GEAP automatically identifies microarray file formats, retrieves contents, detects file corruption, and resolves package dependencies. finding
- ★ GEAP covers 32 statistical options and supports quality assessment, differential expression (single and dual-channel), and gene ontology analysis. resource
- ★ Certain R functions were reimplemented in C# (parallel file downloads, lazy-loaded table processing with hash-set deduplication, plotting/filtering) to improve performance over pure R. method
- ★ GEAP stores library and project data in a local SQLite database, which provides faster binary data performance than reading from the file system. finding
- GEAP supports GO analysis via topGO across all Bioconductor organism database packages (currently ~20 model organisms) plus custom GAF annotations. resource
- GEAP generates R source code tracking user actions/parameters during an analysis session for reproducibility. method
- GEAP runs on Windows 7-10 (.NET Framework 4.6.1) and Ubuntu 18.04+ (Mono 6.8.0), released under MIT license (rgeap under GNU LGPL). resource
| Assay | System | Perturbation | Readout | Platform |
|---|---|---|---|---|
| software performance benchmark (data storage read/compression) | SQLite database file vs. File System (binary microarray data) | storage backend (SQLite vs file system) | performance/compression boost (%) | SQLite |
| microarray data processing/validation (general software testing) | multiple microarray datasets of different sources, backgrounds, platforms, and sample sizes | none (software validation across datasets) | successful data processing/analysis capability | GEAP/rgeap, R 4.0.3, Bioconductor packages (affy, oligo, makecdfenv, beadarray, illuminaio, limma, arrayQualityMetrics, topGO) |
- ▲ SQLite-based storage provides additional performance boost for binary data compared to File System storage ~35%
- ▲ GEAP reimplements time-critical R operations (multi-threaded downloads, lazy table loading, hash-set based duplicate detection) in C# for speed gains
- ▼ Data table processing complexity reduced via hash sets O(1) instead of O(n)
- other 32 statistical options (Number of statistical options covered by GEAP)
- other ~35% additional performance boost (SQLite vs File System binary data read performance)
- other 20 model organisms (Number of Bioconductor organism database packages supported for GO analysis)
- other O(1) vs O(n) (Complexity reduction for data table duplicate/error detection using hash sets)
- other minimum 2 GB RAM, 2.4 GHz processor, 1 GB disk; recommended 8 GB RAM, dual-core 2.4 GHz, 10+ GB disk (GEAP system requirements)
Statistical methods review
Model: sonnetA 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 software description paper presenting GEAP, a microarray analysis platform; it does not report original biological statistical analyses. Statistical capabilities are delegated to established R/Bioconductor packages, principally limma (linear model fitting with empirical Bayes moderation) for differential expression and topGO for gene ontology enrichment. Performance advantages over R are described qualitatively and with one informal benchmark figure (SQLite approximately 35% faster for binary data), without formal hypothesis testing or reported dispersion.
| Test | Applied to | n | Assumptions |
|---|---|---|---|
| Linear model fitting with empirical Bayes moderated statistics (limma) | Differential expression analysis embedded within GEAP software | — | not stated |
| Statistical evaluation of gene ontology enrichment (topGO package) | Gene ontology analysis embedded within GEAP software | — | not stated |
-
Performance gains over R are supported by a single informal benchmark (SQLite ~35% faster for binary data) and qualitative statements↳ Could also: Formal benchmarking with repeated timed runs across multiple representative datasets, reporting mean ± SD or median with IQR, and a paired Wilcoxon signed-rank or t-test on runtime measurements — Quantified benchmarks with dispersion and an inferential test would allow readers to evaluate both the magnitude and consistency of the reported speedup across dataset sizes and hardware configurations
-
GEAP's analytical correctness is validated qualitatively across an unspecified set of datasets 'from different sources, backgrounds, platforms, and sample size'↳ Could also: A structured comparison using established public reference datasets (e.g., MAQC/SEQC benchmarks) with quantitative concordance metrics such as DEG list overlap (Jaccard index) or ROC analysis against known ground-truth — Reference datasets with known expression changes enable objective, reproducible evaluation of whether a software pipeline produces results consistent with established analyses, complementing qualitative feature descriptions
-
Differential expression is implemented via the limma linear model framework, the only DE method described↳ Could also: For microarray data, RMA-normalized data could also be analyzed with other frameworks such as ROTS (Reproducibility-Optimized Test Statistic) or a standard moderated t-test with Benjamini-Hochberg correction; for count-based platforms, DESeq2 or edgeR would be alternatives — Presenting concordance between limma and at least one alternative DE approach on a representative dataset would help users understand how method choice affects the reported DEG lists
-
Gene ontology enrichment is performed via topGO, with the specific internal test statistic (e.g., Fisher's exact test, Kolmogorov-Smirnov) not stated in the provided text↳ Could also: clusterProfiler, goseq (which accounts for gene-length bias), or GSEA (rank-based permutation testing) are also widely used for ontology and pathway enrichment — Explicitly naming the topGO test statistic and graph-pruning algorithm used (e.g., 'elim' or 'weight'), and noting how it differs from rank-based alternatives, would aid reproducibility and help users select the method most appropriate for their data
Citation network
Where this publication sits in the reproducibility-weighted citation graph — what it is built on, and what is built on it. Citation data from OpenAlex.
No assessed neighbours yet — the network grows as more papers are assessed.
What was reproduced
The exact results taken into scope, with each reported value next to the value our attempt produced.
scope.md — pmid-34927664
Paper: Nunes IJG, Feltes BC et al. (2021) Gene Expression Analysis Platform (GEAP): A highly customizable, fast, versatile and ready-to-use microarray analysis platform. Genet Mol Biol 44(4):e20210077. PMID 34927664 / PMC8754388 / DOI 10.1590/1678-4685-gmb-2021-0077. Repo: https://github.com/nunesijg/rgeap (HEAD 38d0eed, 2022-03-13; v0.4.0). Data: GEO GSE9466 (registry-pinned), one of the datasets used to validate GEAP.
Rule 2 — code provenance & paper type
This is a software/tool paper. GEAP itself is a C#/.NET (Windows) / Mono
(Linux) GUI executable — not in the GitHub repo. The public repo rgeap is
the R backend library that GEAP's GUI calls at runtime via an embedded RTerm
("rgeap is not really an R package aimed to perform standalone differential
expression analyses … developed as a technical solution", Methods). The authors
are the same (Nunes IJG, Feltes BC) → authors' own code. Per Operator P16, even a
third-party tool would be equally valid; here the repo wraps the standard
Bioconductor packages it documents (affy, oligo, limma, beadarray, …), so the
faithful reproduction is to run that same documented pipeline with GEAP's exact
default parameters (which we extracted from the rgeap source) on the paper's own
data GSE9466.
What the paper actually reports (and what it does NOT)
The validation (Results §"Software validation", Figure 3) analysed 15 microarray datasets (3 GSE each from Affymetrix, Agilent, Illumina, GenePix, NimbleGen). Crucially, the paper prints NO numeric pipeline-derived result — Figure 3 is qualitative (scatter/volcano/bar/stack plots; "higher dispersion and flattened volcano plots in groups with fewer samples"). The only specific, checkable, GSE9466-named factual claim is the dual-channel classification.
Results: "Three of the chosen GSE to validate the software were dual-channel (GSE9466, GSE10402, and GSE 17978), which GEAP successfully identified."
GEAP's documented default DE pipeline (extracted from rgeap source)
R/f-limma.R::treat.limma()→ background correctionmethod="auto"(=normexp),normexp.method="saddle",offset=0; thennormalizeBetweenArrays(method="quantile").R/f-diffexpr.R::diffexpr.compare()→lmFit+contrasts.fit+eBayes(defaultebayesMethod="ebayes");topTable(adjust.method="BH"); volcano useslogFCvs-log10(adj.P.Val); default fold-change thresholdlfc = log2(1.2).
GSE9466 — verified dataset facts (GEO, control-plane)
- Title: Transcriptome in yeast by exposure of 7 transition metals; organism
S. cerevisiae (BY4743). Platform GPL6068 = Agilent-011447 Yeast Oligo
Microarray (G4140A), two-color (
Sample_channel_count = "2"). - 84 arrays (GSM240140–GSM240223) = 7 metals (Ag, As, Cd, Cr, Cu, Hg, Zn) × 2 doses (EC10, EC50) × 6 replicates. Each array: ch1 = Control vs ch2 = treated, with dye-swap (ch1 dye alternates CY5/CY3). 6256 probes (series matrix).
- Raw data public:
GSE9466_RAW.tar(Agilent feature-extraction TXT + TIFF).
IN SCOPE (public data, open-source toolchain, tractable on «our HPC»)
- C1 (clean 1:1, EXACT-able): GSE9466 is a dual-channel/two-color microarray.
Verify from GEO platform GPL6068 (Agilent two-color) and series-matrix
Sample_channel_count="2". This is the one GSE9466-specific number the paper claims. - C2 (pipeline reproduction, PARTIAL): Run GEAP's documented default limma two-color pipeline (normexp/saddle bg → quantile between-array norm → dye-swap M-values → lmFit/eBayes/BH, lfc=log2(1.2)) on GSE9466 raw Agilent data. Produce per-contrast DE counts (over/under-expressed), an M-value summary, and a scatter+volcano plot to compare qualitatively with Figure 3 (small-sample two-color → high dispersion / flattened volcano). No printed reference number exists → grade = partial (pipeline runs end-to-end on the paper's own data with the paper's
Assessments & scoring basis
Each contributor’s verdict, the per-question basis, and the auditable, itemised worksheet behind it.
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.
Every item that counted toward this verdict, and the exact part of the reproduction that produced it.
This is a software/tool paper (GEAP) whose validation is entirely qualitative — Figure 3 prints no numeric pipeline result. The one falsifiable GSE9466-specific claim, that the dataset is dual-channel, is fully derivable and reproduced exactly (84 arrays channel_count=2; RGList R+G 10807×84), and the documented limma two-color pipeline runs end-to-end with internally-consistent counts (14 contrasts, 4540–6647 DE probes), so there is no deviation and no fabrication signal. The honest limits are on the paper's side / our scope, not a defect: C2 cannot exceed partial because no reference number exists, and the GUI plus performance benchmarks ('almost twice the performance') were dropped as out-of-scope/unbuildable, leaving the tool's headline speed claim unverified — hence q7/q8 yellow.
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-4-8Measured resources invested to assess this paper — sanitised (machine class only, no job ids/paths). Compute = HPC accounting (SLURM); tokens = the AI agent's session.