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

Specific signature biomarkers highlight the potential mechanisms of circulating neutrophils in aneurysmal subarachnoid hemorrhage.

Front Pharmacol · 2022
L1 80/100 3/4
Why this verdict

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

Reproduced on the brainbox compute brainarbeit.com
✓ What held up
  • Same input data as the authors
  • Reported values are derivable from the shared data
  • Any deviation was negligible
  • The central claim held under reproduction
What did not (or only partly)
  • 🟡Reported values were only indirectly comparable
  • 🟡A deviation arose in the data or preprocessing
  • 🟡A deviation was attributed to the published material
  • 🟡Overall, the reproduction showed a material discrepancy
How its reproducibility compares
80/100
Reproducibility score
0.3 SD above mean
vs. all fields · 1173 studies
🎯 Scores higher than 56% of all assessed papers rank 484 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 to reproduce the core 1:1. The paper is a secondary GEO analysis; its 'code' link (GseaVis) is only a GSEA plotter, so per P16 we re-ran the standard pipeline (limma/clusterProfiler/glm+pROC) on the paper's own data GSE36791. PRIMARY ANCHORS REPRODUCE: sample groups exact (43 aSAH/18 control); DEGs 571 vs reported 602 (within-tol, -5%, up/down ratio preserved, paper's exact thresholds); the published 6-gene signature (CST7/HSP90AB1/PADI4/PLBD1/RAB32/SLAMF6) is all present and gives combined diagnostic AUC 0.938 vs reported 0.906 (within-tol), each gene 0.85-0.91. PARTIAL: GO themes are broadly immune/leukocyte-consistent but not the paper's neutrophil-specific top terms, because those come from a WGCNA blue module (out of scope, params unspecified); KEGG could not run (rest.kegg.jp blocked on «infra» compute nodes). NOT ATTEMPTED / not comparable: WGCNA module reconstruction, de-novo LASSO+SVM-RFE gene selection (re-applied published genes instead), CIBERSORT immune infiltration (80/20 skip), and external-validation AUCs (our script re-fit per-cohort apparent AUC rather than applying the trained model, and 2/3 validation cohorts had unresolved group labels). No fabrication concern: every reproduced value is derivable from the public data.

💻 Code ↗ 🗄 Data: GSE36791

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 80
    assessed: 2026-06-15 ⛓ da0f6dd9226e
✎ 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
2026-06-15
Rubric version
v1.0
Assessed by
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-06-15
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: opus
Founding hypothesis

The study investigates the potential biomarkers and molecular mechanisms of circulating neutrophils in the peripheral blood transcriptome of aneurysmal subarachnoid hemorrhage (aSAH) patients, hypothesizing that inflammation-related genes drive the systemic neutrophil immune response after intracranial aneurysm rupture.

Core claims
  • Six genes (CST7, HSP90AB1, PADI4, PLBD1, RAB32, SLAMF6) are signature diagnostic biomarkers for aSAH identified by LASSO and SVM-RFE. finding
  • A neutrophil-related co-expression module (blue module) is significantly associated with aSAH and enriched for neutrophil activation/degranulation and NET formation. finding
  • MAPK14, ITGAM, TLR4, and FCGR1A are hub genes involved in neutrophil-related peripheral immune activation in aSAH. finding
  • Neutrophils are upregulated post-aSAH and PADI4 is positively correlated with neutrophils; the NETs pathway is significantly upregulated. mechanism
  • A six-gene machine learning diagnostic model discriminates aSAH from controls with robust ROC performance. method
  • CMap analysis predicts small-molecule compounds (e.g., STOCK1N-35874) able to reverse neutrophil-related gene expression as candidate aSAH drugs. resource
  • 602 DEGs (315 up, 287 down) distinguish aSAH from control peripheral blood. finding
Experimental setups
Assay System Perturbation Readout Platform
Microarray gene expression / bulk transcriptome analysis (DEG identification via limma) Peripheral blood cells of 43 aSAH patients vs 18 controls (GSE36791) none (disease vs control observational) Differentially expressed genes (|log2FC|>0.5, FDR p<0.05) Illumina HumanHT-12 V4.0 expression beadchip (GPL10558)
WGCNA gene co-expression network analysis Whole peripheral blood transcriptome, aSAH (GSE36791) none Gene modules correlated with aSAH trait
PPI network analysis / hub gene identification (STRING, Cytoscape cytoHubba) 197 overlapping genes (blue module ∩ DEGs) none Hub genes by degree centrality STRING (min interaction score 0.7)
Machine learning feature selection (LASSO + SVM-RFE) and ROC validation aSAH datasets GSE36791 (training); GSE73378, GSE15629, GSE13353 (validation) none Signature genes and diagnostic AUC
CMap connectivity map drug-signature matching (XSum, topN=200) Cell line drug-treatment expression profiles vs neutrophil-related genes drug (in silico) CMap reversal scores of small-molecule compounds
GSEA pathway enrichment Whole transcriptome, aSAH vs control (GSE36791) none Normalized enrichment scores of pathways MSigDB gene sets
Immune cell deconvolution (CIBERSORT, LM22) and ssGSEA aSAH vs control peripheral blood (GSE36791) none Fraction/abundance of 22 immune cell types and immune function
GO and KEGG functional enrichment analysis Genes in blue, tan, turquoise modules (aSAH) none Enriched biological processes and pathways
Key results
  • 602 DEGs identified between aSAH and control (315 up, 287 down) 602 DEGs
  • Six overlapping signature genes selected by LASSO (25 genes) and SVM-RFE (8 genes)
  • CST7, PADI4, PLBD1, RAB32 upregulated; HSP90AB1 and SLAMF6 downregulated in aSAH
  • Six-gene model AUC in training set GSE36791 AUC=0.906
  • Model validation AUCs in external datasets AUC 0.67 (GSE13353), 0.636 (GSE15629), 0.572 (GSE73378)
  • Neutrophils significantly elevated in aSAH vs control by CIBERSORT p<0.001
  • NET formation pathway most upregulated in aSAH by GSEA NES=2.12
  • CD8 T cells negatively correlated with neutrophils r=-0.72
Key statistics
  • correlation cor=0.48, p=9e-05 (blue); cor=-0.56, p=2e-06 (turquoise); cor=-0.5, p=4e-05 (tan) (WGCNA module-trait correlations with aSAH)
  • count 602 DEGs (315 up, 287 down) (aSAH vs control DEGs in GSE36791)
  • other AUC=0.906 (Training-set ROC of six-gene model)
  • pvalue p<0.001 (Neutrophil proportion higher in aSAH (CIBERSORT))
  • correlation -0.72 (CD8 T cell vs neutrophil correlation)
  • other NES=2.12 (NET formation), 2.01 (FcγR phagocytosis), 1.96 (TLR), 1.81 (chemokine), 1.69 (leukocyte transendothelial migration), 1.61 (platelet activation) (GSEA upregulated pathways in aSAH)
  • other soft threshold power=12, scale-free fit index 0.9 (WGCNA network construction)
  • count 186 nodes and 95 edges from 197 overlapping genes (PPI network of blue module ∩ DEGs)

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 bioinformatics study analyzed public microarray transcriptome data (GSE36791; 43 aSAH vs. 18 controls) using WGCNA to identify co-expression modules, limma-based differential expression analysis, and machine-learning feature selection (LASSO and SVM-RFE) to derive a six-gene diagnostic signature. Model discrimination was assessed by AUC/ROC in the discovery set and three external validation GEO datasets. Immune cell composition was estimated computationally via CIBERSORT and ssGSEA, with inter-group differences assessed by Wilcoxon rank-sum tests and correlations by Pearson and Spearman methods.

Replicationbiological Sample size43 aSAH patients and 18 headache controls (no aneurysm) from GSE36791; three external GEO datasets used for validation (sample sizes not specified in text) GroupsaSAH patients vs. headache controls without intracranial aneurysm Pairingunpaired Randomization/blindingnot stated Dispersionnone Exact p-valuesyes Effect sizesyes Confidence intervalsno Multiplicity correctionBenjamini-Hochberg FDR (for DEG identification via limma and for GSEA); no correction stated for six Wilcoxon comparisons of signature genes
Statistical tests used
Test Applied to n Assumptions
limma moderated t-test with Benjamini-Hochberg FDR correction Differential expression analysis (43 aSAH vs. 18 controls, GSE36791) 43 aSAH, 18 controls stated
Spearman correlation WGCNA module-trait (aSAH status) associations; gene-immune-cell correlation analysis 61 samples (43+18) not stated
Wilcoxon rank-sum test Comparison of expression levels of six signature genes between aSAH and control groups 43 aSAH, 18 controls not stated
ROC/AUC analysis Discrimination performance of the six-gene model in training (GSE36791) and three external validation sets Training: 61; validation sets sizes not stated in excerpt na
LASSO regression (penalized logistic regression via glmnet) Feature selection from 197 candidate genes to derive diagnostic signature 197 genes, 61 samples not stated
SVM-RFE with five-fold cross-validation Feature selection from 197 candidate genes, cross-validated 197 genes, 61 samples not stated
Pearson correlation General inter-variable correlations (stated in statistical analysis section) null not stated
GSEA (gene set enrichment analysis) with FDR adjustment Whole-transcriptome pathway enrichment, ranked by limma log2FC; p < 0.05 and FDR-adjusted p < 0.05 61 samples not stated
Approaches that could also have been used
  • Six individual Wilcoxon rank-sum tests were performed to compare signature gene expression between aSAH and control groups without a stated multiplicity correction
    Could also: Apply a Benjamini-Hochberg FDR or Bonferroni correction across the family of six simultaneous comparisons — When multiple tests are conducted on the same dataset, a family-wise correction is a standard approach to account for the increased probability of at least one false positive; reporting adjusted p-values alongside unadjusted ones is common practice in gene-expression studies
  • Model performance was evaluated solely by AUC from ROC curves in training and validation sets
    Could also: Additionally report calibration metrics (e.g., Hosmer-Lemeshow test, calibration plots) or Brier scores alongside AUC — AUC measures discrimination (rank-ordering) but not calibration (agreement between predicted probabilities and observed outcomes); supplementing AUC with calibration metrics provides a more complete picture of diagnostic model performance
  • Spearman correlation was used to relate co-expression modules to the binary aSAH/control trait in WGCNA
    Could also: Use a point-biserial correlation or a logistic regression-based module-trait association, which are specifically designed for a continuous predictor against a binary outcome — Spearman correlation is applicable and widely used in WGCNA pipelines; point-biserial correlation or biserial approaches are alternatives that explicitly model the binary nature of the trait and can yield slightly different sensitivity when group sizes are unequal
  • Immune cell deconvolution was performed with CIBERSORT (LM22 signature matrix)
    Could also: Use complementary deconvolution tools such as xCell, TIMER, or MCP-counter in parallel — Different deconvolution algorithms use distinct reference matrices and statistical models; running multiple tools and comparing concordant results is a common strategy to increase confidence in immune infiltration estimates, particularly for cell types like neutrophils that can be difficult to deconvolve from microarray data
  • Feature selection combined LASSO and SVM-RFE, taking the intersection as the final signature
    Could also: Also apply elastic net regression (combining L1 and L2 penalties) or random forest variable importance as additional or alternative selectors — Elastic net can handle correlated predictors more stably than pure LASSO; random forest importance is non-parametric and captures non-linear relationships; comparing multiple selectors and reporting stability across methods is a common approach to assess robustness of the selected gene set
  • The discovery dataset (n = 61) was used both for WGCNA module identification and for training the diagnostic model, with generalization assessed in separate external datasets
    Could also: Perform an internal cross-validation split or bootstrapping on the discovery set to report bias-corrected training AUC alongside the external validation AUCs — Reporting only training AUC (0.906) alongside lower external validation AUCs (0.57–0.67) without an internal cross-validated estimate makes it harder to separate optimism from true generalization; a cross-validated training AUC would provide an intermediate benchmark
Software: R/limma R 4.1.1; limma version not stated · R/WGCNA R 4.1.1; WGCNA version not stated · R/clusterProfiler not stated · R/glmnet (LASSO) not stated · R/e1071 and caret (SVM-RFE) not stated · Cytoscape/cytoHubba not stated · STRING database not stated · R/ggplot2 and GseaVis not stated

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.

Citations
16
Impact: medium
Foundation confidence
None of its references are in our reproducibility record yet — its foundation cannot be assessed.
Topics

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 — pmid-36438795

Paper: Weng, Cheng & Zhang (2022) Front Pharmacol — "Specific signature biomarkers highlight the potential mechanisms of circulating neutrophils in aneurysmal subarachnoid hemorrhage." DOI 10.3389/fphar.2022.1022564.

Nature of the study: A pure secondary-analysis / bioinformatics paper on a public GEO microarray dataset. No wet-lab experiments. The "Code" link in the brief (github.com/junjunlab/GseaVis) is a third-party GSEA visualization R package, NOT the authors' analysis code — there is no authors' repo. Per rule 2 (P16) we reproduce by re-running the described standard pipeline (limma / clusterProfiler / glmnet / pROC) on the paper's own data (GSE36791). Equally valid.

Data: GSE36791 (training) — Illumina HumanHT-12 V4.0 beadchip (GPL10558), peripheral blood, 43 aSAH / 18 control. Validation: GSE73378, GSE15629, GSE13353.

In scope (pipeline-derived, attempted)

id result reported pipeline
C1 DEG count aSAH vs control 602 (315 up / 287 down), |log2FC|>0.5 & FDR<0.05 GEOquery + limma
C2 GO/KEGG enrichment themes top BP = neutrophil activation/degranulation/granulocyte migration; KEGG NET formation, FcγR phagocytosis, chemokine signalling clusterProfiler + org.Hs.eg.db
C3 Diagnostic AUC of the 6-gene signature on training set AUC = 0.906 glm logistic on CST7/HSP90AB1/PADI4/PLBD1/RAB32/SLAMF6 + pROC
C4 per-gene ROC (audit) + validation-set AUC (best-effort) GSE13353 0.67 / GSE15629 0.636 / GSE73378 0.572 pROC on re-applied signature

Out of scope (not attempted, with reason)

  • WGCNA module construction ("blue module") — soft-threshold / module colours are seed- and parameter-sensitive and the paper does not pin them. The paper runs GO on the blue module; we run GO on the DEG set instead → expect thematic, not exact, agreement (recorded as partial).
  • De-novo LASSO (25 genes) + SVM-RFE (8 genes) selection — the exact final 6-gene set depends on the WGCNA module input, glmnet lambda/seed and e1071 RFE internals, none pinned. We instead re-apply the published 6 genes (the clean 1:1) for C3/C4 rather than re-deriving them.
  • CIBERSORT immune infiltration (neutrophil p<0.001; CD8-T vs neutrophil r=-0.72) — LM22 signature + permutation; high-effort, low marginal value under 80/20. Skipped.
  • GSEA NES values — visualization-only (GseaVis); the NES numbers depend on ranking metric + gene-set version; skipped.

Anchor: C1 (DEG count) is the primary clean comparison; C3 (signature AUC) is the secondary re-apply anchor.

C0
Reported
43 aSAH / 18 control
Reproduced
43 / 18
exact
C1
Reported
602 DEGs (315 up / 287 down)
Reproduced
571 DEGs (299 up / 272 down)
within tolerance
C2
Reported
top GO BP neutrophil activation/degranulation; KEGG NET/FcgammaR/chemokine
Reproduced
GO on DEGs: immune/T-cell/leukocyte activation (BP=421); KEGG blocked on «infra»
partial
C3
Reported
6-gene signature AUC 0.906
Reproduced
0.938 (combined glm; per-gene 0.845-0.907)
within tolerance
C4
Reported
validation AUC 0.67/0.636/0.572
Reproduced
not comparable (in-sample refit; 2/3 groups unresolved)
m.public.grade.not-comparable

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 80/100

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)
🤝
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.

87.5 k
tokens (I/O) · 6.8 M incl. cache
19 min
runtime · 0.04 CPU-h
4.8 GB
peak RAM
1
HPC jobs
hummel
machine