TGF-β-mediated activation of fibroblasts in cervical cancer: implications for tumor microenvironment and prognosis.
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.
- Nothing in this column.
- 🟡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
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 for the SHIPPED-data portion. The headline prognostic model's downstream TCGA results reproduce 1:1: risk-score time-ROC AUC (0.75/0.78/0.76/0.74/0.72) and risk-group KM (p<1e-4, groups 145/146) and the TGF-score KM (p=0.0014) all match to rounding when recomputed from the repo's own per-sample tables with survival/timeROC. The 3-gene model (ITGA5/SHF/SNRPN) and its coefficients are taken from the shipped artifact (gene_coef.csv); they could NOT be independently re-derived because (a) the TCGA expression matrix CESC_TPM.txt is not shipped (only clinical+CNV) and (b) gene selection used a stochastic resampling loop with an un-recorded seed. The external validation on GSE44001 (genuinely downloaded + the published coefficients applied) is the weak point: high-risk does trend to worse DFS (KM p=0.043) but time-ROC AUC is only ~0.6 at 1-5y across fixed-coef/z-score/re-fit variants, below the paper's 'good AUC' claim, and ITGA5's weight collapses to ~0 when re-fit on GSE44001 -- flagged as a possible over-statement of external robustness. NOT attempted: de-novo LASSO/WGCNA (un-shipped TPM), scRNA 9-cell-type landscape (Seurat, hard 20%), and all wet-lab Fig 9 assays (out of scope). «host» holds only small result files + pointers; data/compute on «infra»/«our HPC».
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 79assessed: 2026-06-14 ⛓ d9c6c45c1465
✎ 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: opusThis study tests whether aberrant activation of TGF-β signaling in cancer-associated fibroblasts shapes the immunosuppressive tumor microenvironment and drives progression of cervical cancer, and whether TGF-β-related genes can yield a clinically useful prognostic model.
- ★ TGF-β signaling activity is significantly increased in cervical cancer fibroblasts compared to normal fibroblasts and promotes their proliferation/differentiation. finding
- ★ Strong TGF-β-mediated communication between fibroblasts and macrophages and NK/T cells contributes to an immunosuppressive microenvironment. mechanism
- ★ A three-gene prognostic model (ITGA5, SHF, SNRPN) derived from TGF-β-related WGCNA modules predicts cervical cancer survival across multiple datasets. resource
- ★ ITGA5 and SNRPN are upregulated and SHF is downregulated in cervical cancer cells relative to normal cervical epithelial cells. finding
- ★ ITGA5 knockdown suppresses viability, migration, and invasion of cervical cancer cells. finding
- ★ Single-cell analysis of the cervical cancer TME identifies nine major cell types, with fibroblasts central to TGF-β activation. finding
- WGCNA identifies gene modules significantly associated with the TGF-β signaling pathway used for downstream prognostic modeling. method
- TGF-β signaling activity correlates with suppression of inflammatory pathways, immune escape, and metabolic regulation in fibroblasts. mechanism
| Assay | System | Perturbation | Readout | Platform |
|---|---|---|---|---|
| single-cell RNA-seq analysis (Seurat/Harmony/UMAP clustering) | human cervical cancer TME (GSE208653: 2 normal + 3 HPV-infected CC samples) | none | cell-type identification and TGF-β signaling activity (AUCell score) | 10x Genomics (Read10X); GSE208653 |
| pseudo-time trajectory analysis | fibroblasts from normal vs CC samples (GSE208653) | none | differentiation trajectory and gene expression dynamics | Monocle |
| cell-cell communication analysis | cervical cancer TME single-cell subpopulations | none | ligand-receptor TGF-β signaling interactions | CellChat |
| WGCNA and Cox/LASSO prognostic modeling | TCGA-CESC bulk RNA-seq (291 tumor samples) and GSE44001 (300 samples) | none | prognostic gene modules, risk score, survival (K-M, ROC) | WGCNA, glmnet, timeROC |
| quantitative real-time PCR | Hela CC cells and Ect1/E6E7 normal cervical epithelial cells | none / si-ITGA5 knockdown | mRNA levels of ITGA5, SHF, SNRPN (2^-ΔΔCT, GAPDH normalizer) | SYBR Green qPCR (Beyotime) |
| CCK-8 cell viability assay | Hela CC cells | si-ITGA5 vs si-NC transfection | cell viability (OD 450 nm) | CCK-8 (Beyotime); Bio-Rad iMark reader |
| scratch/wound healing migration assay | Hela CC cells | si-ITGA5 vs si-NC transfection | wound closure (%) / migration | Olympus DP27 microscope |
| transwell invasion assay (Matrigel) | Hela CC cells | si-ITGA5 vs si-NC transfection | number of invaded cells (crystal violet) | Corning 8 µm transwell; Olympus DP27 |
- ▲ TGF-β signaling AUCell score markedly higher in CC fibroblasts than normal fibroblasts
- ▲ TGF-β-related genes ID2, PPP1R15A, SMAD7, XIAP significantly hyperexpressed in CC fibroblasts
- – CC fibroblasts located at the end of the pseudo-time differentiation trajectory while normal fibroblasts at the start
- – Three-gene model (ITGA5, SHF, SNRPN) shows good predictive ability across TCGA-CESC and GSE44001
- – ITGA5 and SNRPN higher and SHF lower expression in CC cells vs normal cervical epithelial cells
- ▼ ITGA5 knockdown suppressed viability, migration and invasion of CC cells
- – Nine major cell types identified in the cervical cancer TME with fibroblast markers COL1A2, DCN, COL1A1 highly expressed
- ▲ TGF-β signaling activity positively correlated with negative-regulation-of-immune inflammatory pathways
- count 604,000 new cervical cancer cases worldwide in 2020 (WHO global CC incidence)
- count 342,000 deaths (global CC mortality 2020)
- count 291 tumor samples (TCGA-CESC samples used for analysis)
- count 300 tumor samples (GSE44001 validation samples retained)
- count 2 normal + 3 HPV-infected CC samples (GSE208653 single-cell dataset)
- count nine cell types (major cell types identified in TME)
- other >90% (share of global CC burden in less developed nations)
- pvalue p < 0.05 (threshold defined as statistically significant)
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 bioinformatics and translational study combined single-cell RNA-seq analysis (GSE208653; 5 samples), bulk RNA-seq prognostic modeling (TCGA-CESC; n=291 tumors split 70/30 training/validation), and an independent microarray cohort (GSE44001; n=300 tumors) with in vitro functional validation in HeLa cells. Pathway activity was scored via AUCell and gene co-expression modules were identified by WGCNA; a three-gene prognostic risk score was constructed by sequential univariate Cox, LASSO Cox, and multivariate Cox regression. Survival differences between risk groups were assessed by Kaplan–Meier analysis with log-rank test, and model discrimination was characterized by time-dependent ROC curves.
| Test | Applied to | n | Assumptions |
|---|---|---|---|
| Wilcoxon rank-sum test | Two-group comparisons of continuous variables (e.g., TGF-β AUCell scores in normal vs CC fibroblasts) | not stated per comparison | not stated |
| Student's t-test | Two-group comparisons of continuous variables (stated alongside Wilcoxon in statistical analysis section) | not stated per comparison | not stated |
| Pearson correlation | Correlation of TGF-β AUCell scores with inflammatory, proliferative, and metabolic pathway AUCell scores in CC fibroblasts | not stated | not stated |
| Spearman correlation | General correlations as stated in the statistical analysis section | not stated | not stated |
| Univariate Cox proportional hazards regression | Screening prognostic relevance of WGCNA module genes in TCGA-CESC training set | ~204 (70% of 291) | not stated |
| LASSO Cox regression (glmnet) | Penalized feature selection from univariate Cox candidate genes | ~204 (70% of 291) | not stated |
| Multivariate Cox regression | Final risk score coefficient estimation for ITGA5, SHF, SNRPN | ~204 (70% of 291) | not stated |
| Kaplan–Meier analysis with log-rank test | Overall survival comparison between high- and low-risk groups in TCGA-CESC training set, TCGA-CESC validation set, and GSE44001 | 291 (TCGA-CESC total), 300 (GSE44001) | not stated |
| Time-dependent ROC curve (timeROC package) | Discriminative performance of the prognostic model at multiple survival time horizons | not stated per dataset | na |
| ssGSEA (GSVA package) | 28-immune-cell type infiltration scoring across TCGA-CESC samples stratified by risk group | 291 | na |
| AUCell scoring | TGF-β signaling pathway activity per cell in scRNA-seq data; also applied to inflammatory, proliferative, and metabolic gene sets | 5 samples (GSE208653: 2 normal, 3 CC) | na |
-
Pathway activity in individual cells was quantified with AUCell scores, which use an area-under-the-recovery-curve statistic over ranked genes↳ Could also: UCell or single-sample GSVA could also score pathway activity per cell — UCell uses a Mann–Whitney U-based rank statistic that is not sensitive to dataset size or expression-level normalization choices; GSVA is the established bulk-RNA counterpart extended to single-cell contexts — comparing two or more scoring methods can reveal whether conclusions are robust across approaches
-
Pearson correlation was used to relate TGF-β AUCell scores to other pathway scores in fibroblasts, while Spearman was stated as the general correlation method elsewhere in the same paper↳ Could also: Spearman rank correlation could also be applied to AUCell score correlations for consistency — AUCell scores are bounded and may be non-normally distributed; Spearman is more robust to non-normality and outliers and is already the paper's stated default, so applying it uniformly would make the correlation analyses internally consistent
-
Multiple individual gene-level comparisons between normal and CC groups (and between high/low TGF-β subgroups) were made without a stated multiplicity correction↳ Could also: Benjamini–Hochberg false discovery rate (FDR) correction could also be applied across the family of simultaneous gene-level tests — When many genes are tested simultaneously, FDR control at a chosen threshold (commonly 5%) is a standard approach in omics research that keeps the expected proportion of false positives interpretable without being as conservative as Bonferroni
-
Prognostic gene selection used a sequential pipeline of univariate Cox → LASSO Cox → multivariate Cox regression↳ Could also: Elastic net Cox regression (combining L1 and L2 penalties) could also perform joint feature selection and coefficient shrinkage in a single step — Elastic net can handle correlated predictors more stably than pure LASSO, which may arbitrarily retain one gene among a group of highly correlated candidates; it is a frequently used alternative when collinearity among gene expression features is expected
-
Internal validation used a single 70/30 random split of the TCGA-CESC dataset↳ Could also: Repeated k-fold cross-validation (e.g., 5- or 10-fold, repeated multiple times) or bootstrap-based optimism correction could also estimate internal model performance — A single random split can produce variable performance estimates depending on the specific partition drawn; cross-validation averages over multiple splits to reduce this variance and is a common complement to external cohort validation
-
Survival differences between risk groups were assessed with the standard (unweighted) log-rank test↳ Could also: A Cox proportional hazards model comparison, or a weighted log-rank test (e.g., Fleming–Harrington), could also be used — The standard log-rank test is most powerful when hazard ratios are proportional and constant over time; a Cox model provides an effect size estimate (hazard ratio with confidence interval) in addition to a p-value, and weighted variants can better detect early or late divergence in survival curves — both are common in cancer survival studies
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.
Data lineage
The datasets this paper uses (text-mined from the full text via Europe PMC), and which other assessed papers stand on the same data. A shared dataset is a factual link — not a judgement.
Downstream reach in the literature
95 downstream papers · 2 datasetsHow widely the datasets deposited by this paper are reused across the whole literature (Europe PMC), beyond our assessed set. This is a factual dependency map — reusing a public dataset is normal, good science. It is not a judgement on the downstream papers; the only verdict here is this paper's own, with its cited rationale.
- LncRNA EPB41L4A-AS1 regulates glycolysis and glutami... 2019 · 120 cites
- A multi-omics-based investigation of the prognostic... 2022 · 46 cites
- Genomic Characterization of Six Virus-Associated Can... 2018 · 35 cites
- Expression of fibroblast growth factor receptor fami... 2016 · 32 cites
- Eight-lncRNA signature of cervical cancer were ident... 2021 · 32 cites
- A novel cuproptosis-related gene signature for predi... 2022 · 29 cites
- Interactions of Indoleamine 2,3-dioxygenase-expressi... 2023 · 55 cites
- Proteogenomic characterization of molecular and cell... 2025 · 10 cites
- Pan-Cancer Analyses Refine the Single-Cell Portrait... 2025 · 8 cites
- A pan-cancer single-cell transcriptional analysis of... 2024 · 5 cites
- HPV16 E6-induced M2 macrophage polarization in the c... 2024 · 5 cites
- Elucidating the tumor microenvironment interactions... 2025 · 3 cites
What was reproduced
The exact results taken into scope, with each reported value next to the value our attempt produced.
Scope — pmid-40124621
Paper: TGF-β-mediated activation of fibroblasts in cervical cancer: implications for tumor microenvironment and prognosis. PeerJ 2024; DOI 10.7717/peerj.19072. PMID 40124621 · PMCID PMC11929507. Code: https://github.com/21kunzhang/raw-data (pushed 2024-12-20, public, no license). Data: GEO GSE208653 (scRNA, 5 samples), GSE44001 (microarray, 300 tumors, GPL14951, DFS endpoint), TCGA-CESC (TPM expression + clinical + CNV).
Datasets — what is / isn't shipped in the repo
- scRNA GSE208653 RAW (10x: barcodes/features per sample) — SHIPPED under
00_origin_datas/GEO/GSE208653_RAW/(5 samples: 3 CA_HPV, 2 NO_HPV). - TCGA-CESC clinical + gene-level CNV — SHIPPED (
00_origin_datas/TCGA/). - TCGA-CESC expression
CESC_TPM.txt— NOT shipped (script reads it; absent). → any de-novo step needing TCGA expression (LASSO gene selection, WGCNA, risk score from expression) requires GDC/Xena download. - GSE44001 expression/clinical — NOT shipped; script downloads via a custom
getGEOExpData()helper → reproduce via GEOquery. - Many intermediate result tables are SHIPPED (gene_coef.csv, tcga.risktype.cli.txt, TGF.score.txt, WGCNA_Modules.csv, aucs.csv, …) — these let us reproduce downstream steps without the missing raw expression.
Pipeline stages (numbered folders) and scope decision
| stage | analysis | pipeline | in scope? |
|---|---|---|---|
| 01_landscape | Seurat scRNA QC/cluster/UMAP, 9 cell types | Seurat+harmony | partial (C8, heavy — hard 20%) |
| 02_AUCell | AUCell TGF/escape scoring, DEG | AUCell | descriptive, no single pinnable number → not attempted |
| 03_pseudotime | monocle BEAM trajectory | monocle | descriptive → not attempted |
| 04_immu_metab | DEG + GO/KEGG + Pearson corr | clusterProfiler | descriptive → not attempted |
| 05_CellChat | ligand–receptor TGF communication | CellChat | descriptive → not attempted |
| 06_TGF | TGF ssGSEA score → KM(OS) | survival | IN SCOPE (C5) p=0.0014 shipped |
| 07_WGCNA | WGCNA modules vs TGF score | WGCNA | IN SCOPE (C7), heavier, needs TCGA TPM |
| 09_model | univ-Cox→LASSO-Cox→risk score, KM, ROC | glmnet/survival/timeROC | IN SCOPE (C1-C4, C6) core |
| 10_immu | ssGSEA immune vs risk group | GSVA | descriptive → not attempted |
| Raw experimental data | EdU, CCK-8, transwell invasion, wound healing, qPCR (ITGA5 KD in Hela) | wet-lab | OUT OF SCOPE (manual/experimental) |
What we attempt (80/20), in priority order
- C1/C2 — confirm the 3-gene model + coefficients (from shipped gene_coef.csv; note de-novo LASSO is stochastic resampling and needs un-shipped TPM → not exactly re-derivable, flagged).
- C3/C4 — reproduce TCGA risk-score time-ROC AUC (1–5y) and risk-group KM p-value
from the shipped per-sample risk scores (
tcga.risktype.cli.txt). Pure downstream recompute, no download → strongest internal check. - C5 — reproduce TGF-score KM(OS) p from shipped
TGF.score.txt+ OS, median split. - C6 — external validation on GSE44001: apply the published 3-gene coefficients, compute risk score, time-ROC (DFS) + KM. Genuine external reproduction (needs GEO DL).
- C7 (optional/heavier) — WGCNA on TCGA-CESC expression: confirm 17 merged modules at β=6 and brown↔TGF positive correlation. Needs TCGA TPM download.
- C8 (hard 20%) — scRNA 9 cell types: clustering-resolution sensitive; attempt only if budget remains.
Out of scope (not attempted), with reason
- All Fig 9 cellular validation assays (EdU/CCK-8/transwell/wound/qPCR) — wet-lab, not pipeline-derived.
- AUCell / pseudotime / CellChat / DEG-enrichment figures — descriptive, no single reported scalar to compare 1:1 (would be qualitative only).
Hard-rule compliance
All compute on «our HPC» SLURM; all downloads (GSE44001, optional TCGA TPM) land on
«infra» reproductions/pmid-40124621/. «host» holds only small result files +
pointers. See AUDIT.md for honesty flags.
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.
The headline computational claims reproduce essentially 1:1: recomputed from the repo's own shipped per-sample tables, TCGA risk-score time-ROC AUC (0.750/0.784/0.761/0.743/0.722), risk-group KM (p=2.5e-6, groups 145/146) and TGF-score KM (p=0.0014) all match the reported values to rounding, with no fabrication signal on the core. The genuine deviation is on the authors' side in external validation: GSE44001 reproduces only AUC ~0.53-0.65 (across three methods) against the paper's 'good AUC' claim, and ITGA5's weight collapses to ~0 — the direction/significance still hold (KM p=0.043) so it is a moderate over-statement, not a flipped conclusion. Two availability caveats (unshipped CESC_TPM.txt + un-recorded resampling seed) mean the upstream model selection is not byte-reproducible, but that is a data-availability limit, not a defect. Overall: solid reproduction with one explainable, documented over-statement of external robustness → 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.