Comprehensive analysis of transcriptomics and radiomics revealed the potential of TEDC2 as a diagnostic marker for lung adenocarcinoma.
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.
- ✓Reported values were directly comparable
- ✓No authors-side cause for any deviation
- ✓Reported values are derivable from the shared data
- ✓Any deviation was negligible
- ✓The central claim held under reproduction
- 🟡Could not use the authors’ exact input data
- 🟡A deviation arose in the data or preprocessing
- 🟡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 to reproduce the CENTRAL claim 1:1 from public data; verdict 'partial' = central claim reproduced strongly (within-tol), coverage intentionally partial. This is a clean rebuild after an administrative batch requeue (reason 'no-reproduction-dir' — the requeue process archived the prior reproduction/ dir into requeue_archive/ and emptied the canonical path). Compute re-ran FRESH on «our HPC» «job» (node n095, partition std, COMPLETED 24s, exit 0:0) reading the pre-staged «infra» inputs; outputs are byte-identical to the prior deterministic run («job», node n111) — rank/closed-form statistics, no RNG. The repo (github.com/taochao1/Raw-data @631739f0, authors' OWN R code) ships method scripts + figures + some result tables, but its preprocessing load()s expression-matrix RData that are NOT shipped and depends on an unpublished custom helper pkg (mg*, getGEOExpDataByCel, clean_TNMStage, gencode.pcg) -> verbatim re-execution blocked (env_unresolvable for preprocessing). So I reproduced INDEPENDENTLY with standard tools on the same public data. The paper's central diagnostic-marker result (single-gene ROC AUC of UBE2T/TEDC2/RCC1/FAM136A for tumor-vs-normal, the exact pROC roc(tissue~gene) method) reproduced across THREE cohorts: TCGA-LUAD (UCSC Xena STAR-TPM) and GEO GSE31210 + GSE30219 (GPL570). Validation-cohort sample counts matched the paper EXACTLY (GSE31210 226t/20n; GSE30219 293t/14n); TCGA 528t/59n vs paper 513t/59n (data-freeze version diff). 11 of 12 AUCs within 0.03 of reported (e.g. TCGA UBE2T 0.988 vs 0.989, GSE30219 FAM136A 0.934 vs 0.94); the one outlier, TEDC2 on GSE31210, reproduced at 0.89 vs reported 0.94 (within 0.05, still strong) — likely series-matrix processed values vs the authors' own CEL re-RMA. Shipped-output audit: the repo's tcga.degs.txt has EXACTLY 1007 up + 1593 down rows and the hub-gene file EXACTLY 214 genes, identical to the paper -> internal consistency holds, no fabrication detected at the output level. Independent DEG re-run (limma-equivalent Welch t + BH on Xena STAR-TPM, all 60660 genes) gave 1254 up / 1925 down — ~22% high because no protein-coding-only filter, but the down>up asymmetry reproduced (ratio 1.535 vs reported 1.582). NOT ATTEMPTED: radiomics model (Fig6), survival/Cox (FigS4), CIBERSORT/ESTIMATE (Fig5), GO/KEGG/GSEA, full WGCNA+ML feature-selection re-run (needs unshipped matrix+custom pkg), wet-lab assays (Fig8). No completeness claim; all grades provisional pending human audit.
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 80assessed: 2026-06-16 ⛓ 89bfdda4a724
✎ 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
- 2026-06-22
- Rubric version
- v1.0
- Assessed by
-
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-06-16no 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: sonnetThe molecular mechanism of lung adenocarcinoma (LUAD) remains unclear when transcriptomics and radiomics are combined, and the paper tests whether integrating transcriptomic screening with radiomic feature modeling can identify a non-invasive diagnostic marker (TEDC2) for LUAD.
- ★ WGCNA identified 214 key genes in the blue module most correlated with LUAD finding
- ★ Intersection of RF, LASSO, and SVM-RFE algorithms identified four diagnostic genes: UBE2T, TEDC2, RCC1, and FAM136A method
- ★ The four diagnostic genes show high individual diagnostic accuracy for LUAD (AUC 0.987-0.989) finding
- ★ Diagnostic marker expression correlates significantly and negatively with stromal and immune scores finding
- ★ A TEDC2-based radiomics model built from seven radiomic features achieves high diagnostic performance (AUC up to 0.96) finding
- ★ Knockdown of TEDC2 slows proliferation, migration, and invasion of LUAD cell lines finding
- ★ Cell cycle progression is hyperactive in LUAD mechanism
- A combined four-biomarker logistic regression model achieves AUC of 0.963 finding
| Assay | System | Perturbation | Readout | Platform |
|---|---|---|---|---|
| WGCNA (weighted gene co-expression network analysis) | TCGA-LUAD and GEO (GSE31210, GSE30219) transcriptome cohorts | none | gene modules correlated with LUAD phenotype | R 'WGCNA' package |
| Differential expression analysis | Merged GEO LUAD tumor vs normal samples | none | differentially expressed genes (DEGs) | R 'limma' package |
| Machine learning feature selection (RF, LASSO logistic regression, SVM-RFE) | TCGA-LUAD transcriptome data | none | diagnostic gene candidates | R packages randomForest, glmnet, e1071 |
| CIBERSORT and ESTIMATE immune infiltration analysis | TCGA-LUAD gene expression matrix | none | immune/stromal/ESTIMATE scores correlated with diagnostic genes | CIBERSORT (LM22), ESTIMATE |
| Radiomics feature extraction from CT images | TCGA-LUAD gross tumor volume ROIs | none | 107 radiomic features (shape, first-order, texture) | PyRadiomics v3.0, ITK-SNAP v3.8.0 |
| LASSO regression radiomics modeling | TEDC2-associated radiomic features | none | Rad score / radiomics model AUC | R 'glmnet' (cv.glmnet) |
| RT-qPCR | A549 (LUAD) and BEAS-2B (normal lung epithelial) cell lines | none | mRNA expression of UBE2T, TEDC2, RCC1, FAM136A | SYBR Green RT-PCR kit (Vazyme) |
| Transwell migration/invasion assay and CCK8 viability assay | A549 LUAD cells | TEDC2 siRNA knockdown | cell migration, invasion, and proliferation/viability | microplate reader (Bio-Rad), inverted microscope |
- ▲ 214 key genes identified in the blue WGCNA module, of which 192 were upregulated in LUAD
- – Intersection of RF, LASSO, and SVM-RFE yielded four diagnostic genes: UBE2T, TEDC2, RCC1, FAM136A
- ▲ ROC AUC values for UBE2T, TEDC2, RCC1, FAM136A AUC 0.989, 0.989, 0.989, 0.987
- ▲ Combined four-gene logistic regression model diagnostic performance AUC 0.963
- – Diagnostic marker expression higher in tumor vs para-cancerous tissue and negatively correlated with stromal/immune scores
- ▲ TEDC2-based radiomics model (7 features) diagnostic performance AUC 0.96
- ▼ TEDC2 knockdown reduced proliferation, migration, and invasion in LUAD cells
- – 2,600 DEGs identified between LUAD and normal samples 1,593 down, 1,007 up
- fold_change AUC = 0.989 (UBE2T diagnostic ROC)
- fold_change AUC = 0.989 (TEDC2 diagnostic ROC)
- fold_change AUC = 0.987 (FAM136A diagnostic ROC)
- fold_change AUC = 0.963 (combined 4-gene logistic regression model)
- fold_change AUC = 0.96 (TEDC2 radiomics model (ROC and PR curves))
- count 214 key genes (blue module genes from WGCNA (MM > 0.7, GS > 0.4))
- count 2,600 DEGs (1,593 down-regulated, 1,007 up-regulated) (differential expression in LUAD vs normal)
- correlation R2 = 0.85, soft threshold = 7 (WGCNA scale-free topology fit)
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.
The paper applies a multi-stage discovery design using three public transcriptomics cohorts (TCGA-LUAD, GSE31210, GSE30219) to identify LUAD diagnostic genes via WGCNA co-expression network analysis followed by intersection of three machine-learning feature-selection algorithms (Random Forest, LASSO logistic regression, SVM-RFE). Differential expression between tumor and normal tissue was tested genome-wide using the limma moderated t-test with FDR adjustment. Diagnostic accuracy of candidate genes and a LASSO-selected radiomic model were evaluated by ROC curve analysis (AUC), and functional validation of the top gene (TEDC2) was performed in LUAD cell lines using RT-qPCR, CCK8 proliferation, and Transwell migration/invasion assays.
| Test | Applied to | n | Assumptions |
|---|---|---|---|
| limma moderated t-test with FDR adjustment (adj. p < 0.05, |log2FC| > 1) | Genome-wide differential expression analysis between LUAD tumor and normal samples (merged GEO cohorts GSE31210 + GSE30219) | GSE31210: 226 tumor + 20 normal; GSE30219: 293 tumor + 14 normal | not stated |
| Student's t-test or Wilcoxon rank-sum test (choice not specified per comparison) | Comparisons of continuous variables including gene expression levels and cell-line assay results | — | not stated |
| Pearson correlation analysis | Correlation between diagnostic marker expression and CIBERSORT immune cell fractions, ESTIMATE immune/stromal scores, and radiomic features | — | not stated |
| Log-rank test | Comparison of survival time between patient groups | — | not stated |
| Cox regression | Survival analysis (stated in statistical analysis section; survival R package) | — | not stated |
| LASSO logistic regression with 5-fold cross-validation (glmnet, alpha=1, nlambda=1000, deviance metric) | Radiomic feature selection correlated with TEDC2 expression; also transcriptomic feature selection (nlambda=100) | 107 radiomic features from TCGA-LUAD CT ROIs; TCGA-LUAD: 513 tumor + 59 normal for transcriptomics | not stated |
| ROC curve analysis (AUC via R/ROCR) | Diagnostic accuracy of individual genes UBE2T, TEDC2, RCC1, FAM136A; combined four-gene logit model; TEDC2-based radiomic model | TCGA-LUAD: 513 tumor + 59 normal | na |
| GSEA (gene set enrichment analysis via clusterProfiler/gseaplot2) | Pathway activation assessment of LUAD full expression profile | — | not stated |
-
Cell-line assay results (CCK8, Transwell) were reported as averages across three experiments without a stated measure of dispersion↳ Could also: Report mean ± SD or mean with 95% CI alongside significance test results — With small n (3 experiments), a dispersion measure allows readers to assess effect magnitude and variability; SD is standard in cell-biology reporting and supports reproducibility assessments
-
Multiple Pearson correlation analyses were performed between diagnostic markers and immune scores/radiomic features without a stated multiple-testing correction for that family↳ Could also: Apply Benjamini-Hochberg FDR correction across the family of simultaneous correlation tests; alternatively use Spearman rank correlation — FDR control reduces the expected proportion of false discoveries when many correlations are tested together; Spearman correlation additionally relaxes the normality and linearity assumptions of Pearson correlation
-
Diagnostic accuracy was summarized by point-estimate AUC values alone↳ Could also: Report 95% confidence intervals for each AUC (e.g., via DeLong's method) and summary calibration statistics (e.g., Brier score) — AUC CIs quantify estimation precision, which matters when comparing models; calibration statistics complement discrimination measures by describing how well predicted probabilities match observed event rates
-
The radiomic model was evaluated using internal 5-fold cross-validation on the same TCGA-LUAD dataset from which features were extracted↳ Could also: Evaluate on a fully independent external imaging cohort or apply nested cross-validation with a separate outer loop for performance estimation — Internal cross-validation after feature selection on the same sample can yield optimistic performance estimates; an external cohort or nested CV provides a less biased estimate of generalizability to new patients
-
Three machine-learning algorithms were each applied to the full training cohort and their gene-selection outputs were intersected↳ Could also: Reserve a held-out test partition before running any of the three algorithms, then evaluate the intersection genes on that unseen partition — Evaluating all three algorithms and their intersection on the same data they were trained on may overfit the intersection to that cohort; a consistent hold-out set across all algorithms provides a more conservative estimate of which intersection genes generalize
-
Student's t-test or Wilcoxon rank-sum test was used for individual gene-expression comparisons across multiple genes without a stated correction↳ Could also: Apply the same limma moderated t-test framework (already used for DEG analysis) with FDR correction uniformly across all per-gene comparisons — A unified linear-model framework provides consistent variance estimation across genes and integrates naturally with the existing FDR correction pipeline, reducing the number of separate test families to track
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.
What was reproduced
The exact results taken into scope, with each reported value next to the value our attempt produced.
Scope — pmid-39553728
Paper: Huang Q, Zhang P, Guo Z, Li M, Tao C, Yu Z. Comprehensive analysis of transcriptomics and radiomics revealed the potential of TEDC2 as a diagnostic marker for lung adenocarcinoma. PeerJ 2024. PMID 39553728 / PMCID PMC11569783 / DOI 10.7717/peerj.18310.
Code: https://github.com/taochao1/Raw-data @ commit
631739f0ecbbbc7016f0d0e723add0f8fb0a0a1a (v1.1.0, 2024-03-06). The authors'
OWN code (R), not a third-party tool. Contains scripts/ (limma/WGCNA/ML/ROC/
radiomics), bundled intermediate RData + result tables + figures.
Data: TCGA-LUAD (513 tumor / 59 normal), GEO GSE31210 (226 tumor / 20 normal, GPL570), GSE30219 (293 tumor / 14 normal, GPL570), plus radiomics CT features (TCIA TCGA-LUAD). All public.
In-scope pipeline-derived results (computational)
| # | Result | Pipeline | Reproducibility |
|---|---|---|---|
| R1 | DEG count TCGA tumor-vs-normal: 1,007 up / 1,593 down (2,600), FDR<0.05 & | log2FC | >1 (Fig S2) |
| R2 | 4 diagnostic genes UBE2T, TEDC2, RCC1, FAM136A from LASSO∩SVM-RFE∩RF intersection (Fig 2/3) | glmnet/caret/randomForest | partially — stochastic, seeds set; needs TCGA matrix + 192 WGCNA hub genes |
| R3 | Single-gene diagnostic AUC, TCGA training (Fig 3B): UBE2T/TEDC2/RCC1 0.989, FAM136A 0.987 | pROC roc(tissue~gene) | needs TCGA TPM matrix |
| R4 | Single-gene AUC GSE30219 (Fig 4A): UBE2T 0.96, TEDC2 0.95, RCC1 0.89, FAM136A 0.94 | pROC | fully independent — public GPL570 data; PRIMARY target |
| R5 | Single-gene AUC GSE31210 (Fig 4C): UBE2T 0.96, TEDC2 0.94, RCC1 0.94, FAM136A 0.93 | pROC | fully independent — public GPL570 data; PRIMARY target |
| R6 | WGCNA: soft-power 7, R²=0.85, 15 modules, blue module 214 hub genes (MM>0.7,GS>0.4) | WGCNA | shipped hub-gene file auditable; full re-run needs TCGA matrix |
| R7 | Radiomics LASSO model, 7 features, train AUC 0.96 / val AUC 0.90 (Fig 6) | PyRadiomics+glmnet | needs TCIA CT + segmentations; harder |
Out of scope (not attempted — wet-lab / manual)
- RT-qPCR (A549 vs BEAS-2B), CCK-8, Transwell migration/invasion (Fig 8) — wet-lab.
- Manual CT segmentation / radiologist steps.
Reproduction strategy
The preprocessing scripts load() expression-matrix RData (luad.tcga.exp.RData,
GSE31210.RData, gse31210.exp.RData) that are NOT shipped in the repo (only
clinical .cli RData + enrichment outputs + figures are), and depend on an
unpublished custom helper package (mg_FPKM2TPMs, getGEOExpDataByCel,
exp_probe2symbol_v2, clean_TNMStage, gencode.pcg). So verbatim re-execution is
blocked (env_unresolvable for preprocessing). The underlying datasets are public,
so we reproduce independently with standard tools:
- R4/R5 (PRIMARY): download GSE31210 & GSE30219 series matrices (GPL570, RMA log2)
- GPL570 platform annotation from GEO; compute pROC-equivalent AUC (Mann-Whitney) of each gene's expression vs tumor/normal labels. AUC is a rank statistic, invariant to the unknown normalization/probe-collapse, so this is a faithful 1:1 check. NB: TEDC2 legacy symbol C16orf59, RCC1 legacy CHC1 — handled via alias map.
- R1 (shipped-output audit, DONE): repo
results/Files/tcga.degs.txthas exactly 1,007 Up + 1,593 Down rows; hub-gene file = 214 genes → matches paper exactly (paper numbers == repo's own outputs; fabrication check passes at repo level). - R3 (harder): TCGA-LUAD TPM via recount3/GDC + pROC — pursued after primary.
All compute on «our HPC» («infra» workdir
«path»).
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 paper's central diagnostic claim reproduces strongly and independently: 11/12 single-gene ROC AUCs across three cohorts fall within 0.03 of reported, and the one outlier (TEDC2 on GSE31210, 0.89 vs 0.94) stays a strong AUC, attributable to series-matrix vs CEL re-RMA normalization. DEG counts came out ~22% high (1254/1925 vs 1007/1593) purely because our independent run skipped the protein-coding filter, yet the repo's shipped files match the paper's 1007/1593/214 exactly — internal consistency holds, no fabrication. Deviations sit on our-method / input-preprocessing side, not the authors', and severity is negligible-to-moderate; verdict is solid-with-explainable-deviations rather than 1:1 because the authors' scripts couldn't be re-executed (unshipped RData + unpublished helper package).
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.