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

M6A-mediated molecular patterns and tumor microenvironment infiltration characterization in nasopharyngeal carcinoma.

Cancer Biol Ther · 2024
L1 76/100 3/4
Why this verdict

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

Reproduced on the brainbox compute brainarbeit.com
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: 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 🟡
Concordant (toward reproduced)
Code + data deposited & functional
-2 pts
From: Data & code availability Available & functional
Total score +3
✓ What held up
  • Reported values were directly comparable
  • Reported values are derivable from the shared data
  • The central claim held under reproduction
What did not (or only partly)
  • 🟡Could not use the authors’ exact input data
  • 🟡A deviation arose in the data or preprocessing
  • 🟡A deviation was attributed to the published material
  • 🟡The deviation was non-trivial in magnitude
  • 🟡Overall, the reproduction showed a material discrepancy
How its reproducibility compares
76/100
Reproducibility score
at the mean
vs. all fields · 1173 studies
🎯 Scores higher than 48% of all assessed papers rank 586 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. Authors' own repo ships the processed GSE102349 data (counts, clinical-PFS, published cluster labels). On «our HPC» (edgeR 4.0.16, ConsensusClusterPlus 1.66.0) three clearly-specified pipeline steps were re-run. STRONGEST 1:1: consensus clustering reproduces the published 64/24 m6A subtyping almost exactly (63/25, K=2, 98.9% concordance, 1 of 88 samples relabeled). edgeR DEG count 1558 is bracketed by our filtered (1406) and unfiltered (1772) runs with matching up/down direction; exact integer is edgeR-version/prefilter dependent, and all 4 downstream LASSO hub genes reappear as DEGs. Univariate Cox calls fewer prognostic m6A genes (4 vs 10) but 3 of 4 overlap the paper's set and the misses are just over p=0.05 - explained by using log2(CPM+1) as a proxy for the authors' TPM (TPM matrix NOT shipped and not exactly regenerable: per-gene lengths are lost when counts are collapsed to symbols). NOT ATTEMPTED: ssGSEA/CIBERSORT/ESTIMATE/GSVA/pRRophetic/TIDE/SubMap/GSEA/TF-prediction (Figs 2-5,9-10; mostly directional not single-number), the full LASSO signature (needs non-regenerable TPM + two RNG seeds), and the wet-lab 255-sample qPCR/IHC validation (Fig 8; out of scope, manual). No fabrication indicator: every checked number is derivable from the shipped data within margins explained by the TPM-proxy and edgeR version/prefilter ambiguity.

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 76
    assessed: 2026-06-15 ⛓ 51aa23dabd02
✎ 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

What are the m6A-related molecular patterns in nasopharyngeal carcinoma (NPC) and how does m6A modification regulate tumor microenvironment (TME) cell infiltration, therapeutic response, and prognosis?

Core claims
  • NPC samples can be divided into two distinct m6A-related molecular subclasses based on prognostic m6A regulators finding
  • Cluster 1 is characterized by immune-related and metabolism pathway activation with better response to anti-PD1/anti-CTLA4 and chemotherapy, while cluster 2 shows stromal activation, low HLA/immune checkpoint expression, and worse therapeutic response finding
  • Cluster 2 patients have poorer progression-free survival (PFS) than cluster 1 patients finding
  • An m6A-related prognostic signature/risk model and nomogram were constructed to predict PFS in NPC patients resource
  • REEP2, TMSB15A, DSEL, and ID4 are upregulated in NPC tumor samples versus non-tumor tissue finding
  • High expression of REEP2 and TMSB15A is associated with poor survival in NPC patients finding
  • REEP2, TMSB15A, DSEL, and ID4 interact with m6A regulators mechanism
  • m6A modification plays an important role in regulating TME heterogeneity and complexity in NPC mechanism
Experimental setups
Assay System Perturbation Readout Platform
bulk mRNA RNA-seq (public dataset analysis) NPC patient tumors (GSE102349, 113 patients; 88 with PFS) none mRNA expression, consensus clustering, PFS Illumina HiSeq 2000
whole genome expression microarray (public dataset analysis) NPC primary tumor tissues (18) and non-cancerous nasopharyngeal tissues (18) (GSE53819) none mRNA expression, subclass validation Agilent-014850 Whole Human Genome Microarray 4x44K G4112F
immune cell infiltration estimation (ssGSEA, CIBERSORT, ESTIMATE) NPC tumors (GSE102349) none immune/stromal scores, relative abundance of 22/24 immune cell types
immunotherapy response prediction (TIDE, SubMap) NPC m6A-related subclasses (GSE102349) none predicted response to anti-PD-1/anti-CTLA4, HLA and immune checkpoint expression
chemotherapy response prediction (pRRophetic) NPC m6A-related subclasses (GSE102349) none IC50 of antitumor drugs GDSC database
immunohistochemistry (IHC) 255 primary NPC samples and adjacent normal nasopharyngeal tissue (Kunming Medical University) none protein expression of REEP2, TMSB15A, DSEL, ID4 and immune cell markers (CD20, CD27, CD68, CD4, CD45RO, CD8) antibodies: DSEL NBP2-56370 Novus, ID4 LS-B9923 Lsbio, REEP2 bs-15198R bioss, TMSB15A sc-271649 Santa Cruz
LASSO/Cox prognostic risk model and nomogram construction NPC patients (GSE102349, 7:3 training/test split) none risk score, PFS, time-dependent ROC/AUC
transcription factor prediction network analysis REEP2, TMSB15A, DSEL, ID4 hub genes none TF-hub gene interaction network NetworkAnalyst/JASPAR, Cytoscape 3.9.1
Key results
  • 88 NPC patients clustered into cluster 1 (64 patients) and cluster 2 (24 patients) 64 vs 24
  • 10 of 21 m6A regulators (IGF2BP1, ALKBH5, YTHDF2, ELAVL1, WTAP, LRPPRC, HNRNPA2B1, CBLL1, RBM15B, YTHDF1) significantly associated with PFS
  • All 10 prognostic m6A regulators upregulated in cluster 2 versus cluster 1
  • Cluster 2 patients show poorer PFS than cluster 1 patients (Kaplan-Meier)
  • 1558 DEGs identified between the two m6A-related subclasses 1558 genes
  • REEP2, TMSB15A, DSEL, ID4 upregulated in NPC versus non-tumor samples
  • High REEP2 and high TMSB15A expression associated with poor survival
Key statistics
  • count 129,079 new cases and 73,987 deaths in 2018 (global NPC incidence/mortality)
  • count 113,659 (85.5%) new cases in Asia in 2020 (NPC geographic distribution)
  • count 1558 DEGs (DEGs between two m6A subclasses)
  • count n = 255 (primary NPC clinical samples for IHC)
  • count 88 (NPC patients with PFS analyzed from GSE102349)
  • fold_change |log2 FC| > 1, FDR < 0.05 (DEG identification threshold (edgeR))
  • pvalue P < .05 (univariate Cox significance threshold for m6A regulators)
  • correlation correlation coefficient > 0.3 (Pearson correlation threshold between prognostic factors and m6A regulators)

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 applied consensus clustering to 88 RNA-seq NPC samples (GSE102349) using 10 prognostic m6A regulators selected by univariate Cox regression, identifying two molecular subclasses. Tumor microenvironment infiltration differences between subclasses were assessed with Wilcoxon rank-sum or Kruskal-Wallis tests on ssGSEA and ESTIMATE scores. A LASSO Cox prognostic signature was built on a 7:3 internal train/test split and evaluated via Kaplan-Meier/log-rank tests and time-dependent ROC; protein-level findings were confirmed in 255 clinical IHC samples with chi-square tests for categorical comparisons.

Replicationbiological Sample size88 NPC patients selected from 113 in GSE102349 (those with available PFS data); 36 samples from GSE53819 (18 tumor + 18 non-cancerous) for validation; 255 clinical IHC samples; no formal power calculation mentioned GroupsTwo m6A-related molecular subclasses (cluster 1, n=64 vs cluster 2, n=24); high-risk vs low-risk prognostic score groups; NPC tumor vs non-cancerous tissue (GSE53819) Pairingunpaired Randomization/blindingnot stated DispersionSD Effect sizesyes Multiplicity correctionFDR (Benjamini-Hochberg via edgeR) for DEGs; Bonferroni correction for SubMap immunotherapy prediction; adjusted p-value for GSVA pathway comparisons
Statistical tests used
Test Applied to n Assumptions
Univariate Cox proportional hazards regression Screening 21 m6A regulators for association with PFS to select features for clustering and prognostic modeling 88 NPC patients (GSE102349) not stated
Consensus clustering (consensusClusterPlus) Unsupervised subtype identification from expression profiles of 10 prognostic m6A regulators; k selected by CDF plateau 88 NPC patients (GSE102349) na
Wilcoxon rank-sum test Comparing GSVA pathway scores, ESTIMATE immune/stromal scores, and CIBERSORT immune cell proportions between the two m6A subclasses 88 NPC patients not stated
Kruskal-Wallis test Comparing GSVA pathway/TME scores between subclasses (reported as alternative to Wilcoxon rank-sum depending on context) 88 NPC patients not stated
edgeR (negative binomial GLM) Differential expression between two m6A subclasses; cutoffs |log2FC|>1 and FDR<0.05 88 NPC patients (GSE102349, RNA-seq) not stated
LASSO Cox regression with ten-fold cross-validation Feature selection and construction of m6A-related prognostic risk signature from DEGs Training set (~62 patients, 7:3 split of 88) not stated
Kaplan-Meier survival analysis with log-rank test PFS comparison between m6A subclasses and between high-risk vs low-risk groups in training, test, and whole sets; also IHC-based survival for REEP2/TMSB15A 88 NPC patients (GEO); 255 for IHC cohort (n for survival analysis not explicitly restated) not stated
Time-dependent ROC / AUC Evaluating predictive accuracy of the prognostic risk model across training, test, and whole sets 88 NPC patients na
Chi-square test Testing significance of differences in categorical clinical variables 255 primary NPC clinical samples not stated
Pearson correlation Correlation between prognostic signature genes and m6A regulators; threshold r>0.3 and p<0.05 88 NPC patients not stated
Spearman correlation Correlation between expression of REEP2, TMSB15A, DSEL, ID4 and other genes for GSEA group classification 88 NPC patients not stated
SubMap algorithm with Bonferroni correction Predicting and cross-dataset validation of anti-PD1 and anti-CTLA4 immunotherapy response between m6A subclasses 88 NPC patients (GSE102349); 18 NPC (GSE53819) na
Approaches that could also have been used
  • Univariate Cox regression screened 21 m6A regulators at p<0.05 without correction for simultaneous testing
    Could also: Apply FDR (Benjamini-Hochberg) correction to the 21 univariate Cox p-values before selecting regulators for downstream clustering — Testing 21 hypotheses simultaneously raises the expected number of false discoveries; FDR adjustment is the standard in multi-gene survival screening and is already employed at the DEG step of this same study, so applying it here would make the filtering criteria consistent throughout the pipeline
  • Prognostic signature performance was evaluated using a single internal 7:3 random train/test split of the 88-patient cohort
    Could also: Repeated k-fold cross-validation or bootstrap-based internal validation could also be used to assess prognostic model performance — With n=88 a single random split allocates only ~26 patients to the test set, making performance metrics sensitive to which patients happen to fall in that partition; repeated cross-validation or bootstrap uses all available patients for both training and evaluation, yielding more stable and less optimistic AUC and concordance estimates
  • Consensus clustering was used to identify the two molecular subclasses, with k chosen by the CDF plateau criterion
    Could also: Non-negative matrix factorization (NMF) is another widely used approach for identifying molecular subtypes from expression data — NMF and consensus clustering optimize different objective functions and can yield subtypes with different boundaries; showing concordance—or quantifying discordance—between methods is a common sensitivity check and can help establish whether the two-subtype solution is robust to algorithmic choice
  • Pearson correlation was used for gene-regulator associations while Spearman correlation was used separately for gene-gene associations in the GSEA step
    Could also: Spearman correlation could be used consistently for both correlation analyses throughout the paper — Spearman correlation makes no assumption of bivariate normality and is robust to outliers and non-linear monotone relationships; applying it uniformly would reduce assumption sensitivity and methodological inconsistency between analysis steps
  • Continuous clinical variables (age, tumor size) were summarized as mean ± SD
    Could also: Median with interquartile range (IQR) could also summarize continuous clinical variables — Clinical measurements such as age and tumor size are often right-skewed; median and IQR describe the distribution more robustly when normality is not verified, and are the format recommended by many clinical reporting guidelines (e.g., CONSORT, STROBE)
  • edgeR was used for differential expression between the two RNA-seq subclasses
    Could also: DESeq2 is another standard tool for RNA-seq differential expression analysis — DESeq2 and edgeR use different normalization strategies and dispersion estimation models; comparing DEG lists from both tools and reporting the overlap is a common practice for identifying the most reproducible findings, since genes called by both methods tend to be more robust
Software: R 3.4.4 · GraphPad Prism 8 · consensusClusterPlus (R package) · survival (R package) · GSVA (R package) · edgeR (R package) · glmnet (R package, LASSO) · survivalROC (R package) · rms (R package, nomogram) · clusterProfiler (R package) · pRRophetic (R package) · Cytoscape 3.9.1

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
11
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-38532632

Paper: Wang et al. 2024, M6A-mediated molecular patterns and tumor microenvironment infiltration characterization in nasopharyngeal carcinoma, Cancer Biol Ther. Repo: https://github.com/DrWangfeng/nasopharyngeal-carcinoma (authors' OWN R code; P16 N/A). Data: GSE102349 (113 NPC patients; 88 with PFS), GSE53819 (18 tumor + 18 normal). The repo SHIPS the processed data: GSE102349_counts_symbol.txt, ..._clinical_PFS.txt, ..._Cluster_res.txt (published labels), GSE53819_NPC_exprSet.txt, etc.

In scope (pipeline-derived, attempted)

id result method (pipeline) reported location
C1 # prognostic m6A regulators univariate Cox (survival pkg), p<0.05 10 of 21 Results, Fig 1g / Table S1
C2 m6A consensus clusters ConsensusClusterPlus (hc, euclidean, seed99, maxK6) 2 clusters: 64 / 24 Fig 1c
C3 DEGs between clusters edgeR, |log2FC|>1 & FDR<0.05 1558 (928 up, 630 down) Fig 6a,b / Table S12

Attempted-but-fragile (reported, not primary)

  • LASSO 4 hub genes (REEP2, TMSB15A, DSEL, ID4): chain = TPM → edgeR DEG → univ-Cox (caret seed 359) → cv.glmnet (seed 9). Depends on TPM which is not exactly regenerable (gene lengths lost on symbol collapse). We check only whether the 4 hub genes appear in our DEG set, not the full LASSO. Not graded as a primary claim.

Out of scope (not attempted)

  • Wet-lab / clinical validation cohort (255 NPC qPCR/IHC, Fig 8) — manual, not pipeline.
  • ssGSEA/CIBERSORT/ESTIMATE/GSVA/pRRophetic/TIDE/SubMap/GSEA/TF-prediction (Figs 2–5, 9–10): in principle pipeline-derived but mostly directional ("higher in cluster1") rather than a single checkable number; skipped under 80/20.
  • t-SNE plots (visual, non-numeric).

Key honest caveat

Authors' C1/C2 use TPM (GSE102349_TPM_symbol.txt, NOT shipped; count2TPM.R needs per-gene lengths from featurecounts, also not shipped). We substitute log2(CPM+1) from the shipped symbol-level counts as a monotone proxy. C3 (edgeR) uses raw counts directly and needs no TPM → cleanest data point. The edgeR DEG script itself (GSE102349_edgeR_diff.txt) is also not shipped; we implement it per the README spec.

Figures / tables: Fig 1gTableFig 1cFig 6aFig 7a
C1
Reported
10 prognostic m6A regulators (univariate Cox p<0.05)
Reproduced
4 (METTL3, YTHDF2, IGF2BP1, ELAVL1); 3 of 4 within the paper's set
partial
C2
Reported
2 m6A clusters, sizes 64 / 24 (Fig 1c)
Reproduced
K=2 optimal, sizes 63 / 25, 98.9% concordance vs authors' shipped labels
within tolerance
C3
Reported
1558 DEGs (928 up, 630 down) edgeR |log2FC|>1 & FDR<0.05 (Fig 6a,b)
Reproduced
1406 (filterByExpr, 863/543) and 1772 (no prefilter, 994/778) bracket 1558; same direction
within tolerance
C4
Reported
4 LASSO hub genes REEP2, TMSB15A, DSEL, ID4 (Fig 7)
Reproduced
all 4 recovered as DEGs (log2FC 1.26-2.33)
within tolerance

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 76/100

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: 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 🟡
Concordant (toward reproduced)
Code + data deposited & functional
-2 pts
From: Data & code availability Available & functional
Total score +3

This is a solid partial reproduction of the m6A-NPC study. The headline molecular subtyping reproduces almost 1:1 (consensus clustering 63/25 vs reported 64/24, 98.9% concordance, 1/88 relabeled), DEGs are bracketed (1406–1772 vs reported 1558, same direction), and all four LASSO hub genes reappear as DEGs. The one real deviation — 4 vs 10 prognostic m6A genes (C1) — sits on the input/normalization side: it is driven by a log2(CPM+1) proxy that was necessary because the authors did not deposit the (non-regenerable) TPM matrix, not by any computational error or fabrication. Net: deviations are moderate, explainable, and shared between our method choice and an authors'-side data-availability gap; the central conclusion holds.

🤝
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.

97 k
tokens (I/O) · 6.3 M incl. cache
12 min
runtime · 0.02 CPU-h
1.3 GB
peak RAM
1
HPC jobs
hummel
machine