Specific Marker Gene Analysis for Primary Central Nervous System Lymphoma Based on Methylation Difference and Development of Detection Primers.
The main results reproduced, with only marginal, non-material deviations.
Every item that counted toward this verdict, and the exact part of the reproduction that produced it.
- ✓Reported values are derivable from the shared data
- ✓The central claim held under reproduction
- 🟡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
- 🟡The deviation was non-trivial in magnitude
- 🟡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 headline result 1:1 on PUBLIC data, despite the paper's 'data on request' statement (its Methods cite public TCGA/GEO accessions). Reproduced the 450K branch for the paper's central named marker, RHEB, on the same TCGA HumanMethylation450 cohorts (DLBC=lymphoma/PCNSL, GBM+LGG=CNS, PCPG+COAD=non-CNS) via scikit-learn ROC, on «our HPC»/«infra» («job», COMPLETED 4m22s). RESULT: RHEB promoter is markedly hypermethylated in lymphoma (DLBC) vs other tumours; ROC AUC=0.930 gene-mean / 0.958 best probe cg22589161 -> the paper's ROC>0.8 claim REPRODUCES and is exceeded. The beta-pattern reproduces for two of three tiers (lymphoma>0.4 yes, non-CNS<0.15 yes at the promoter); the intermediate 'CNS 0.2-0.3' tier is weaker/probe-dependent in our data (CNS often <0.2 for top probes). No fabrication: the named result is independently reproducible from public data. NOT a clean match only in that (a) the paper's <0.15 holds at the promoter CpGs but NOT for the whole-gene 26-probe mean (gene-body dilution -> non-CNS ~0.31), and (b) the neat three-tier gradient is partly a CpG-selection effect. NOT ATTEMPTED (honest 80/20 skips): the RRBS/mHapSuite branch (14,867 sites; needs BAM alignment + mHap generation across many GEO bisulfite-seq series), the full t-test differential scan (26 sites), the 12-gene intersection, and the other 6 marker genes (FAM83B, B3GAT1-DT, PLOD2, DNMBP-AS1, GNG7, TEAD1). Wet-lab qPCR primers / plasma cohort are out of scope.
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 62assessed: 2026-06-14 ⛓ 9fbe1379639d
✎ 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: opusCan DNA methylation differences, identified by combining 450K microarray and RRBS analyses, yield a PCNSL-specific molecular marker (and detection primers) capable of distinguishing primary central nervous system lymphoma from other CNS diseases in patient plasma?
- ★ RHEB promoter hypermethylation is a PCNSL-specific biomarker distinguishing PCNSL from other CNS diseases. finding
- ★ A pair of methylation-specific primers targeting the RHEB promoter region was developed and can distinguish PCNSL patients via plasma-based detection. resource
- ★ Combining 450K microarray and RRBS analyses identified differential methylation sites, yielding 12 candidate marker genes (7 in promoter/CDS regions). method
- ★ RHEB is highly methylated in PCNSL tumors, and its promoter methylation suppresses RHEB expression, reversible by demethylation drug treatment. mechanism
- Candidate marker genes (RHEB, FAM83B, B3GAT1-DT, PLOD2, DNMBP-AS1, GNG7, TEAD1) showed ROC AUC values above 0.8 for classifying CNS versus PCNSL. finding
- Findings are preliminary, proof-of-concept, and require validation in larger cohorts. finding
| Assay | System | Perturbation | Readout | Platform |
|---|---|---|---|---|
| 450K methylation microarray (Illumina HumanMethylation450 BeadChip) analysis | Public datasets: DLBCL, GBM/LGG, PCPG, COAD (TCGA/NCI GDC) | none | differential CpG methylation beta values / sites | Illumina HumanMethylation450 BeadChip (485,512 probes) |
| RRBS (reduced-representation bisulfite sequencing) analysis | CNS tumors, normal tissue, DLBCL (CCLE and GEO datasets) | none | differential methylation sites (gene promoters, MM/PDR/entropy metrics) | mHapSuite / GEO & CCLE data |
| ROC curve analysis | TCGA and GEO datasets | none | AUC for CNS vs PCNSL classification | Python sklearn module |
| Kaplan-Meier survival analysis (log-rank test) | Published GDC cohort datasets | none | overall survival stratified by DNA methylation level | R v4.1.1 |
| Methylation-specific RT-qPCR | Plasma from clinical patients (PCNSL and other intracranial tumors) | none / methylation drug treatment | RHEB-U and RHEB-M methylation/expression levels | PerfectStart Green qPCR SuperMix; M5 First Strand cDNA kit |
| In vitro methylation and demethylation cell assay with qPCR | PCNSL cell lines (HKBML, DS, TK) and CNS cell lines (SF-268, SNB-19, U251) | demethylation drug (decitabine 5 uM, 24 h) / MethylCode bisulfite methylation kit | RHEB CDS (mRHEB), RHEB-U, RHEB-M expression/methylation | MethylCode bisulfite conversion kit (Thermo Fisher); decitabine (MCE) |
- ▲ 450K analysis: 16,043 differential sites in PCNSL vs CNS and 10,538 in CNS vs non-CNS, with 26 overlapping differential methylation sites; these 26 sites had higher methylation in PCNSL. 26 overlapping sites
- – RRBS analysis identified 14,867 common differential methylation sites between the two comparison groups. 14,867 sites
- ▲ Intersection of 450K (26) and RRBS (14,867) yielded 12 shared marker genes; 7 associated with promoter/CDS, showing higher methylation in PCNSL than other CNS groups. 12 genes; 7 markers
- – ROC analysis of marker genes (TCGA and GEO) gave AUC values above 0.8, with RHEB promoter methylation showing the highest diagnostic performance. ROC/AUC > 0.8
- ▼ In plasma qPCR, RHEB-U expression was significantly downregulated in PCNSL vs CNS patients, indicating RHEB promoter is highly methylated in PCNSL.
- – In HKBML (PCNSL) cells after demethylation, mRHEB and RHEB-U increased significantly while RHEB-M decreased, confirming methylation suppresses RHEB expression.
- – In U251 (CNS) cells, mRHEB and RHEB-U did not increase after demethylation, but methylation treatment raised RHEB-M and lowered RHEB-U.
- – Only one CpG island exists in the RHEB promoter region; only RHEB among candidate sites was located in a promoter region (others in CDS). 1 CpG island
- count 26 overlapping differential methylation sites (450K PCNSL vs CNS (16,043) intersect CNS vs non-CNS (10,538))
- count 14,867 common differential methylation sites (RRBS overlap between PCNSL vs CNS and CNS vs non-CNS)
- count 12 shared marker genes (intersection of 450K and RRBS differential sites)
- correlation ROC/AUC > 0.8 (CNS vs PCNSL classification, TCGA and GEO data)
- pvalue p-value cutoff 5% plus absolute MM change of 0.1 (significance threshold for differential methylation (Student's t-test))
- other beta <0.15 in non-CNS, 0.2-0.3 in CNS, >0.4 in PCNSL (CpG marker selection criteria)
- count 22 clinical samples (PCNSL (n=5) and other intracranial CNS tumors (n=17) for plasma primer validation)
- count RRBS CNS vs non-CNS: 14,471 hyper-, 8094 hypo-methylated; PCNSL vs CNS: 17,104 hyper-, 8665 hypo-methylated (volcano plot differential site counts)
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 study combined 450K methylation array data (TCGA) and RRBS data (CCLE/GEO) to identify differentially methylated CpG sites distinguishing PCNSL from CNS and non-CNS tumors; Student's t-test was applied for differential methylation analysis, with β-value thresholding for candidate selection. Marker performance was evaluated by ROC/AUC analysis and Kaplan–Meier survival curves with log-rank testing on publicly available datasets. Clinical validation used qPCR on plasma from 22 patients (5 PCNSL, 17 other CNS), and in vitro sensitivity was assessed through methylation/demethylation drug treatment of PCNSL and CNS cell lines.
| Test | Applied to | n | Assumptions |
|---|---|---|---|
| Two-sample Student's t-test | Differential methylation analysis of gene promoters; 450K chip and RRBS data (PCNSL vs CNS; CNS vs non-CNS) | 450K: DLBCL N=48, GBM/LGG N=685, PCPG N=187, COAD N=100; RRBS n per group not individually stated | not stated |
| ROC curve / AUC (with 95% CI) | Marker gene classification performance for PCNSL vs CNS (Figures 4A and 4B; TCGA and GEO datasets) | Database-level; exact per-analysis n not stated | na |
| Log-rank test | Kaplan–Meier overall survival comparison between high- and low-methylation groups for seven marker genes (Figure 4C) | GDC portal queue data; exact n not stated | not stated |
| qPCR relative expression comparison (test not specified) | RHEB-M and RHEB-U primer qPCR in plasma of PCNSL (n=5) vs other CNS tumors (n=17); Figures 5B and 5C | N=22 (5 PCNSL, 17 CNS) | not stated |
| qPCR relative expression comparison before/after drug treatment (test not specified) | mRHEB, RHEB-U, RHEB-M primer detection in HKBML and U251 cell lines pre/post methylation and demethylation treatment (Figures 6B and 6C) | Cell line experiments; number of replicates not stated | not stated |
-
Differential methylation across thousands of CpG sites was assessed with Student's t-test at p < 0.05, without a multiple-testing correction↳ Could also: Benjamini–Hochberg false discovery rate (FDR) correction, or the Storey q-value method, applied across the full set of tested sites — When thousands of sites are tested simultaneously, controlling the FDR is a widely adopted approach to quantify and limit the expected proportion of false positives among declared hits; reporting FDR-adjusted p-values alongside the unadjusted threshold would make the false-positive rate explicit
-
Student's t-test was used directly on β-values from 450K array data to compare methylation between groups↳ Could also: Limma with M-value transformation, or a dedicated methylation analysis package such as minfi or methylKit — β-values are bounded [0,1] and often bimodally distributed, which can violate t-test assumptions; M-values (logit-transformed β) are approximately normally distributed and are the recommended input for linear-model frameworks such as limma in methylation array workflows
-
The clinical validation comparison of RHEB methylation levels between PCNSL (n=5) and other CNS tumors (n=17) used an unspecified test↳ Could also: Mann–Whitney U (Wilcoxon rank-sum) test — With a small, unbalanced sample (n=5 vs n=17), distributional assumptions of parametric tests are difficult to assess; a non-parametric rank-based test makes no normality assumption and is commonly preferred in small-cohort clinical biomarker studies
-
Kaplan–Meier curves were compared with the log-rank test, stratified by high vs low methylation level↳ Could also: Cox proportional hazards regression — Cox regression can adjust for covariates such as age, performance status, or treatment, which is particularly relevant in PCNSL where these factors influence survival; it also quantifies the association as a hazard ratio with confidence interval, providing an effect-size estimate alongside the significance test
-
Seven candidate marker genes were each evaluated individually with separate ROC analyses↳ Could also: A multi-marker logistic regression model or LASSO-penalized regression combining several methylation features, with AUC evaluated via cross-validation — Combining multiple markers in a single predictive model can capture complementary information; cross-validated AUC from a multi-marker panel is a standard approach for assessing whether a panel outperforms any individual marker
-
Results for the qPCR cell-line experiments (Figures 6B and 6C) were presented descriptively without a stated statistical test or reported replicates↳ Could also: A paired t-test or Wilcoxon signed-rank test on repeated biological replicates, with the number of replicates and a measure of variability (e.g., SD or SEM) stated explicitly — Reporting the number of independent biological replicates, a dispersion measure, and a formal test for the before-vs-after drug treatment comparison would allow readers to assess reproducibility and the magnitude of the observed changes
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 — pmid-41097902
Paper: Yan F, Wang Y, Fan X, Wei P, Shan Y. Specific Marker Gene Analysis for Primary Central Nervous System Lymphoma Based on Methylation Difference and Development of Detection Primers. Brain Behav 2025. PMID 41097902 · PMCID PMC12528802 · DOI 10.1002/brb3.70996.
Pipeline branches in the paper
| # | Reported result | Pipeline / tool | Public data? | In scope? |
|---|---|---|---|---|
| A | 450K differential methylation across PCNSL/CNS/non-CNS → 26 overlapping sites; RHEB + 6 other marker genes; β pattern (non-CNS <0.15, CNS 0.2–0.3, PCNSL >0.4); ROC > 0.8 | Illumina HM450 β-values, Student t-test, GenomicFeatures overlap, sklearn ROC | Yes — TCGA HM450 (DLBC/GBM/LGG/PCPG/COAD) public via GDC / UCSC Xena | YES (primary) |
| B | RRBS methylation analysis → 14,867 overlapping sites; mHapSuite metrics (MM, PDR, CHALM, MCR, MBS, MHL, entropy, LD R²) on bisulfite-seq (GSE70175 etc.) | mHapSuite (github.com/yoyoong/mHapSuite) on RRBS/WGBS BAMs | Partly — GEO bisulfite-seq series public, but needs alignment + mHap generation | Partial / optional (hard 20%) |
| C | GO/KEGG enrichment (ClusterProfiler) of marker genes | ClusterProfiler | derived from A | optional |
| D | Kaplan–Meier survival of RHEB | KM/log-rank, R 4.1.1 | TCGA clinical | optional |
| E | qPCR methylation-specific primers (Table 1), cell-line screening (HKBML/DS/TK…), 22 plasma clinical samples | Wet-lab | n/a | OUT of scope (not a pipeline) |
Decision (80/20)
Reproduce branch A for the headline gene RHEB on the same public TCGA HM450 data the paper used: per-cohort RHEB promoter β means (DLBC=lymphoma/PCNSL, GBM+LGG=CNS, PCPG+COAD=non-CNS) and ROC (lymphoma vs rest) via scikit-learn — exactly the paper's described method. This is the clearest, fully-public, low-hanging numeric claim.
- Branch B (mHapSuite/RRBS) is the hard ~20%: it needs per-sample BAM alignment and mHap generation across many GEO bisulfite-seq series; not attempted in full (would not change the headline conclusion about RHEB; flagged honestly).
- Branch E is wet-lab, out of scope by definition.
- Data-availability note: the paper's statement says data is "from the corresponding author on reasonable request", yet Methods cite public GEO/TCGA accessions — so the computational inputs ARE publicly obtainable. We reproduce on those public inputs.
Third-party-tool note (P16)
The cited repo (mHapSuite) is a generic third-party methylation tool, not the authors' bespoke code. Per the brief, running an existing tool / standard pipeline on the paper's own public data is an equally valid reproduction.
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 named result — RHEB promoter hypermethylation distinguishing PCNSL/lymphoma with ROC>0.8 — reproduces and is exceeded (AUC 0.930 gene-mean, 0.958 at cg22589161) from public TCGA HM450 despite the 'data on request' notice, so there is no fabrication concern. Deviations are minor and lie on the input/selection side: cohorts were redefined from TCGA labels (n/sample list not deposited), and the paper's intermediate 'CNS 0.2–0.3' tier is optimistic (CNS often <0.2) — a CpG-selection effect, not an output-logic failure. The broader claims (RRBS 14,867 sites, 26 sites, 12-gene intersection, 6 other genes) were not attempted, so the reproduction is solid but partial rather than a clean 1:1.
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.