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

FRMD8 inhibits tumor metastasis in BRCA1-associated TNBC by negatively regulating tmTNF-α.

Cell Mol Biol Lett · 2025
L1 73/100 3/4
Why this verdict

The main results reproduced, with only marginal, non-material deviations.

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: Q7 · Core claim 🟡
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 🟡
Minor / cosmetic deviation
+1 pts
From: Q2 · Endpoint comparability 🟡
Concordant (toward reproduced)
Code + data deposited & functional
-2 pts
From: Data & code availability Available & functional
Total score +6
✓ What held up
  • Same input data as the authors
  • Reported values are derivable from the shared data
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
  • 🟡The deviation was non-trivial in magnitude
  • 🟡The central claim did not (fully) hold under reproduction
  • 🟡Overall, the reproduction showed a material discrepancy
How its reproducibility compares
73/100
Reproducibility score
at the mean
vs. all fields · 1173 studies
🎯 Scores higher than 41% of all assessed papers rank 664 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 1:1. The paper's pipeline-derived survival claim (Fig 1J: FRMD8 protective for MFS in 4 of 7 public GEO microarray datasets) was re-run on the exact 7 public datasets with the paper's stated method (quantile-normalize, mean-split FRMD8 high/low, survfit/logrank/Cox) on «our HPC». Result MATCHES the reported 4/7 when FRMD8's multiple probes are averaged, and is 3/7 with the single canonical probe 210043_at (the paper doesn't specify probe handling; both reported). Direction reproduced. KEY AUDIT FINDING (no fabrication, soft overstatement): in a faithful re-run only 1 of 7 KM comparisons (GSE5327, p=0.049) is statistically significant; the other protective calls are non-significant directional trends, so the prognostic evidence is weaker than the framing implies. NOT attempted (the optional 20%): DFS panel Fig 1K (5/7, needs different survival-column curation + TCGA-TNBC via GDC), OS panel Fig 1I, the bulk RNA-seq of FRMD8 KO/OE cells (no public FASTQ accession located -> no 1:1 number), and all single-cell/spatial/wet-lab/AlphaFold3 results (out of scope, non-pipeline). The auto-mined code_url (griffithlab/rnaseq_tutorial) is a third-party tutorial the paper's RNA-seq follows; per BRIEF P16 reproducing the described pipeline on the paper's public data is valid.

💻 Code ↗ 🗄 Data: GSE5327

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 73
    assessed: 2026-06-14 ⛓ f493a427309d
✎ 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
2026-06-14
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: sonnet
Founding hypothesis

Low expression of FRMD8 in BRCA1-mutant TNBC enhances tumor metastatic potential by promoting surface tmTNF-α (via impaired iRHOM2 degradation and reduced ADAM17-mediated cleavage), and this axis can be therapeutically targeted.

Core claims
  • Low FRMD8 expression in BRCA1-mutant breast cancer cells significantly enhances metastatic potential to various organs finding
  • FRMD8 low inhibits cleavage of tmTNF-α and promotes surface tmTNF-α expression finding
  • FRMD8 regulates tmTNF-α by inhibiting iRHOM2 degradation, acting mainly through the endocytic pathway mechanism
  • FRMD8 low/iRHOM2 low greatly facilitates in vivo metastasis of TNBC finding
  • Combined paclitaxel and etanercept treatment reverses FRMD8 and iRHOM2 expression and inhibits metastatic potential in vivo finding
  • FRMD8 anchors to the intracellular N-terminus of iRHOM2, forming a ternary complex with ADAM17 that enhances ADAM17 stability and sheddase activity mechanism
  • CRISPR-Cas9 sgRNA library screening of 39 candidate genes in a metastasis-poor cell line identifies FRMD8 as a metastasis-suppressor gene method
  • Integration of hospital RNA-seq and public GEO/TCGA datasets identifies 39 genes lowly expressed in metastatic primary tumors and associated with survival method
Experimental setups
Assay System Perturbation Readout Platform
bulk RNA-seq human TNBC tumor tissue (7 metastatic vs 7 non-metastatic; 3 FRMD8-low vs 3 FRMD8-high FFPE samples) none differentially expressed genes HISAT2, edgeR, htseq-count, stringtie
public dataset meta-analysis multiple GEO datasets (GSE5327, GSE1456, GSE4922, etc.) and TCGA breast cancer cohorts none survival-associated gene expression, hazard ratio of FRMD8 GEOquery, preprocessCore, coxph, survfit
single-cell RNA-seq and spatial transcriptomics breast cancer tissue (GSE161529, GSE180286, GSE176078 scRNA-seq; GSE210616 spatial) none cell type annotation, ligand-receptor cell-cell communication, spatial DE genes Seurat, Harmony, SingleR, CellChat, SpatialDE, clusterProfiler
CRISPR-Cas9 sgRNA library screening 759 mouse TNBC cell line injected into nude mice mammary glands knockout of 39 candidate genes presence of organ-specific sgRNAs (lung, spleen, liver, kidney, brain) by PCR PCR with sgRNA-F1/library-R1 primers
in vivo metastasis assay 759/642 mouse TNBC cells and human MCF-7, BT-549, MDA-MB-231, MDA-MB-436 cell lines in nude mice FRMD8 knockdown (sgRNA1-3) GFP fluorescence radiance in metastatic organs IVIS Spectrum, Living Image software
in vivo drug treatment 759 Frmd8 sgRNA1 + iRhom2 shRNA cells in nude mice paclitaxel, etanercept, or combination primary tumor volume and organ metastasis IVIS Spectrum
IHC and H&E staining mouse and human breast tumor tissue none target protein expression intensity (DAB integrated optical density) Pannoramic MIDI scanner, Python scikit-image
immunofluorescence cultured breast cancer cells and tissue sections (FRMD8/iRHOM2 constructs) FRMD8 wild-type/Δ150-200, Flag-iRHOM2-Δ overexpression, iRhom2 knockdown/overexpression fluorescence intensity and red-green colocalization (overlap coefficient K1) Zeiss LSM780 microscope, Python (NumPy, scikit-image, Matplotlib)
Key results
  • Low FRMD8 expression significantly enhances metastatic potential of BRCA1-mutant TNBC cells to various organs
  • FRMD8 low increases cell surface tmTNF-α expression across breast cancer subtypes and inhibits its cleavage
  • FRMD8 protects iRHOM2 from endocytic degradation, thereby regulating ADAM17 cleavage activity
  • FRMD8 low/iRHOM2 low combination substantially enhances in vivo TNBC metastasis
  • Paclitaxel + etanercept combination treatment reverses FRMD8/iRHOM2 expression and reduces metastatic potential in vivo
  • 895 genes were significantly downregulated in metastatic vs non-metastatic TNBC hospital samples |log2FC| > 6, p < 0.05
  • 718 genes identified as significantly associated with survival across public datasets p < 0.05
  • Intersection of hospital and public dataset gene lists yielded 39 library genes lowly expressed in metastatic tumors and linked to survival 39 genes
Key statistics
  • fold_change |log2FC| > 6 (differential expression cutoff between 5 metastatic and 5 non-metastatic TNBC hospital samples)
  • pvalue p < 0.05 (significance threshold for 895 downregulated DEGs in hospital RNA-seq analysis)
  • pvalue p < 0.05 (significance threshold for 718 survival-associated genes from public dataset survival models)
  • count 39 genes (intersection of hospital DEGs and public survival-associated genes used as sgRNA library)
  • count 6 sgRNAs per gene (sgRNA library design for CRISPR-Cas9 screening)
  • other MOI 0.1 (lentiviral library infection of 759 cells)
  • count 1 × 10^6 cells per mouse (mammary injection dose for sgRNA library screening in nude mice)
  • other 5 mg/kg paclitaxel, 2 mg/kg etanercept, every 2 weeks (in vivo drug treatment dosing regimen)

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.

The study combined hospital FFPE RNA-seq (edgeR differential expression, n=5+5 TNBC patients) with Kaplan-Meier/log-rank survival screening across 15 public GEO datasets and TCGA, then intersected the resulting gene lists to nominate 39 CRISPR library targets validated in a mouse mammary injection model. FRMD8 prognostic value was quantified by Cox proportional hazards meta-analysis across public cohorts. Single-cell and spatial transcriptomic data were processed with Seurat/Harmony/CellChat pipelines; IHC and immunofluorescence signals were quantified computationally via Python. Statistical tests applied to in vitro cell-line and in vivo treatment-group comparisons are not described in the provided text, which ends mid-Methods.

Replicationbiological Sample size7 metastatic + 7 non-metastatic patient tumors collected; 5+5 used for edgeR DEG analysis; 3+3 FFPE samples used for FRMD8-high/low RNA-seq at USTC; 10 nude mice for sgRNA library screening; drug treatment group sizes not stated in provided text GroupsTNBC metastatic vs non-metastatic patients; FRMD8-high vs FRMD8-low expressors; FRMD8/iRHOM2 knockdown vs control cell lines (759, MCF-7, BT-549, MDA-MB-231, MDA-MB-436); four drug treatment groups in mouse model (CON, paclitaxel, etanercept, combination) Pairingunpaired Randomization/blindingnot stated Dispersionunclear Effect sizesyes Multiplicity correctionnone stated
Statistical tests used
Test Applied to n Assumptions
edgeR negative binomial differential expression (raw p < 0.05, |log2FC| > 6) Hospital FFPE RNA-seq: TNBC with metastasis vs without metastasis 5 metastatic vs 5 non-metastatic TNBC patients not stated
Kaplan-Meier survival analysis with log-rank test (R surv_pvalue function), p < 0.05 Survival-related gene screening across 15 public GEO datasets and TCGA not stated
Cox proportional hazards regression (R coxph function) for hazard ratio estimation; forest plot Meta-analysis of FRMD8 prognostic value across multiple public datasets not stated
Integrated optical density (IOD) via color deconvolution (HED model, Python/scikit-image) IHC DAB staining intensity quantification across tissue sections na
Fluorescence intensity ratio (channel sum / DAPI sum) and overlap coefficient K1 via Otsu thresholding (Python/scikit-image) Immunofluorescence expression normalization and red/green colocalization analysis na
IVIS radiance quantification (photons/s/cm2/sr) via ROI analysis in Living Image software GFP-based organ metastasis quantification in mouse experiments na
Approaches that could also have been used
  • The hospital RNA-seq DEG screen applied a raw p < 0.05 threshold via edgeR without FDR correction, yielding 895 downregulated genes from a transcriptome-wide comparison
    Could also: Benjamini-Hochberg FDR correction on the edgeR output (which computes adjusted p-values by default) could also be applied as the primary filter — FDR adjustment is standard practice for transcriptome-wide differential expression to bound the expected false-discovery proportion; edgeR produces FDR values natively and they are widely reported in RNA-seq pipelines as a complement to or replacement for raw p-values
  • Survival-related genes in public datasets were identified by applying a raw p < 0.05 log-rank threshold across many genes and multiple datasets with no stated multiple-testing correction
    Could also: A Benjamini-Hochberg FDR correction within each dataset, or a stringent Bonferroni threshold, could also be applied before intersecting with the DEG list — Screening many genes for survival association simultaneously inflates the number of expected false positives; FDR or family-wise correction would reduce this inflation and may sharpen the specificity of the candidate list entering the intersection step
  • FRMD8 prognostic value was meta-analysed by pooling Cox HR estimates across heterogeneous public GEO cohorts, visualised as a forest plot
    Could also: A random-effects meta-analysis model (e.g., DerSimonian-Laird) with explicit between-study heterogeneity statistics (I², Cochran Q) could also be reported — Public GEO cohorts vary in treatment, platform, and patient population; a random-effects framework quantifies between-study variance and produces more conservative pooled estimates when heterogeneity is present, which is often the case across independently collected breast cancer cohorts
  • IHC signal was quantified as total absolute IOD summed across the whole image without background subtraction
    Could also: Mean optical density per unit tissue area after background subtraction, or a semi-quantitative H-score (proportion of cells × staining intensity), could also be used — Absolute whole-image IOD is influenced by section area and tissue density; area-normalised or background-corrected measures improve comparability across specimens with differing section sizes or cellularity
  • Red/green immunofluorescence colocalization was quantified using the overlap coefficient K1 after Otsu binary thresholding
    Could also: Pearson's or Spearman's pixel-intensity correlation coefficients (e.g., via Coloc2/Fiji), or Manders' overlap coefficients M1/M2, could also be used — K1 is sensitive to the chosen intensity threshold; correlation-based or Manders' coefficients computed over the full pixel dynamic range reduce threshold dependence and align with widely used colocalization reporting conventions in cell biology
  • The FFPE RNA-seq comparison of FRMD8-high vs FRMD8-low breast cancers was based on n=3 per group selected from ten available samples
    Could also: A pre-study power calculation for RNA-seq differential expression (e.g., via the RNASeqPower or pwr R packages) could also be reported to contextualise the detectable effect size given this sample size — With n=3 per group, statistical power to detect moderate fold-changes is limited; a power statement would help readers interpret the completeness of the resulting DEG list and the likelihood of false negatives
Software: edgeR (R) · HISAT2 · FastQC · flexbar · samtools · stringtie · htseq-count · IGV (Integrative Genomics Viewer) · R/GEOquery + preprocessCore (normalize.quantiles) · R/survival (survfit, surv_pvalue, coxph) · R/ggplot2 · Seurat (R) · Harmony (R) · SingleR (R) · CellChat (R) · SpatialDE · clusterProfiler (R) · Python 3.8 / NumPy, scikit-image, Matplotlib 3.8 · Living Image (PerkinElmer IVIS)

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
1
Impact: low
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.

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.

21334 RRID in Article (http://semanticscience.org/resource/SIO_001029)
no other assessed paper uses this yet
72263 RRID in Article (http://semanticscience.org/resource/SIO_001029)
no other assessed paper uses this yet

What was reproduced

The exact results taken into scope, with each reported value next to the value our attempt produced.

Scope — pmid-40619383

Title: FRMD8 inhibits tumor metastasis in BRCA1-associated TNBC by negatively regulating tmTNF-α. (Xu et al., Cell Mol Biol Lett 2025; DOI 10.1186/s11658-025-00754-2)

Note on the auto-mined metadata

The scaffold's code_url = github.com/griffithlab/rnaseq_tutorial and data_accession = GSE5327 are text-mining artifacts, but they are not wrong:

  • The paper's bulk RNA-seq methods literally follow the griffithlab RNA-seq tutorial stack (FastQC, flexbar, HISAT2, samtools, StringTie, htseq-count, edgeR, IGV). The repo is a generic third-party tutorial, not the authors' own code — per BRIEF rule P16 this is still a valid reproduction target.
  • GSE5327 (Wang et al. 2005, 76 ER−/LN− breast tumours, GPL96) is one of ~14 public GEO microarray datasets the paper mines for FRMD8 survival association (it appears in the Fig 1J metastasis-free-survival panel).

Computational vs wet-lab

The paper is wet-lab heavy (cell lines, mouse metastasis models, luciferase reporters, IHC, co-IP, AlphaFold3 structure). Those are out of scope (not a reproducible pipeline from shipped data).

IN SCOPE (pipeline-derived, public data, fully specified method)

FRMD8 expression ↔ patient survival across public GEO microarray datasets (Figure 1 panels I/J/K). Method as printed:

  • normalize each dataset with preprocessCore::normalize.quantiles;
  • split samples into FRMD8-high / FRMD8-low by mean FRMD8 expression;
  • survival::survfit + survminer::surv_pvalue for the KM comparison.
  • Endpoints: OS (Fig 1I), MFS (Fig 1J), DFS (Fig 1K).

Reproducible claims (the only quantitative statements the text pins down):

  • MFS: "FRMD8 was identified as a protective factor in four of the seven analyzed datasets" — Fig 1J datasets: GSE103091, GSE25055, GSE25065, GSE6532, GSE69031, GSE9195, GSE5327.
  • DFS: "FRMD8 acted as a protective factor for DFS in five out of seven" — Fig 1K datasets: GSE103091, GSE1456, GSE26304, GSE48390, GSE71258, GSE86166, TCGA-TNBC.

Primary target = MFS panel (4/7) because all 7 datasets are public GEO series with downloadable survival metadata. DFS panel attempted if cheap (TCGA-TNBC needs a different fetch path → likely the optional 20%).

OUT OF SCOPE / not attempted (with reason)

  • Bulk RNA-seq of FRMD8 KO/OE cells (Fig 3/4) — raw FASTQ not located in a public accession from the text; the tutorial pipeline is reproducible but without the paper's own reads there is no 1:1 number to compare.
  • Single-cell (GSE161529/180286/176078) & spatial (GSE210616) re-analysis — large, multi-step, exact reported values not pinned → hard 20%, skipped.
  • All wet-lab assays, AlphaFold3 structure — non-pipeline.

Exact-value caveat (fabrication-relevant)

The manuscript text gives counts ("4 of 7", "5 of 7") but not the per-dataset p-values / hazard ratios (those live only inside the figure images). So the reproduction checks the qualitative claim (how many datasets show FRMD8 as a protective factor) and records each per-dataset HR/p we compute as provisional evidence for a human auditor.

Figures / tables: Fig 1J
MFS-count
Reported
FRMD8 protective for metastasis-free survival in 4 of 7 GEO datasets (Fig 1J)
Reproduced
4/7 protective (HR<1) when averaging all FRMD8 probes; 3/7 with single canonical probe 210043_at; protective DIRECTION reproduced either way, but only 1/7 (GSE5327) significant at p<0.05
within tolerance
GSE5327-MFS
Reported
FRMD8-high better metastasis-free survival (Fig 1J KM)
Reproduced
HR(high vs low)=0.243 [0.052-1.124], logrank p=0.0494, n=58, events=11
within tolerance
GSE9195-MFS-probe-sensitivity
Reported
counted among the 4 protective MFS datasets
Reproduced
single-probe 210043_at HR=2.345 (NOT protective) vs probe-averaged HR=0.427 (protective) - this dataset is what flips the count 3/7<->4/7
partial

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 73/100

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: Q7 · Core claim 🟡
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 🟡
Minor / cosmetic deviation
+1 pts
From: Q2 · Endpoint comparability 🟡
Concordant (toward reproduced)
Code + data deposited & functional
-2 pts
From: Data & code availability Available & functional
Total score +6

The headline Fig 1J claim — FRMD8 protective for MFS in 4 of 7 public GEO datasets — is reproducible 1:1 from the exact public data using the paper's stated pipeline when FRMD8's probes are averaged (3/7 with single probe 210043_at; GSE9195 is the swing dataset). No fabrication: the count is derivable. The deviations are on our-method / authors-underspecified side (unstated probe-collapse rule, self-curated MFS-event columns), not data unavailability. The main caveat is a soft authors-side overstatement: only GSE5327 (p=0.049) is statistically significant, so the protective conclusion holds only directionally and the prognostic evidence is weaker than the framing implies.

🤝
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.

167.5 k
tokens (I/O) · 10.6 M incl. cache
24 min
runtime · 0.08 CPU-h
3.2 GB
peak RAM
4
HPC jobs
hummel
machine