Corpus 1,286 assessed · 1,187 scored · 648 reproduced ≥75 · 174 flagged ·∅ 73.9/100
← New search

IRSN-23 gene diagnosis enhances breast cancer subtype classification and predicts response to neoadjuvant chemotherapy: new validation analyses.

Breast Cancer · 2025
L1 71/100 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.

Concordant (toward reproduced)
All content-critical questions reproduced
-4 pts
From: Q7 · Core claim 🟢
Code + data deposited & functional
-2 pts
From: Data & code availability Available & functional
Every question reproduced
-1 pts
From: “every question reproduced”
Total score -7
✓ What held up
  • Same input data as the authors
  • Reported values were directly comparable
  • No relevant deviation in data/preprocessing
  • 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
  • Overall, the reproduction was clean
What did not (or only partly)
  • Every checked point held up.
How its reproducibility compares
71/100
Reproducibility score
at the mean
vs. all fields · 1187 studies
🎯 Scores higher than 38% of all assessed papers rank 699 of 1187 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 -> 1:1 reproduction. The paper's lead pipeline result (OUH-cohort pCR enrichment in IRSN-23 Gp-R vs Gp-NR without anti-HER2 = 29% vs 1%, P=1.70E-5) was reproduced EXACTLY from the authors' own shipped supclust DLDA model applied to their shipped OUH data (repo SNlaboratory/irsn23 @ db0f6b4), run on «our HPC» with R 4.5.3 + supclust 1.1-1. pCR rates 29%/1% match exactly; the P-value matches to the same order of magnitude (test-convention difference). Training/Validation reference tables also reproduced deterministically. NOT attempted (hard ~20%): the external GEO meta-analysis (N=1282, GSE25066 + 9 other series), pooled odds ratios 3.23-5.72 (Fig 1B), AUC/DeLong, GSEA, subtype heatmaps -- the repo ships only the fitted model + OUH data, not the per-GEO normalization/scoring pipeline, so that would require rebuilding the whole external pipeline. Repo's Validation.data.Robj is corrupted (bad magic 'X') but is not needed for the reproduced claims. No fabrication flags.

💻 Code ↗ 🗄 Data: GSE25066

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 71
    assessed: 2026-06-14 ⛓ e898702cc6d3
✎ 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
2026-09-19

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

The study tests whether the IRSN-23 immune-related gene signature reproducibly classifies breast cancer patients into chemotherapy-sensitive (Gp-R) and less-sensitive (Gp-NR) groups and predicts pathological complete response (pCR) to neoadjuvant chemotherapy, and how it impacts breast cancer subtype classification.

Core claims
  • IRSN-23 Gp-R patients have significantly higher pCR rates than Gp-NR patients without anti-HER2 therapy, across the OUH cohort and multiple independent public datasets finding
  • IRSN-23 predictive performance is reproducible across different microarray platforms (Affymetrix U133, X3P, Agilent, Illumina) finding
  • In patients receiving anti-HER2 therapy, pre-treatment IRSN-23 does not significantly predict pCR, but reassessment two weeks post-treatment does finding
  • IRSN-23 combined with Oncotype Dx or PAM50 refines breast cancer subtype classification based on tumor microenvironment (offensive factor PAM50, defensive factor IRSN-23) finding
  • The immune subtype identified via IRSN-23 is correlated with better prognosis after NAC finding
  • IRSN-23 is a 23-probe (19-gene) immune-related signature developed by DLDA supervised analysis, patented, and available on GitHub resource
  • The 23-probe IRSN-23 score correlates highly (r=0.99) with the 19-gene aggregated version used to adapt the signature to other platforms method
  • TMB and HRD scores derived from WES were compared against IRSN-23 as predictors of chemotherapy sensitivity method
Experimental setups
Assay System Perturbation Readout Platform
DNA microarray (gene expression) breast tumor tissue, OUH patients (n=146) NAC (paclitaxel-FEC) ± trastuzumab IRSN-23 immune score and pCR after NAC Affymetrix Human Genome U133 Plus 2.0 (GPL570)
DNA microarray (gene expression) breast tumor tissue, GSE25066 public dataset (n=488) NAC IRSN-23 classification and pCR rate GPL96/GPL570/GPL571
DNA microarray (gene expression) breast tumor tissue, public GEO datasets (GSE28844, GSE42822, GSE140494, GSE37946, GSE66399) NAC ± anti-HER2 therapy IRSN-23 classification and pCR rate GPL570
DNA microarray (gene expression) breast tumor tissue, GSE4779 (n=102) NAC IRSN-23 classification and pCR rate Affymetrix Human X3P Array (GPL1352)
DNA microarray (gene expression) breast tumor tissue, GSE21974, GSE34138, GSE130788 NAC ± anti-HER2 therapy IRSN-23 classification and pCR rate Agilent/Illumina (GPL6480, GPL6884)
RNA-Seq breast tumor tissue, OUH patients (n=43) none concordance of IRSN-23 score with microarray-derived score
Whole exome sequencing breast tumor tissue, OUH patients (n=81) none TMB and HRD score (NtAI, LST, HRD-LOH) SureSelect Human All Exon V6 (Agilent)
Immunohistochemistry / FISH breast tumor tissue, OUH patients none ER, PR, HER2 status, TILs, Foxp3+ Tregs, CD8+ T cells, IL17F+ cells
Key results
  • OUH cohort without trastuzumab: pCR rate higher in Gp-R than Gp-NR 29% vs 1%, P=1.70E-5
  • Pooled validation datasets without anti-HER2 therapy: pCR rate higher in Gp-R than Gp-NR N=1103, 40% vs 12%, P=2.02E-26
  • Pooled validation datasets with anti-HER2 therapy: pCR rate higher in Gp-R than Gp-NR N=304, 49% vs 35%, P=0.017
  • GSE4779 (Affymetrix X3P array): pCR rate higher in Gp-R than Gp-NR 54% vs 24%, P=3.53E-03
  • Agilent (GSE21974) and Illumina (GSE34138) datasets confirmed higher pCR in Gp-R than Gp-NR P=3.22E-03 and P=3.63E-05 respectively
  • GSE130788: pre-treatment IRSN-23 not significant for pCR, but post-treatment reassessment (2 weeks) was significant pre-treatment P=0.084 (ns); post-treatment P=0.025
  • 23-probe IRSN-23 score correlated strongly with 19-gene IS version in OUH internal validation r=0.99 (n=59)
  • Immune subtype identified via IRSN-23 associated with better prognosis after NAC
Key statistics
  • correlation r=0.99 (23-probe IRSN-23 score vs 19-gene IS, OUH internal validation)
  • pvalue P=1.70E-5 (OUH Gp-R vs Gp-NR pCR rate without trastuzumab (abstract))
  • pvalue P=2.35E-05 (OUH Gp-R vs Gp-NR pCR rate without anti-HER2 therapy (results section))
  • pvalue P=2.02E-26 (pooled validation without anti-HER2 therapy, Gp-R 40% vs Gp-NR 12% pCR)
  • pvalue P=0.017 (pooled validation with anti-HER2 therapy, Gp-R 49% vs Gp-NR 35% pCR)
  • pvalue P=8.39E-13 (GSE25066 Gp-R 38% vs Gp-NR 10% pCR)
  • count 1282 cases (1261 analyzed after exclusion) (external validation set from 8 GEO datasets)
  • pvalue P=3.53E-03 (GSE4779 (Affymetrix X3P array) Gp-R 54% vs Gp-NR 24% pCR)

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 validates the IRSN-23 immune gene signature for predicting pathological complete response (pCR) to neoadjuvant chemotherapy in breast cancer using an institutional dataset (n=146) and independent public GEO datasets (N=1282). Group comparisons (Gp-R vs Gp-NR) used chi-square or Fisher's exact tests across multiple individual and pooled datasets; Kaplan-Meier survival curves were compared with log-rank tests; and AUC comparisons employed the DeLong test. All tests were two-tailed at α=0.05, and results were reported primarily as pCR proportions with exact p-values using R version 3.6.3.

Replicationbiological Sample sizeexact patient counts stated per dataset; no formal power calculation or sample size justification reported GroupsGp-R (immune score > 0) versus Gp-NR (immune score ≤ 0) within each dataset and breast cancer subtype; secondary comparisons across PAM50/IHC subtypes for survival Pairingunpaired Randomization/blindingnot stated Dispersionnone Exact p-valuesyes Effect sizesno Confidence intervalsno Multiplicity correctionno
Statistical tests used
Test Applied to n Assumptions
chi-square test or Fisher's exact test (two-tailed) pCR rate comparisons between Gp-R and Gp-NR in each individual dataset and in pooled analyses varies by dataset: OUH without anti-HER2 n=112; GSE25066 n=488; GSE28844 n=32; GSE42822 n=66; GSE140494 n=91; GSE4779 n=102; GSE21974 n=32; GSE34138 n=178; pooled without anti-HER2 N=1103; pooled with anti-HER2 N=304 not stated
DeLong test comparison of area under the curve (AUC) values not stated
log-rank test comparison of Kaplan-Meier distant recurrence-free survival curves by breast cancer subtype not stated
Pearson's correlation coefficient concordance between 23-probe and 19-gene IRSN-23 scores (OUH internal validation n=59) and as the linkage method for unsupervised hierarchical clustering (heat maps) n=59 for probe/gene score concordance; n=43 for microarray vs RNA-Seq concordance not stated
Gene Set Enrichment Analysis (GSEA) via DAVID Functional Annotation Bioinformatics genes with principal component loadings greater than 0.10 in absolute value na
Approaches that could also have been used
  • pCR rates between Gp-R and Gp-NR were compared using chi-square or Fisher's exact tests, reporting only p-values
    Could also: logistic regression reporting odds ratios (or risk ratios) with 95% confidence intervals — odds ratios with CIs quantify both the magnitude and precision of the association rather than only its statistical significance, and are directly amenable to formal meta-analytic pooling across the contributing datasets
  • pooled analysis combined raw patient counts across independent datasets by simple aggregation into a single 2×2 table
    Could also: conduct a random-effects meta-analysis pooling log odds ratios across studies using inverse-variance weighting (e.g., DerSimonian-Laird) — a formal meta-analysis accounts for between-study heterogeneity, yields a pooled effect estimate with a CI, and allows assessment of heterogeneity (e.g., I²) across the contributing datasets, which differ in platform, chemotherapy regimen, and patient mix
  • numerous separate hypothesis tests were performed across multiple datasets and subgroups without a stated multiplicity adjustment
    Could also: apply Benjamini-Hochberg false discovery rate (FDR) correction or Bonferroni correction across the family of tests — when many comparisons are made, FDR or family-wise correction controls the expected proportion of false positives and helps readers calibrate which findings are most robust
  • Kaplan-Meier survival differences between subtypes were assessed with log-rank tests
    Could also: Cox proportional hazards regression reporting hazard ratios with 95% CIs — hazard ratios quantify the magnitude of the survival difference and permit adjustment for clinical covariates such as tumor stage, ER status, and treatment, which is informative given the heterogeneous patient populations
  • concordance between microarray-derived and RNA-Seq-derived IRSN-23 scores was assessed with Pearson's correlation (n=43)
    Could also: use intraclass correlation coefficient (ICC) or Bland-Altman analysis — ICC directly quantifies agreement rather than linear association; Bland-Altman plots additionally visualize systematic bias and limits of agreement across the measurement range, which is relevant when assessing whether the two platforms can be used interchangeably
  • pCR proportions were reported as point estimates without uncertainty bounds
    Could also: report 95% confidence intervals around each proportion (e.g., Wilson or Clopper-Pearson intervals) — CIs convey estimation uncertainty alongside the point estimate and are especially informative for small-n subgroups such as GSE28844 (n=32) or GSE21974 (n=32), where the point estimate alone may be imprecise
Software: R 3.6.3 · DAVID Functional Annotation Bioinformatics · R/genefu · R/supclust

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
2
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.

GPL6884 GEO in Results (http://purl.org/orb/Results)
also used by 1 paper:
GSE25066 GEO in Methods (http://purl.org/orb/Methods)
also used by 1 paper:
GPL1352 GEO in Results (http://purl.org/orb/Results)
no other assessed paper uses this yet
GPL571 GEO in Results (http://purl.org/orb/Results)
no other assessed paper uses this yet
GSE130788 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE140494 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE21974 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE28844 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE34138 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE37946 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE42822 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE4779 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE66399 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
NCT01042379 NCT in Discussion (http://purl.org/orb/Discussion)
no other assessed paper uses this yet

Downstream reach in the literature

100 downstream papers · 1 datasets

How 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.

What was reproduced

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

Scope — pmid-40128415 (IRSN-23 new validation analyses)

Paper: Sota Y et al. IRSN-23 gene diagnosis enhances breast cancer subtype classification and predicts response to neoadjuvant chemotherapy: new validation analyses. Breast Cancer (2025). PMID 40128415 · PMCID PMC11993443 · DOI 10.1007/s12282-025-01687-6.

Method (from Methods + repo): IRSN-23 = a 23-probe (19-gene) immune-related signature scored with DLDA (Diagonal Linear Discriminant Analysis) via the R package supclust. The pre-fit classifier and the OUH cohort data are shipped in the authors' repo https://github.com/SNlaboratory/irsn23 (pinned commit db0f6b4f2229464dfe07453696460153afd01d30). The repo's How to Use IRSN-23.Rmd loads the model + OUH.data and tabulates predicted group (Gp-R = responder / Gp-NR = non-responder) against the actual pCR result.

In scope (pipeline-derived, reproducible from shipped artifacts)

result pipeline source data status
OUH prospective cohort: pCR rate Gp-R vs Gp-NR, without anti-HER2 (paclitaxel-FEC) supclust DLDA IRSN.23.model$predict → 2×2 vs pCR OUH.data (shipped in repo) REPRODUCED
OUH Training confusion table (Sota 2014 Ann Oncol, re-shipped) same OUH.data$Training reproduced (deterministic)
OUH Validation confusion table (Sota 2014 Ann Oncol, re-shipped) same OUH.data$Validation reproduced (deterministic)
significance of OUH Gp-R vs Gp-NR pCR difference (P value) Fisher/chi-square on the 2×2 derived from above reproduced (within order of magnitude)

Out of scope / NOT attempted (the hard ~20%)

  • External GEO meta-analysis (N = 1282, GSE25066 + 9 other series), pooled odds ratios 3.23–5.72 (Fig 1B), AUC/DeLong comparisons, GSEA, subtype-class heatmaps. The repo ships only the fitted model + the OUH cohort; it does not ship the per-GEO download/normalization/probe-mapping pipeline used for the 10 external datasets. Reproducing those would require independently reconstructing the entire normalization + scoring pipeline for 10 microarray series across 3 platforms (GPL96/570/571, GPL1352, GPL6480, GPL6884) — that is the unspecified, high-effort 20% and is intentionally not attempted.
  • Wet-lab / clinico-pathological / patent items: out of scope (non-pipeline).

Headline reproduced claim: the OUH-cohort pCR enrichment in Gp-R (the paper's lead quantitative result, abstract + Results), computed end-to-end from the shipped model on the shipped data.

C1
Reported
29% vs 1% (OUH pCR rate Gp-R vs Gp-NR, without anti-HER2)
Reproduced
29% (12/42) vs 1% (1/70)
exact
C2
Reported
P = 1.70E-5
Reproduced
chi-square ~1.4E-5 ; Fisher two-sided ~2.4E-5
within tolerance
C3
Reported
OUH Training table (Sota 2014, re-shipped)
Reproduced
Gp-NR 30/2 ; Gp-R 10/16
partial
C4
Reported
OUH Validation table (Sota 2014, re-shipped)
Reproduced
Gp-NR 35/0 ; Gp-R 15/9
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 71/100

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.

Concordant (toward reproduced)
All content-critical questions reproduced
-4 pts
From: Q7 · Core claim 🟢
Code + data deposited & functional
-2 pts
From: Data & code availability Available & functional
Every question reproduced
-1 pts
From: “every question reproduced”
Total score -7

The paper's lead result — OUH-cohort pCR enrichment in IRSN-23 Gp-R vs Gp-NR without anti-HER2 (29% vs 1%, P=1.70E-5) — reproduces 1:1 from the authors' own shipped DLDA model applied to their shipped OUH data; pCR rates match exactly and significance holds (P~1.4E-5/2.4E-5, a pure test-convention difference). No deviation sits on the authors' side or in the core computation, and there are no fabrication flags. Caveats are scope/integrity only: the external GEO meta-analysis (Fig 1B, OR 3.23-5.72) was not attempted because its per-series pipeline isn't shipped, and Validation.data.Robj is corrupted — neither affects the cleanly reproduced headline.

🤝
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 [email protected].

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.

91.7 k
tokens (I/O) · 5.6 M incl. cache
11 min
runtime · 0.02 CPU-h
2.6 GB
peak RAM
3
HPC jobs
hummel
machine