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

CoINcIDE: A framework for discovery of patient subtypes across multiple datasets.

Genome Med · 2016
L1 87/100 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: 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 🟡
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
  • No authors-side cause for any deviation
  • Reported values are derivable from the shared data
  • Any deviation was negligible
What did not (or only partly)
  • 🟡Could not use the authors’ exact input data
  • 🟡A deviation arose in the data or preprocessing
  • 🟡The central claim did not (fully) hold under reproduction
  • 🟡Overall, the reproduction showed a material discrepancy
How its reproducibility compares
87/100
Reproducibility score
0.7 SD above mean
vs. all fields · 1173 studies
🎯 Scores higher than 72% of all assessed papers rank 301 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 the data-processing pipeline 1:1. CoINcIDE (Planey & Gevaert, Genome Med 2016) ships its method as the R/Bioconductor pkg Coincide (github.com/kplaney/CoINcIDE @5845d379) plus dissertation analysis scripts, with breast data in the author's curatedBreastData Bioconductor pkg. Built a conda R env on «our HPC» and reproduced THREE clear pipeline outputs: (C1) the package's own shipped unit test for the gene-expression NA-filter+KNN-impute step reproduced EXACTLY (99 genes, 29 patients; imputation error 0.00503 vs shipped 0.00517, within-tol because the error-estimation sample() is unseeded). (C2) the paper's headline '17 datasets' — mergeDatasetList(minNumGenes=10000,minNumPatients=40) on curatedBreastData printed 'Found 17 passing the input thresholds' -> 17 EXACT, and the kept GEO study IDs match Table 1. (C3) total samples 2237 vs reported 2235 (within-tol, Δ=2/0.09%; curatedBreastData is now v2.30.0/2024 vs the 2016 curation). NOT attempted (hard 20%): C4, the full meta-cluster discovery (PAM50 centroid clustering -> CoINcIDE computeEdges with 500 null permutations -> community detection -> '5 breast meta-clusters'), and the ovarian analyses, GSEA/druggable-gene/survival meta-features (the Broad MSigDB gene-list RData is NOT in the repo -> data_unavailable for those), and the simulation figures. Honesty note: the author's ExpressionSet wrapper procExprSet() crashes on modern Biobase (dim-changing S4 assignments + a column-vs-row pData indexing bug 2016 Biobase tolerated); we ran the author's OWN lower-level functions (filterAndImputeSamples, collapseDupProbes, removeDupPatients, filterGenesByVariance) on plain matrices — the identical computation without the broken S4 bookkeeping — then the genuine mergeDatasetList. Faithful re-run, not a re-implementation. No fabrication signs: every reproduced value is derivable from the shipped data+code and matches the paper to within curation drift.

💻 Code ↗ 🗄 Data: GSE12470

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 87
    assessed: 2026-06-14 ⛓ 63c2143bb991
✎ 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: opus
Founding hypothesis

Can patient disease subtypes be discovered as replicable 'meta-clusters' across multiple high-dimensional molecular datasets without requiring between-dataset batch-correction transformations, by finding consensus across dataset-specific clusterings rather than clustering a single concatenated matrix?

Core claims
  • CoINcIDE is a methodological framework that discovers replicable patient subtypes (meta-clusters) across multiple datasets by finding consensus across dataset-specific clusterings, requiring no between-dataset transformations. method
  • CoINcIDE extends the unsupervised meta-analysis paradigm by building on the In-Group-Proportion (IGP) metric to compute cluster-cluster similarities and assemble a network whose communities are the final subtypes. method
  • curatedBreastData is a high-quality curated database collection of 24 breast cancer gene expression datasets (over 2,500 samples) with linked outcome and treatment information, released as an R package. resource
  • CoINcIDE validates known breast cancer subtypes (e.g., PAM50 subtypes) across multiple datasets. finding
  • CoINcIDE discovers novel breast and ovarian cancer subtypes with prognostic significance across multiple datasets. finding
  • CoINcIDE identifies novel hypothesized ovarian cancer therapeutic targets across multiple datasets. finding
  • Avoiding concatenation/batch correction preserves true between-dataset signaling variances (e.g., latent subtypes from different clinical trials) and retains per-dataset consensus information that concatenation loses. mechanism
  • Both CoINcIDE and curatedBreastData are distributed as R packages. resource
Experimental setups
Assay System Perturbation Readout Platform
Gene expression microarray clustering (consensus k-means clustering analysis) Human breast cancer primary tumor tissue (curatedBreastData: 2,235 pre-treatment samples, 17 datasets, 15 studies, 6 platforms) none Dataset-specific patient clusters / meta-cluster subtype assignments Affymetrix U133A/U133 Plus 2.0, Agilent, Arcturus, AFFY Human Phase3, MLRG (GEO; ~11,000 genes avg)
Gene expression microarray clustering (CoINcIDE meta-cluster analysis) Human ovarian cancer primary tumor tissue (curatedOvarianData: 3,105 samples, 24 datasets, 10 platforms) none Meta-cluster subtype assignments, prognostic significance, hypothesized therapeutic targets Illumina HumanRef-8, Agilent, Affymetrix, Operon, ABI (GEO/TCGA; ~15,000 genes avg)
Semi-supervised clustering using PAM50 centroids (CoINcIDE evaluation) Human breast cancer datasets (17 datasets) none Patient assignment to 5 PAM50 subtypes (Normal, Basal, Luminal A, Luminal B, HER2) via Pearson correlation PAM50 50-gene centroid set (min 35/50 genes; UNC Microarray Database)
De novo PAM50 gene-set clustering (CoINcIDE) Human breast cancer datasets none De novo dataset-specific clusters / meta-clusters PAM50 gene set
Simulated clustered-dataset benchmark (Eigen decomposition simulation) Simulated gene expression datasets derived from a real 4-tissue-type dataset (7 sets x 10 datasets, 200 simulated genes) Added random normal noise (mean 0, SD 0 to 2.4) True positive rate and false positive rate of CoINcIDE meta-cluster edges (same-tissue connections)
Concatenated-matrix clustering comparison Human breast cancer datasets (17 datasets, minimum 35-gene PAM50 set) none Replicability and biological intuitiveness of discovered subtypes vs CoINcIDE
Key results
  • CoINcIDE re-discovers known PAM50 breast cancer subtypes under semi-supervised conditions across multiple datasets
  • CoINcIDE discovers breast and ovarian cancer subtypes with prognostic significance across multiple datasets
  • CoINcIDE identifies novel hypothesized ovarian cancer therapeutic targets across multiple datasets
  • Final meta-cluster patient assignments are robust to small changes in the mean similarity threshold (first breast cancer analysis gave the same results at thresholds 0.15 and 0.25)
  • Simulation benchmark across 7 dataset sets with increasing noise yielded measurable TPR/FPR for CoINcIDE meta-clusters
Key statistics
  • count over 2,500 breast cancer gene expression samples (curatedBreastData collection size)
  • count 2,235 pre-treatment patient tumor sample microarrays (breast cancer samples after filtering (17 datasets, 15 studies, 6 platforms))
  • count 3,105 samples (ovarian cancer samples (24 datasets, 10 platforms) after filtering)
  • other sample sizes ranged from 46 to 286; average ~11,000 genes per platform (breast cancer dataset characteristics)
  • other sample sizes ranged from 28 to 578; average ~15,000 genes per platform (ovarian cancer dataset characteristics)
  • other recommended averaged nearest neighbor fraction of 0.7 and P value threshold of 0.01 (CoINcIDE edge-assignment thresholds)
  • count 500 null centroid iterations (suggested iterations to compute IGP-based P value)
  • other similarity thresholds of 0.15 and 0.25 (breast PAM50); 0.5 (ovarian) (mean similarity thresholds chosen from density curves)

Statistical methods review

Model: opus

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 is a methods/framework paper presenting CoINcIDE, an unsupervised pipeline that derives patient subtypes by clustering each dataset separately and then finding consensus 'meta-clusters' across datasets without batch correction. Within-dataset clustering uses consensus clustering (Hartigan-Wong k-means with resampling and a PAC score); cross-dataset cluster similarity is quantified with Pearson-correlation-based centroid assignment, a nearest-neighbor fraction, and a mean similarity metric, with a permutation P value built on the In-Group-Proportion (IGP) null distribution. Edges passing user thresholds form a network, and meta-clusters are detected with the Girvan-Newman community detection algorithm. Performance is assessed on simulated datasets via averaged true/false positive rates, and the framework is applied to curated breast and ovarian gene-expression collections.

Replicationbiological Sample sizeDataset-level sample sizes are tabulated (breast: 17 datasets/15 studies, 2,235 samples, n 46–286; ovarian: 24 datasets, 3,105 samples, n 28–578). Inclusion required sample size >30 (breast) and gene-overlap thresholds. No formal power/sample-size calculation described. Groupspatient clusters/meta-clusters across multiple independent gene-expression datasets Pairingna Randomization/blindingna Dispersionnone Exact p-valuesno Effect sizesno Confidence intervalsno Multiplicity correctionnone stated
Statistical tests used
Test Applied to n Assumptions
Permutation/null-distribution P value based on the In-Group-Proportion (IGP) method (clusterRepro), comparing real cluster-centroid assignments to null centroid sets Step 2 cluster-cluster similarity significance for each cluster pair across datasets 500 null centroid iterations not stated
Pearson's correlation (centroid assignment and mean-correlation similarity metric) assigning patients to nearest cluster centroid and computing cluster-cluster similarity; also for PAM50 subtype assignment not stated
Consensus clustering with Hartigan-Wong k-means (one random start, 90% resampling) and a rounded Proportion of Ambiguous Clusters (PAC) score Step 1 single-dataset clustering not stated
Girvan-Newman community detection (igraph) Step 4 meta-cluster identification from the cluster similarity network na
True positive rate / false positive rate computation simulation evaluation of CoINcIDE meta-clusters averaged over 50 simulation iterations per set; 10 datasets per set na
Approaches that could also have been used
  • Significance of cluster-cluster similarity was assessed with a per-pair permutation P value thresholded at 0.01 across many cross-dataset comparisons.
    Could also: A Benjamini-Hochberg FDR or other multiple-comparison adjustment could also be applied across the family of cluster-cluster tests. — Applying a multiplicity correction would also characterize the expected proportion of spurious edges when many pairs are tested, complementing the fixed per-pair threshold.
  • Method performance was summarized as TPR and FPR averaged over 50 simulation iterations.
    Could also: One could also report dispersion (SD/IQR) or 95% confidence intervals around these averaged rates, and/or ROC/AUC across thresholds. — Adding a measure of variability or a full ROC curve would also convey the stability of performance across simulation replicates and threshold choices.
  • Within-dataset clustering used consensus clustering with Hartigan-Wong k-means and the PAC score to choose cluster number.
    Could also: Model-based clustering (e.g., Gaussian mixtures with BIC) or hierarchical/PAM clustering with silhouette/gap statistics could also be used. — These alternatives would also provide principled, distributional or internal-validity criteria for selecting cluster number and could be compared as a sensitivity check.
  • Cluster similarity and centroid assignment relied on Pearson's correlation.
    Could also: Spearman correlation or cosine/Euclidean-based distances could also serve as the similarity metric. — Rank-based or alternative distance metrics would also be more robust to outliers and monotonic-but-nonlinear relationships, offering a check on metric sensitivity.
  • Network-level thresholds (similarity, nearest-neighbor fraction, P value) were chosen via inspection of similarity density curves and recommended defaults.
    Could also: A formal sensitivity analysis sweeping thresholds, or stability-selection across resamples, could also be reported systematically. — Systematically tabulating how meta-clusters change across the threshold grid would also document robustness of the final subtypes to user-chosen parameters.
  • Meta-clusters were detected using the Girvan-Newman community detection algorithm.
    Could also: Modularity-based methods such as Louvain/Leiden or label propagation could also be used for community detection. — These alternatives would also scale efficiently and provide a comparison point for the stability of detected communities.
Software: R package: CoINcIDE · R package: curatedBreastData (Bioconductor) · R package: curatedOvarianData (Bioconductor) · R package: clusterRepro (IGP null distribution) · R package: igraph (Girvan-Newman) · R package: HGNChelper

Result convergence & founder nodes

Findings this paper shares with others that ran a comparable experiment. A node’s strength is how many independent papers report it (replication breadth) — not how often it is cited, so a heavily-replicated but under-cited founder still stands out.

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
37
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.

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.

GPL80 GEO in Methods (http://purl.org/orb/Methods)
also used by 2 papers:
GPL8300 GEO in Methods (http://purl.org/orb/Methods)
also used by 1 paper:
GPL887 GEO in Methods (http://purl.org/orb/Methods)
also used by 1 paper:
GSE2109 GEO in Methods (http://purl.org/orb/Methods)
also used by 1 paper:
GSE9891 GEO in Methods (http://purl.org/orb/Methods)
also used by 1 paper:
GPL15718 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GPL2986 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GPL3291 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GPL6104 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE12470 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE13876 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE14764 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE17260 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE18520 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE19829 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE20565 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE26193 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE26712 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE30161 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE32062 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE32063 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE44104 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE49997 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE6008 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet
GSE6822 GEO in Methods (http://purl.org/orb/Methods)
no other assessed paper uses this yet

Downstream reach in the literature

194 downstream papers · 2 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-26961683 (CoINcIDE; Planey & Gevaert, Genome Med 2016)

Paper: CoINcIDE: A framework for discovery of patient subtypes across multiple datasets. PMID 26961683 · PMCID PMC4784276 · DOI 10.1186/s13073-016-0281-4. Method code: https://github.com/kplaney/CoINcIDE (R/Bioconductor pkg Coincide

  • dissertation analysis scripts). Data pkg: curatedBreastData (Bioconductor experiment-data pkg authored by same author), curatedOvarianData (Bioconductor).

CoINcIDE = "Clustering INtra and Inter DatasEts": per-dataset consensus clustering → cluster centroids → significant edges between cluster centroids across datasets (null permutation test) → community detection on the network → "meta-clusters" (patient subtypes reproducible across datasets).

In scope (pipeline-derived; attempted)

id result (reported) pipeline tractability
C1 filterAndImputeSamples() self-test: 99/100 genes pass NA filter, 29/30 patients pass, imputation avg-abs-error ≈ 0.0052 Coincide pkg testthat unit test on a synthetic matrix; shipped expected output tests/testthat/createTestTrainSetsOutput.txt. Fully deterministic counts (impute seed fixed rng.seed=362436069). HIGH — self-contained, fast
C2 17 breast datasets used (Table 1 / Results; processing filename literally ..._17_studies...) curatedBreastDataprocExprSetList(minVar=.001)mergeDatasetList(minNumGenes=10000, minNumPatients=40) keeps datasets ≥40 patients & ≥10k genes HIGH — deterministic filter on shipped data
C3 2,235 breast pre-treatment tumor samples (Results) sum of samples across the 17 retained datasets HIGH — deterministic
C4 5 breast PAM50 meta-clusters (Fig 3a / Fig 5a) PAM50 centroid clustering per dataset → CoINcIDE computeEdges (500 null iters, p<0.01) → community detection LOW (hard 20%) — full multi-step pipeline, 10-yr-old code, hours; attempt only as stretch

Out of scope (not attempted, with reason)

  • Ovarian analyses (24 datasets, 3/6 meta-clusters; Fig 7): needs curatedOvarianData + full pipeline — same hard-20% class as C4; one disease suffices for a clear data point.
  • GSEA / druggable-gene / survival meta-feature results: require Broad MSigDB gene-list RData object that is NOT in the repo (README: "if the RData object is not in the Github repo, it's because I was nervous about its size … reach out to obtain it"). → data_unavailable for those specific results; not pipeline- reproducible from shipped artifacts.
  • Simulation-study figures: synthetic-data sanity checks, not a reported empirical claim with a single pinnable value.
  • Wet-lab / manual interpretation (subtype biological labels): non-pipeline.

Heavy-compute note

All compute on «our HPC» («infra» SLURM) via «host»; conda prefix env built in-job on «infra». «host» holds only small results + pointers.

Repo: github.com/kplaney/CoINcIDE @ default branch master (last push 2016-08-28). Pinned commit SHA resolved at run time and recorded in code/code.json.

Figures / tables: TableFig 3aFig 5a
C1a
Reported
99 genes pass NA filter (of 100)
Reproduced
99
exact
C1b
Reported
29 patients pass NA filter (of 30)
Reproduced
29
exact
C1c
Reported
0.005165473 imputation avg-abs-error
Reproduced
0.005028749
within tolerance
C2
Reported
17 breast datasets used (>=40 patients & >=10k genes)
Reproduced
17
exact
C3
Reported
2235 breast pre-treatment tumor samples
Reproduced
2237
within tolerance
C4
Reported
5 breast PAM50 meta-clusters (Fig 3a/5a)
Reproduced
not attempted
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 87/100

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

A faithful re-run (author's own Coincide functions on plain matrices to bypass a broken modern-Biobase S4 wrapper): the processing self-test is exact (99 genes/29 patients) and the headline data scale reproduces 1:1 (17 datasets exact, 2237 vs 2235 samples within 0.09%). The two deviations are both technical/expected — curatedBreastData version drift (2024 vs 2016) and an unseeded sample() in the imputation-error estimate. The central 5-meta-cluster discovery (C4) was not attempted (data-prep confirmed, clustering result not), so overall a solid, non-suspicious partial.

🤝
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.

222.2 k
tokens (I/O) · 18.2 M incl. cache
32 min
runtime · 0.12 CPU-h
3.8 GB
peak RAM
4
HPC jobs
hummel
machine