Identification of phenotype-specific networks from paired gene expression-cell shape imaging data.
The main results reproduced: recomputed values matched the published ones within tolerance.
Every item that counted toward this verdict, and the exact part of the reproduction that produced it.
- ✓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
- ✓Overall, the reproduction was clean
- 🟡The central claim did not (fully) hold under reproduction
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 EXACT reproduction of the paper's foundational pipeline step. Paper: Barker et al. 2022 Genome Res, 'Identification of phenotype-specific networks from paired gene expression-cell shape imaging data'. The brief's harvested code link (github.com/dirmeier/diffusr) is a text-mining FALSE POSITIVE; the real authors' code is GitLab gitlab.ebi.ac.uk/petsalakilab/phenotype_networks @ 7f7c9cd3 (MIT, public). The repo ships both the exact input (data/expression/GeneXData.csv, 15304 protein-coding genes x 14 breast-cancer cell lines, derived from Expression Atlas E-MTAB-2706 + E-MTAB-2770) and the authors' reference outputs (ALLgenesprmodule.tab, correlations.txt). I re-ran scripts/wgcna.R faithfully on «our HPC» (R 4.3.3, WGCNA 1.73; signed network, softPower=9, signed TOM, flashClust average, cutreeDynamic deepSplit=2 minClusterSize=30, no merge). Results: 102 WGCNA modules (== paper's 102 GEMs == shipped reference); full gene->module assignment IDENTICAL to the authors' shipped file (Adjusted Rand Index = 1.0 over all 15304 genes, identical module sizes); 34 modules significantly correlated with 8 cell-shape features across 75 module-feature pairs (== paper's '34 modules / 8 features' == 75 rows in shipped correlations.txt). No fabrication signal: every reported number is independently regenerable from shipped data+code. NOT ATTEMPTED (deliberate 80/20 skip of the hard ~20%): DOROTHEA TF activity, DESeq2 differential expression, fgsea enrichment, and the stochastic PCSF/max-flow OmniPath signalling-subnetwork construction (omnipcsf.R) + the RAP1/NF-kB and NAFLD downstream case studies, which depend on live external DB pulls and are not a clean deterministic 1:1.
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 100assessed: 2026-06-15 ⛓ dd89e845699a
✎ 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
- 2026-06-15
- 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 an unbiased, data-driven network-systems approach integrating cell-shape imaging data and RNA-seq expression data identify context-specific gene expression signatures and signaling subnetworks that regulate cell shape in breast cancer, beyond previously known NF-kB pathways?
- ★ A network-based approach integrating RNA-seq and cell-shape imaging data identifies data-derived signaling networks specific to cell-shape regulation in breast cancer. method
- ★ Developmental pathways such as WNT and Notch, along with fine control of NF-kB signaling by kinase and transcriptional regulators, are central to cell-shape regulation. finding
- ★ A gene expression module enriched in the RAP1 signaling pathway mediates between sensing mechanical stimuli and regulation of NF-kB activity, with relevance to breast cancer cell shape. mechanism
- ★ 34 of 102 gene coexpression modules are significantly correlated with cell-shape features. finding
- ★ Breast cancer cell lines cluster into morphologically distinct groups (heterogeneous, luminal-like, basal-like) with distinct gene regulation signatures. finding
- ★ Small-molecule kinase inhibitors targeting proteins in the predicted network significantly perturb cell morphology, validating phenotype-specificity. finding
- ★ A PCSF-derived regulatory network of 691 nodes integrates WGCNA modules, Reactome pathways, TRRUST TFs, and DOROTHEA regulons. resource
- The RAP1 signaling module is up-regulated in basal-like clusters and down-regulated in luminal-like clusters, consistent with negative correlation to neighbor fraction. finding
| Assay | System | Perturbation | Readout | Platform |
|---|---|---|---|---|
| bulk RNA-seq | 13 breast cancer cell lines and one nontumorigenic epithelial breast cell line (MCF-10A) | none | gene expression / coexpression modules (WGCNA) | — |
| high-throughput image analysis / cell-shape imaging | breast cancer cell lines | none | 10 cell-shape variables (size, perimeter, texture of cell and nucleus; ruffliness, neighbor fraction, protrusion area, etc.) | — |
| high-throughput imaging (small-molecule kinase inhibitor screen, LINCS) | breast cancer cell line Hs 578T (also SK-BR-3, MCF-7, MCF-10A) | small-molecule kinase inhibitors (drug); TRAIL as positive control | morphological features (cytoplasmic area/perimeter, nucleus area/length/width/perimeter) | LINCS small-molecule kinase inhibitor data set |
| immunofluorescence imaging | representative breast cancer cell lines per morphological cluster | none | DAPI (nuclei), anti-p65 (NF-kB), DHE staining | — |
| target affinity assay (binding) | kinases / small molecules | small-molecule binding | binding affinities of small molecules to kinases | — |
- – 34 of 102 gene expression modules significantly correlated with one of eight cell-shape features 34/102 modules
- – 17 TF regulons (TRRUST v2) significantly enriched in the modules 17 TFs
- – Six modules shared pathways associated with downstream signaling and regulation of NOTCH 6 modules
- – Kinase inhibitors targeting in-network proteins significantly deviate from control for cytoplasmic area, cytoplasmic perimeter, nucleus area/length/width/perimeter n=37, Wald test P<0.05
- – PCSF network of 691 nodes included 97.11% of genes identified by the analysis 691 nodes, 97.11%
- – Basal-like cluster has lower nuclear/cytoplasmic area, higher ruffliness, lower neighbor fraction vs luminal-like cluster nuc/cyto 0.133±0.05 vs 0.186±0.1; ruffliness 0.235±0.12 vs 0.213±0.14; neighbor 0.258±0.22 vs 0.338±0.26
- – RAP1 signaling module up-regulated in basal-like and down-regulated in luminal-like clusters
- – Pathways from morphology-correlated modules had significantly lower P-values than randomized modules in 1000-resampling null test FDR adjusted P<0.05; 1000 resampled GEMs
- count n = 75,653 (observations/cells used in WGCNA and morphological clustering analyses)
- count 102 GEMs, 34 significantly correlated (gene expression modules identified; 34 correlated at P<0.05 (Student's t-test, Pearson's correlation))
- pvalue P<0.05, Wald test, n=37 (deviation in morphology for kinase inhibitors targeting in-network proteins)
- mean basal nuclear/cytoplasmic area 0.133 ± 0.05; luminal 0.186 ± 0.1 (P<0.001) (one-way ANOVA, Tukey HSD)
- mean basal ruffliness 0.235 ± 0.12; luminal 0.213 ± 0.14 (P<0.001) (morphological cluster comparison)
- mean basal neighbor fraction 0.258 ± 0.22; luminal 0.338 ± 0.26 (P<0.001) (morphological cluster comparison)
- count 691 nodes; 97.11% (PCSF-derived network size and fraction of identified genes included)
- other survival rate 99% (locally contained) vs 27% (metastatic) (breast cancer patient survival rates)
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.
The paper integrates bulk RNA-seq from 14 breast cancer cell lines with high-throughput imaging-derived cell-shape features (n = 75,653 cells) to identify WGCNA co-expression modules correlated with morphology via Student's t-test on Pearson correlations. Cell lines were grouped by k-means clustering on shape features, and cluster differences were tested with one-way ANOVA (Tukey HSD); module-level differential enrichment across clusters used Benjamini-Hochberg-adjusted scores. Transcription factor and pathway enrichment relied on Fisher's exact tests (Enrichr/Reactome) with a resampling-based FDR control, and the resulting network was externally validated against a kinase-inhibitor morphology dataset using Wald tests.
| Test | Applied to | n | Assumptions |
|---|---|---|---|
| Student's t-test on Pearson correlation coefficient | Correlation of 102 WGCNA module eigengenes with 10 cell-shape features; 34 of 102 modules retained at P < 0.05 (Fig. 1B, Supplemental Table S1) | n = 75,653 cells (imaging); 14 cell lines for RNA-seq | not stated |
| Fisher's exact test (via Enrichr) | TF regulon enrichment in gene expression modules; P < 0.1 threshold (Supplemental Table S4) | — | not stated |
| Enrichr over-representation analysis (Fisher's exact test) with Reactome gene sets | Signaling pathway enrichment of TFs regulating morphology-correlated modules (Fig. 1C, Supplemental Table S5) | — | not stated |
| Resampling / permutation test with FDR adjustment | Validation of pathway associations using 1000 resampled GEMs to build pathway-specific null distributions; FDR-adjusted P < 0.05 (Supplemental Table S6) | 1000 resampled module sets | not stated |
| One-way ANOVA with Tukey HSD post-hoc test | Comparison of morphological features (nuclear/cytoplasmic area, ruffliness, neighbor fraction) across three cell-line morphological clusters (Fig. 2A) | n = 75,653 cells | not stated |
| Gene set enrichment analysis (GSEA) with Benjamini-Hochberg FDR correction | Differential enrichment of WGCNA gene expression modules across morphological clusters; adjusted P < 0.01 (Fig. 2B) | — | not stated |
| Wald test | Comparison of morphological effect sizes for kinase inhibitors targeting network proteins vs. non-network proteins; P < 0.05 (Fig. 3A, Supplemental Fig. S4A) | n = 37 kinase inhibitors | not stated |
-
Module-shape correlations were tested with Student's t-test at a nominal P < 0.05 threshold applied across 102 modules and 10 shape features (up to 1020 comparisons)↳ Could also: Apply Benjamini-Hochberg FDR correction simultaneously across all module-feature pairs — Correcting across the full family of comparisons would control the expected false discovery rate and reduce the number of spurious associations reported; the uncorrected approach at P < 0.05 would be expected to produce ~51 false positives by chance alone in 1020 tests
-
Cell-line morphological clusters were identified using k-means clustering, which requires pre-specifying k↳ Could also: Use hierarchical clustering, model-based clustering (e.g., mclust/Gaussian mixture models), or consensus clustering across multiple algorithms — These approaches do not require a pre-specified k, are less sensitive to random initialization, and consensus clustering provides quantitative cluster stability estimates — offering an independent check on the robustness of the three-cluster solution
-
Morphological features were compared across clusters using one-way ANOVA on cell-level observations (n = 75,653 cells from 14 cell lines)↳ Could also: Use a linear mixed-effects model treating cell line as a random effect nested within cluster — Cells from the same cell line share a common microenvironment and genetic background, making them non-independent; a mixed-effects model accounts for within-line correlation and avoids pseudoreplication, where the effective biological n is 14 lines rather than 75,653 cells
-
Dispersion of morphological features across cluster comparisons was reported as mean ± SD↳ Could also: Report 95% confidence intervals around cluster means, or median with IQR given the small number of cell lines per cluster — With only 14 cell lines distributed across three clusters, CI conveys uncertainty in the group-level estimate that SD does not; median and IQR are also robust to outlier cell lines such as HCC1954, which was morphologically clustered with luminal subtypes despite a basal classification
-
TF regulon enrichment in modules used a relatively lenient nominal threshold of P < 0.1 (Fisher's exact test) without stated correction across TF-module combinations↳ Could also: Apply FDR correction across the full matrix of TF-module combinations tested — FDR correction would control the expected proportion of false discoveries across all TF-module pairs and provide a more interpretable threshold, particularly relevant when many combinations are evaluated simultaneously
-
Network validation used Wald tests comparing morphological effects across n = 37 kinase inhibitors split by network membership↳ Could also: Use a label-permutation test that repeatedly shuffles network-membership assignments to build an empirical null distribution — With a modest sample of 37 inhibitors, a permutation test makes no distributional assumptions and directly quantifies whether the observed morphological difference between network-targeted and non-targeted inhibitors exceeds what is expected by chance under random label shuffling
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.
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.
What was reproduced
The exact results taken into scope, with each reported value next to the value our attempt produced.
Scope — pmid-35197309
Paper: Barker CG, Petsalaki E, Giudice G, Sero J, Ekpenyong EN, Bakal C, Petsalaki E. Identification of phenotype-specific networks from paired gene expression–cell shape imaging data. Genome Res 2022;32(4):750–765. DOI 10.1101/gr.276059.121 · PMCID PMC8997347.
Code (authors' own): GitLab gitlab.ebi.ac.uk/petsalakilab/phenotype_networks
@ commit 7f7c9cd3a3838c81c26f5e0d592611cda086ad6d (default branch master, MIT).
NOTE: the brief's harvested code link
github.com/dirmeier/diffusris a text-mining false positive —diffusris a generic network-diffusion R package, not this paper's code. The real analysis repo is the EBI GitLab above, cited in the paper's Data/Code-availability statement ("complete R scripts and data are available as Supplemental Code and at GitLab …").
Data: Expression Atlas FPKM query results E-MTAB-2706 + E-MTAB-2770
(merged in scripts/init_data.R); 14 breast-cancer/epithelial cell lines.
Cell-shape imaging features (Sero et al. 2015, Bakal lab) ship in-repo as
data/phenotype_features/shapefeat*.csv. The repo ships all derived inputs and
reference outputs, so downstream steps are reproducible without re-deriving the
raw imaging.
Pipeline-derived results (in scope)
| Result | Pipeline / script | Reproducible? |
|---|---|---|
| WGCNA gene-expression modules (GEMs) — 102 reported | scripts/wgcna.R (WGCNA: signed network, softPower=9, signed TOM, flashClust average, cutreeDynamic deepSplit=2 minClusterSize=30, no merge) on shipped GeneXData.csv |
YES — chosen target (deterministic; input + reference data/modules/ALLgenesprmodule.tab shipped) |
| Module↔cell-shape correlations — 34 modules / 8 features (P<0.05, |r|≥0.5) | same script; module eigengenes vs shapefeatmedian.csv, corPvalueStudent |
YES — chosen target (reference data/modules/correlations.txt shipped) |
| Expression matrix assembly (15,304 protein-coding genes × 14 lines) | scripts/init_data.R |
YES (shipped GeneXData.csv); used as repro input |
In scope but NOT attempted (the hard ~20%, deliberately skipped per 80/20)
- DOROTHEA TF activity (
scripts/dorothea.R), DESeq2 differential expression (scripts/deSEQX.R), fgsea enrichment (enrich_fgsea.R). - PCSF / OmniPath Prize-Collecting Steiner Forest subnetworks
(
scripts/omnipcsf.R) and maximum-flow signalling subnetworks — multi-step, stochastic (PCSF), and depend on live OmniPath/Reactome pulls; not a clean 1:1. - RAP1 / NF-κB mechanistic module, TRRUST, NAFLD case study — downstream, narrative, partly manual.
Out of scope (not pipeline / not computational)
- Wet-lab cell-shape imaging acquisition (microscopy of 75,653 cells).
- Biological interpretation (WNT/Notch/NF-κB roles, mechanotransduction claims).
Why this target
WGCNA module construction is the foundational, fully-specified, deterministic
step that everything downstream builds on, and the repo ships both the exact input
(GeneXData.csv) and the authors' own reference outputs
(ALLgenesprmodule.tab, correlations.txt) plus the paper's headline numbers
(102 GEMs; 34 modules / 8 features) — giving a clean three-way 1:1 check
(reproduced vs shipped-reference vs paper).
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.
All three reproduced claims are an exact 1:1 match to both the authors' shipped reference outputs and the paper text — 102 GEMs, a gene→module partition with Adjusted Rand Index = 1.0, and 34 modules/8 features/75 significant correlation pairs — with every number independently regenerable from shipped GeneXData.csv via wgcna.R; no fabrication signal. The only caveat is scope: the agent deliberately skipped the stochastic PCSF/OmniPath signalling-network construction and the downstream case studies, which are the paper's titular phenotype-specific networks, so the central conclusion is confirmed only for its foundational module-construction step, not end-to-end. The wrong harvested code link (a text-mining false positive) was correctly overridden with the real authors' GitLab repo, so it caused no error. This is a high-quality reproduction of what was attempted.
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.