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

Single-Cell Transcriptome Analysis Revealed Heterogeneity and Identified Novel Therapeutic Targets for Breast Cancer Subtypes.

Cells · 2023
L1 54/100 3/4
⚑ Flagged for review — a reproduced result did not match the reported value

Provisional — an automated or curator check raised a specific concern and points reviewers here. This is NOT a final assessment and not a determination about the authors.

Why this verdict

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

Reproduced on the brainbox compute brainarbeit.com
✓ What held up
  • Nothing in this column.
What did not (or only partly)
  • 🟡Could not use the authors’ exact input data
  • 🟡Reported values were only indirectly comparable
  • 🟡A deviation arose in the data or preprocessing
  • 🟡A deviation was attributed to the published material
  • 🟡Reported values were not (fully) derivable from the shared data
  • 🟡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
54/100
Reproducibility score
1.1 SD below mean
vs. all fields · 1173 studies
🎯 Scores higher than 14% of all assessed papers rank 997 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

Secondary computational re-analysis of the public Wu et al 2021 BC scRNA-seq atlas (GSE176078); listed code is the third-party tool AltAnalyze (P16). «our HPC» reachable throughout; all compute ran as SLURM jobs on compute node n093 (4 jobs, all COMPLETED exit 0). REPRODUCED EXACTLY: 26 patients (R1b); the GEO 3-way subtype labeling 11 ER+/5 HER2+/10 TNBC (R1c - which the paper itself states is how GEO labels them); and supplementary DE-table sizes 381 (S1, ER+ vs HER2+) and 321 (S4, TNBC vs ER+) plus S2/S3/S5/S6 = 220/386/229/290. NOT REPRODUCIBLE AS DESCRIBED: the headline 49,899-cell count (R1a) - the deposit holds 100,064 cells (full atlas); 49,899 is a ~49.9% downstream subset from the paper's unspecified AltAnalyze ICGS2 cell QC, not derivable from the shipped metadata; and the 13/44/29 DepMap therapeutic-target counts (R3a-c) - a faithful best-effort DE-up x DepMap(<=-0.3 over breast lines) intersection overshoots ~10x (144/110/76) with the wrong rank order, indicating the target-selection rule is under-specified. NOT ATTEMPTED: R4 KMplot survival (external cohort), OPLS-DA (SIMCA, commercial), STRING (web), wet-lab CFU knockdown. No evidence of outright fabrication, but two material methodology-opacity / non-reproducibility findings (49,899 subset; 13/44/29 targets) that a human reviewer should weigh. GSE176078 dataset itself is grade A: complete, internally concordant, delivers exactly what the source atlas promised. All grades PROVISIONAL and human-checkable.

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 50
    assessed: 2026-06-19 ⛓ 6b88477fe2ee
✎ 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
2026-06-30
Rubric version
v1.0
Assessed by
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-06-19
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 integrating single-cell transcriptomic data of EPCAM+ tumor epithelial cells with CRISPR-Cas9 functional screen data delineate the cellular heterogeneity of breast cancer molecular subtypes (ER+, HER2+, ER+HER2+, TNBC) and identify novel subtype-specific therapeutic targets?

Core claims
  • Single-cell transcriptomic analysis of EPCAM+Lin- tumor epithelial cells identifies unique gene signatures that classify ER+, HER2+, ER+HER2+, and TNBC subtypes with high specificity and sensitivity finding
  • Integrating single-cell transcriptomics with CRISPR-Cas9 gene-effect data identified 13 therapeutic targets for ER+, 44 for HER2+, and 29 for TNBC resource
  • ENO1, FDPS, CCT6A, TUBB2A, and PGK1 predict worse relapse-free survival in basal breast cancer and are elevated in aggressive BLIS TNBC finding
  • Targeted depletion of ENO1 and FDPS reduces TNBC cell proliferation, colony formation, migration, and organoid growth while increasing cell death mechanism
  • Several identified targets (e.g., RPS4X, RPL34, VMP1 for ER+; RPS29 for HER2+) outperform current standard-of-care targets (ESR1/ERBB2) in CRISPR gene-effect potency finding
  • FDPS-high TNBC is enriched in cell cycle and mitosis categories, whereas ENO1-high is associated with cell cycle, glycolysis, and ATP metabolic processes finding
  • Integration of single-cell transcriptomics with CRISPR-Cas9 screens provides the first comprehensive dependency map for each BC molecular subtype method
Experimental setups
Assay System Perturbation Readout Platform
single-cell RNA-seq (re-analysis via ICGS2/UMAP/MarkerFinder) 26 breast cancer patients (12 ER+, 3 HER2+, 2 ER+HER2+, 9 TNBC), 49,899 single cells, EPCAM+Lin- epithelial cells none cellular composition and subtype-specific differentially expressed gene markers GSE176078 dataset; AltAnalyze v2.1.3
CRISPR-Cas9 functional screen (data integration) cancer cell lines from Achilles project genome-wide CRISPR-Cas9 knockout gene effect scores (≤-0.3 threshold) for essential genes Achilles project (DepMap)
discriminant analysis (OPLS-DA) / ROC BC subtype gene marker profiles none classification AUC/sensitivity/specificity across subtypes SIMCA v16 (Umetrics)
survival analysis (Kaplan-Meier RFS) 442 basal breast cancer patients none relapse-free survival stratified by median gene expression KMplot database
bulk RNA-seq analysis 360 TNBC patient cohort (BLIS, IM, LAR, MES subtypes) none differential expression and GO enrichment by ENO1/FDPS high vs low KALLISTO 0.4.2.1, GENCODE v33, iDEP.951
Colony Forming Unit (CFU) assay MDA-MB-231 and BT-549 TNBC cell lines siRNA knockdown of ENO1 and FDPS (30 nM) vs scrambled control colony formation (crystal violet absorbance at 590 nm) Lipofectamine 2000; crystal violet
AO/EtBr fluorescence cell death staining MDA-MB-231 and BT-549 TNBC cell lines siRNA knockdown of ENO1 and FDPS number of dead (red/EtBr+) cells Olympus IX73 fluorescence microscope; ImageJ
scratch/wound migration assay and 3D organoid dome culture MDA-MB-231 and BT-549 TNBC cell lines siRNA knockdown of ENO1 and FDPS wound area closure at 24h; number of organoids Matrigel GFR; ImageJ
Key results
  • Identified 13 therapeutic targets for ER+, 44 for HER2+, and 29 for TNBC by integrating scRNA-seq with CRISPR-Cas9 data 13/44/29 targets
  • Gene classifiers discriminated the four BC subtypes with excellent ROC performance AUC: TNBC=0.98, ER+=0.94, HER2+=0.99, ER+HER2+=0.99
  • ENO1, FDPS, CCT6A, TUBB2A, and PGK1 high expression predicted worse RFS in basal BC n=442
  • Highest ENO1 expression in aggressive BLIS TNBC subtype; FDPS, PGK1, CCT6A highest in BLIS and LAR subtypes
  • ENO1 and FDPS siRNA depletion reduced colony formation in MDA-MB-231 and BT-549 TNBC cells
  • RPS4X, RPL34, and VMP1 showed more profound CRISPR gene effects than ESR1 for ER+ BC
  • Differential expression identified subtype-comparison DEG counts (e.g., 381 ER+ vs HER2+; 321 TNBC vs ER+) 381/220/386/321/229/290 DEGs
Key statistics
  • count 49,899 single cells (single cells analyzed from 26 BC patients)
  • other AUC TNBC=0.98, ER+=0.94, HER2+=0.99, ER+HER2+=0.99 (ROC performance of gene classifiers)
  • count 13 ER+, 44 HER2+, 29 TNBC targets (therapeutic targets identified per subtype)
  • pvalue PPI enrichment p ≤ 1.0 × 10^-16 (HER2+ and TNBC gene target PPI network enrichment)
  • pvalue PPI enrichment p = 0.000313 (ER+ gene target PPI network)
  • pvalue GO response to interferon-alpha FDR p = 0.0076 (highest GO enrichment among ER+ targets)
  • count 381 DEGs (differentially expressed genes ER+ vs HER2+)
  • count 442 (basal BC patients in KMplot RFS analysis)

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 combines computational analysis of public single-cell RNA-seq data (differential expression with fold-change and FDR-adjusted p-value cutoffs, OPLS-DA classification with ROC/AUC) with CRISPR-Cas9 dependency screen integration, PPI network enrichment, Kaplan-Meier/log-rank survival analysis in a clinical cohort, bulk RNA-seq differential expression/GO enrichment in a TNBC cohort, and in vitro functional assays (colony formation, apoptosis staining, scratch migration, 3D organoid growth) analyzed with pairwise statistics in GraphPad Prism. Results are reported primarily as fold-change/FDR thresholds, enrichment p-values, log-rank p-values, and mean ± SD for functional assays.

Replicationmixed Sample sizeFunctional assays state experiments were 'repeated at least twice' with data such as CFU presented as mean ± SD from four replicas; clinical/cohort analyses used n=442 (survival) and n=360 (TNBC RNA-seq) as retrieved from existing databases; no formal power/sample-size calculation is described GroupssiRNA knockdown (ENO1, FDPS) vs scrambled control in TNBC cell lines; BC molecular subtypes (ER+, HER2+, ER+HER2+, TNBC); high vs median-split gene expression groups Pairingunclear Randomization/blindingnot stated DispersionSD Confidence intervalsno Multiplicity correctionFDR-adjusted p-value (via AltAnalyze/iDEP.951) for differential expression gene lists
Statistical tests used
Test Applied to n Assumptions
Differential expression analysis with 1.5-fold-change and FDR-adjusted p-value < 0.05 cutoff (AltAnalyze/iDEP.951) Pairwise comparisons among BC molecular subtypes (Figure 2a) and ENO1/FDPS-high vs -low groups in the 360-patient TNBC cohort Varies by comparison (e.g., 381, 220, 386, 321, 229, 290 differentially expressed genes reported per comparison); TNBC cohort n=360 not stated
OPLS-DA discriminant analysis with ROC/AUC evaluation Discrimination of ER+, HER2+, ER+HER2+, and TNBC subtypes based on identified gene markers (Figure 3e,f) 49,899 single cells from 26 BC patients not stated
Kaplan-Meier relapse-free survival analysis with log-rank test Prognostic value of identified TNBC therapeutic targets (Figure 5a–e) n = 442 basal BC patients (KMplot database) not stated
PPI network enrichment analysis (STRING enrichment p-value) Protein-protein interaction networks for ER+, HER2+, and TNBC target gene sets (Figure 4b,d,f) not explicitly stated (based on identified target gene lists: 13, 44, and 29 genes) not stated
Pairwise statistical analyses (specific test not named) in GraphPad Prism v9 In vitro functional assays: colony formation (CFU), apoptosis (AO/EtBr), migration (scratch assay), and organoid growth following ENO1/FDPS knockdown (Figure 6 and related) Experiments repeated at least twice; CFU data from four replicas; migration/organoid counts from three fields not stated
Approaches that could also have been used
  • Differential expression was defined using a fixed 1.5-fold-change plus FDR-adjusted p-value < 0.05 cutoff via AltAnalyze/iDEP.951
    Could also: A model-based differential expression tool with per-gene variance shrinkage (e.g., DESeq2, edgeR, or limma-voom) — These approaches explicitly model count/variance structure across replicates and can provide shrinkage-adjusted fold-change estimates, which some readers find useful alongside fixed-cutoff filtering
  • Functional assay comparisons (colony formation, apoptosis, migration, organoid growth) were analyzed with unspecified 'pairwise statistical analyses' in GraphPad Prism
    Could also: Explicitly naming and justifying the test (e.g., Student's t-test for two groups meeting normality assumptions, or Mann-Whitney U as a non-parametric alternative), and using one-way/two-way ANOVA with a post-hoc correction when more than two groups or cell lines are compared — Naming the specific test and its assumptions supports reproducibility, and an ANOVA-based approach can jointly control error rate when multiple related comparisons (e.g., two cell lines × two targets) are made
  • Survival analysis dichotomized gene expression at the median to form high/low groups for Kaplan-Meier and log-rank testing
    Could also: Cox proportional hazards regression treating gene expression as a continuous variable, or data-driven optimal cutpoint methods — Continuous modeling avoids information loss from dichotomization and allows adjustment for other clinical covariates, which can complement the median-split KM approach
  • No multiplicity-correction method is described for the multiple pairwise comparisons across in vitro functional assays (two targets × two cell lines × several readouts)
    Could also: A formal multiple-comparison correction such as Holm-Bonferroni or FDR across that family of tests — This can help control the family-wise error rate when several related comparisons are drawn from the same experimental system
  • The OPLS-DA classifier's discriminative performance (ROC/AUC) was assessed on the same single-cell dataset used to derive the marker genes
    Could also: Cross-validation (e.g., k-fold) or evaluation on an independent held-out cohort — Independent or cross-validated testing can give a more conservative estimate of how well the classifier generalizes beyond the discovery dataset
  • Dispersion was reported as SD for the colony-formation assay, with other assays not specifying a dispersion measure
    Could also: Consistently reporting SD, SEM, or a 95% confidence interval across all quantified assays — Uniform reporting of a chosen dispersion measure (and CIs in particular) helps convey the precision of small-n functional assay estimates
Software: AltAnalyze v2.1.3 · SIMCA (Umetrics) 16 · STRING database v11.5 · SRA toolkit v2.9.2 · KALLISTO 0.4.2.1 · iDEP iDEP.951 · GraphPad Prism v9 · ImageJ

What was reproduced

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

Scope — pmid-37190091

Paper: Vishnubalaji R, Alajez NM. Single-Cell Transcriptome Analysis Revealed Heterogeneity and Identified Novel Therapeutic Targets for Breast Cancer Subtypes. Cells 2023;12(8):1182. PMID 37190091 · PMCID PMC10137100 · DOI 10.3390/cells12081182.

Nature of the study: This is a secondary computational re-analysis of an existing public scRNA-seq atlas (GSE176078, Wu et al. Nat Genet 2021). The authors did NOT generate new sequencing data for the single-cell part; they downloaded the processed atlas and re-ran a clustering + marker + target-prioritisation pipeline, then added wet-lab validation. The listed "code" is the third-party tool AltAnalyze (https://github.com/nsalomonis/altanalyze) — per BRIEF P16, applying a third-party tool to the paper's data is an equally valid reproduction.

Pipeline(s) named in the paper

  1. AltAnalyze v2.1.3 — CPTT normalisation → ICGS2 (iterative clustering & guide-gene selection 2) → UMAP → MarkerFinder (marker genes per population).
  2. Differential expression between subtypes: 1.5 fold-change, FDR-adj p<0.05 (Tables S1–S6). Tool not fully specified (AltAnalyze / iDEP.951 mentioned).
  3. CRISPR essentiality intersection: DepMap/Achilles gene-effect score ≤ −0.3, crossed with up-regulated DE genes → "therapeutic targets" (13 ER+, 44 HER2+, 29 TNBC).
  4. OPLS-DA classifier in SIMCA v16 (commercial).
  5. STRING v11.5 PPI (web tool).
  6. Survival: 5 TNBC genes (ENO1, FDPS, CCT6A, TUBB2A, PGK1) vs relapse-free survival, n=442 basal BC (external meta-cohort, KM-plotter-style).
  7. Functional validation: ENO1 / FDPS siRNA knockdown, CFU assay (wet lab).

IN SCOPE (pipeline-derived, attempted)

# Result Pipeline Tractability Notes
R1 Dataset structure: 26 patients, subtype composition (12 ER+/3 HER2+/2 ER+HER2+/9 TNBC), 49,899 cells with per-subtype counts (16,350 / 7,824 / 11,487 / 14,238) data download + metadata parse + subsetting HIGH (clean 1:1) Note: GEO labels the 26 as 11 ER+/5 HER2+/10 TNBC — paper RE-CLASSIFIES. Verifiable from the shipped metadata.csv.
R2 ICGS2/UMAP cell populations (immune, fibroblast, pericyte, endothelial, epithelial); MarkerFinder markers AltAnalyze ICGS2 MEDIUM-LOW Paper gives NO numeric cluster counts → only qualitative match possible. Stochastic; not bit-reproducible.
R3 Therapeutic-target counts per subtype (13 / 44 / 29) DE genes ∩ DepMap Achilles ≤−0.3 MEDIUM Needs supplementary DE gene lists (Tables S1–S6) + DepMap CRISPR (public). Intermediate DE counts not reported → reconstruct.
R4 Survival: 5 TNBC genes predict worse RFS (n=442 basal) KM-plotter-style external cohort MEDIUM Reproducible via public KM Plotter; external cohort, semi-in-scope.

OUT OF SCOPE (not attempted; reason)

Result Reason
OPLS-DA classifier (SIMCA v16) Commercial software, not obtainable/scriptable.
STRING PPI network figures Manual web tool; figure-level, no quantitative claim.
ENO1/FDPS knockdown CFU reduction (56–61% / 76–81%) Wet-lab experiment — not computational.
Other immunohistochemistry / protein validation Wet lab.

Reproduction strategy (floor → stretch)

  • Floor (~80%, quick): R1 — download GSE176078 on «infra», parse metadata, verify patient count, subtype composition, total + per-subtype cell counts. Cleanest, most auditable 1:1.
  • Stretch: R3 (target counts) using public supplementary gene lists + DepMap; R4 (survival) via KM Plotter; R2 a qualitative Seurat/AltAnalyze clustering sanity check.

Hard dependency

All data (GSE176078 processed tar.gz, 532.9 MB; DepMap; supplementary tables) lives on «infra», downloaded on «host». No data on «host». Requires the «our HPC» tunnel up.

Figures / tables: Fig 4aFig 4cFig 4eTableTablesFig 5a
R1b
Reported
26 BC patients
Reproduced
26 patients (orig.ident CID3586..CID4535)
exact
R1c
Reported
12 ER+/3 HER2+/2 ER+HER2+/9 TNBC (paper reclassification)
Reproduced
GEO labels reproduced EXACTLY: 11 ER+/5 HER2+/10 TNBC; no ER+HER2+ category in deposit
partial
R1a
Reported
49,899 cells total from GSE176078
Reproduced
100,064 cells in shipped deposit (full Wu 2021 atlas); 49,899 is a non-derivable downstream subset
did not match
R1d-g
Reported
16350/7824/11487/14238 per-subtype cells
Reproduced
GEO-label cells 38241 ER+/19311 HER2+/42512 TNBC (no ER+HER2+); paper split not derivable
did not match
R3d
Reported
381 DE genes ER+ vs HER2+
Reproduced
381 (Table S1 rows)
exact
R3e
Reported
321 DE genes TNBC vs ER+
Reproduced
321 (Table S4 rows)
exact
R3a-c
Reported
13/44/29 therapeutic targets (ER+/HER2+/TNBC) via DE-up x DepMap<=-0.3
Reproduced
best-effort 144/110/76 (mean<=-0.3 over 47 breast lines) or 278/219/129 (any line); ~10x overshoot, wrong rank order
did not match
R4a
Reported
ENO1/FDPS/CCT6A/TUBB2A/PGK1 predict worse RFS (n=442 basal, KMplot)
Reproduced
not attempted (external KMplot meta-cohort; outside deposit-reproduction scope)
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 54/100

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)
🤝
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.

66 k
tokens (I/O) · 3.9 M incl. cache
11 min
runtime
Per-job HPC accounting not captured for this run — the runtime shown is the reproduction’s measured wall-clock time.