Single-Cell Transcriptomic Landscape of Right-Sided Colon Cancer Reveals Cellular and Molecular Features of Metastatic Potential.
The main results reproduced, with only marginal, non-material deviations.
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
- 🟡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
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
PARTIAL, with two human-checkable internal inconsistencies between the paper's Methods and its own reported numbers. This is a re-analysis paper: no authors' code repo (brief's 'code' = the third-party tool inferCNV); reproducible via P16 by running the named standard tools on the paper's data. Data: 8 PUBLIC GEO 10x samples (GSE245552 x5 RCC_LM + GSE188711 x3 RCC_noM) named in Methods 2.1 -- note the Data Availability Statement says 'available upon request', which understates availability since the raw inputs are fully public. I reproduced the DATA-CONSTRUCTION BACKBONE (Methods 2.1) on «our HPC» with the EXACT reported Seurat 4.3.0.1. Findings: (ENV) Seurat version reproduced exactly. (C1) The reported '33583 unique gene symbols (union)' cannot be a union -- a true union of the deposited matrices is 43789, and 33583 is impossibly BELOW GSE188711's own 33694-gene reference; it best matches DETECTED (nonzero) genes = 32476 (96.7%), i.e. the number is mislabeled and not exactly reproducible either way. (C2) HGNC-filtered gene count 23776 vs reported 20389 -- same order, not 1:1 (HGNC-snapshot/version + starting-set dependent). (C3) The stated '>10000 UMI excluded' filter, if applied, yields only 26245 cells BEFORE doublet removal, so the reported post-doublet 31674 is unreachable as written; WITHOUT that cap, 32285 pre-doublet -> ~31674 after light doublet removal, fully consistent. FABRICATION READ: NOT clearly fabricated -- both off-claims reconcile to the public data under small, plausible reinterpretations (detected-genes not union; no UMI cap) -- but the Methods text is IMPRECISE/CONTRADICTED by the paper's own numbers. Flagged for human audit as POSSIBLE MISREPORTING, not asserted fabrication. NOT ATTEMPTED (hard ~20%, see scope.md): inferCNV CNV scores + Friedman p=0.000912 + Fig3 significances, 5 tumor subclusters (Fig2), DEG/GSEA/KEGG-metabolic/Monocle2/CytoTRACE (Figs4-7) -- all need full upstream Harmony+Louvain+MANUAL annotation and are MSigDB-version/manual-root-dependent and figure-only. Also not chased: the exact (stochastic) DoubletFinder count -- v2.0.3 (paper's, Seurat4 API) has no git tag and master 2.0.6 needs Seurat5; the C3 finding does not require it. All heavy compute on «our HPC»/«infra»; «host» holds only small results + pointers.
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 71assessed: 2026-06-16 ⛓ 96c7a347f74b
✎ 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
- 2026-06-16
- Rubric version
- v1.0
- Assessed by
-
🤖 AI curator · claude (ai-curator room) · v1.0 · run #1 2026-06-16no 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: opusDoes the primary tumor of right-sided colon cancer (RCC) harbor distinct cellular and transcriptional features that distinguish tumors prone to liver metastasis (RCC_LM) from non-metastatic tumors (RCC_noM), resolvable at single-cell resolution?
- ★ Liver metastatic potential in RCC is marked by stem-like tumor states, metabolic plasticity, and microenvironmental remodeling. finding
- ★ RCC_LM tumors exhibit higher genomic instability and a significantly higher proportion of cells in G1 phase, implicating altered cell cycle progression as a feature of metastatic potential. finding
- ★ Stem-like tumor cells are significantly enriched in RCC_LM whereas enterocyte-like cells predominate in RCC_noM. finding
- ★ RCC_LM primary tumors display transcriptional programs of epithelial–mesenchymal transition, extracellular matrix remodeling, inflammatory signaling, and metabolic reprogramming (glycolysis and oxidative phosphorylation). mechanism
- ★ Trajectory analysis shows RCC_LM tumors enriched in early pseudotime states, suggesting increased cellular plasticity. finding
- ★ Integrated single-cell atlas of right-sided colon cancer with and without liver metastasis from eight patients. resource
- Malignant epithelial cells can be identified via inferCNV using B cells as a diploid reference, and genomic instability quantified by a CNV score. method
- Differentiation potential of malignant cells assessed by CytoTRACE and state transitions reconstructed by Monocle2 pseudotime. method
| Assay | System | Perturbation | Readout | Platform |
|---|---|---|---|---|
| single-cell RNA-seq (droplet-based, integrated public datasets) | primary right-sided colon tumors, 8 patients (5 RCC_LM, 3 RCC_noM) | none (observational; comparison of metastatic vs non-metastatic) | single-cell transcriptome / cell type composition | 10x Genomics; Cell Ranger v5.0.1 (GSE245552) and v3.0 (GSE188711), hg38/GRCh38 |
| copy number variation inference (inferCNV) | single cells from RCC tumors, B cells as diploid reference | none | CNV score / large-scale chromosomal copy number alterations (genomic instability) | inferCNV v1.16.0 |
| dimensionality reduction, clustering, annotation | integrated single cells and malignant cell subset | none | cell lineages and 5 tumor subclusters via marker genes | Seurat v4.3.0.1, Harmony v1.2.0, t-SNE/UMAP |
| trajectory / pseudotime analysis | malignant epithelial cells | none | pseudotime ordering, branch-dependent gene dynamics | Monocle2 (DDRTree, BEAM) |
| differentiation potential analysis | malignant epithelial cells | none | stemness/differentiation score | CytoTRACE v0.3.3 |
| differential expression and pathway enrichment (GSEA) | epithelial subset, RCC_LM vs RCC_noM | none | DEGs and enriched HALLMARK/REACTOME pathways (NES) | Seurat FindMarkers (Wilcoxon), fgseaMultilevel, MSigDB/msigdbr |
| metabolic pathway activity analysis | cell types from RCC_LM vs RCC_noM | none | weighted relative metabolic pathway activity scores | KEGG metabolism gene sets (Xiao et al. method) |
| doublet detection / quality control | merged single-cell dataset | none | high-confidence singlets retained | DoubletFinder v2.0.3 |
- ▲ RCC_LM tumors show higher genomic instability (higher CNV scores) than RCC_noM.
- ▲ RCC_LM tumors have a significantly higher proportion of cells in G1 phase.
- – Stem-like tumor cells significantly enriched in RCC_LM; enterocyte-like cells predominate in RCC_noM.
- ▲ RCC_LM tumors enriched in early pseudotime states, indicating greater cellular plasticity/stemness.
- ▲ EMT, ECM remodeling, inflammatory signaling, glycolysis and OXPHOS programs enriched in RCC_LM (positive NES).
- – Eight high-quality right-sided colon cancer samples yielded 31,674 single cells across 20,389 genes for analysis. 31,674 cells; 20,389 genes
- count eight patients (5 RCC_LM, 3 RCC_noM) (samples retained for analysis (5 from GSE245552, 3 from GSE188711))
- count 31,674 high-quality single cells (integrated dataset after QC and doublet removal)
- count 20,389 expressed/HGNC-approved genes (final gene set for downstream analyses)
- count 33,583 unique gene symbols (detected after merging the two datasets in Seurat)
- count 41,365 unique symbols (HGNC approved list retrieved via biomaRt for cross-referencing)
- other objective response rates of 40–65% (MSI-H tumors treated with PD-1 blockade (background literature))
- other median overall survival 10.1 months (RCC group on anti-EGFR therapy vs LCC (background literature))
- other up to 50% (proportion of CRC patients affected by liver metastasis (background))
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.
This study integrated public scRNA-seq data from eight right-sided colon cancer patients (five with synchronous liver metastasis [RCC_LM] and three without [RCC_noM]) from two GEO datasets, applying Harmony for batch correction and Seurat for preprocessing, clustering, and differential expression. Differential gene expression between groups in the epithelial subset was assessed with the Wilcoxon rank-sum test (BH-adjusted), while CNV score group differences used Kruskal–Wallis with pairwise Wilcoxon follow-up (BH-adjusted), metabolic pathway activity used a permutation test (5,000 shuffles), and pathway enrichment used fgseaMultilevel; adjusted p < 0.05 was the significance threshold throughout. Pseudotime and differentiation potential were assessed with Monocle2 and CytoTRACE, with results reported as normalized enrichment scores, log2 fold-changes, and empirical p-values.
| Test | Applied to | n | Assumptions |
|---|---|---|---|
| Wilcoxon rank-sum test (Seurat FindMarkers, min.pct = 0.25, logfc.threshold = 0) | Differential gene expression between RCC_LM and RCC_noM in the malignant epithelial subset | Cell-level; 31,674 total cells after QC (epithelial subset size not specified) | not stated |
| Kruskal–Wallis test | CNV score differences among B cells, Proliferating cells, and Tumor cell groups | — | not stated |
| Pairwise Wilcoxon test with Benjamini–Hochberg adjustment | Post-hoc pairwise CNV score comparisons among B cells, Proliferating cells, and Tumor cells; and between RCC_LM and RCC_noM | — | not stated |
| Permutation test (5,000 shuffles of cell labels, empirical null distribution) | Metabolic pathway activity score comparison between RCC_LM and RCC_noM using KEGG gene sets | — | not stated |
| fgseaMultilevel (GSEA) | HALLMARK and REACTOME pathway enrichment using ranked gene list (ranked by log2FC) from RCC_LM vs RCC_noM DEG | — | na |
| BEAM (Monocle2 branch expression analysis module) | Branch-dependent gene expression dynamics along pseudotime trajectory of malignant epithelial cells | — | not stated |
-
Differential gene expression between RCC_LM and RCC_noM was tested using a Wilcoxon rank-sum test applied at the individual-cell level↳ Could also: A pseudobulk approach — aggregating counts per patient and then applying DESeq2 or edgeR — could also be used to test differential expression — Pseudobulk methods treat each patient as the statistical unit, which accounts for the non-independence of cells from the same patient (within-donor correlation) and aligns the degrees of freedom with the actual number of biological replicates (here, 5 vs 3 patients)
-
Batch effects from two source datasets were corrected using Harmony applied to PCA embeddings↳ Could also: Seurat's reciprocal PCA (RPCA) integration or scVI (a variational autoencoder-based method) could also be used for multi-dataset integration — RPCA is computationally efficient for datasets with strong batch effects, while scVI jointly models batch and biological variation in a probabilistic framework and can output corrected count-level data; comparing results across methods can help assess integration robustness
-
Pseudotime trajectory inference was performed using Monocle2 with the DDRTree dimensionality reduction algorithm↳ Could also: Slingshot or PAGA (partition-based graph abstraction) could also perform trajectory inference on the same data — Slingshot is flexible about the number and shape of lineages and does not assume a single root, while PAGA provides a graph-level connectivity estimate before committing to continuous trajectories; using a second method can corroborate Monocle2 findings
-
Cellular differentiation potential was estimated using CytoTRACE, which is based on transcriptional diversity (number of expressed genes per cell)↳ Could also: RNA velocity (e.g., scVelo) could also be used to assess directionality of cell state transitions — RNA velocity infers the future transcriptional state of each cell from the ratio of spliced to unspliced mRNA, providing a complementary, kinetics-based perspective on differentiation direction that does not rely solely on gene count diversity
-
Metabolic pathway activity was assessed per pathway using a permutation test with empirical p-values, without explicit FDR correction across the full set of KEGG pathways evaluated↳ Could also: Applying a BH FDR correction across all simultaneously tested KEGG pathways could also be used — When many pathways are tested in parallel, a pathway-level FDR step (e.g., BH across all pathway p-values) is a standard way to control the expected proportion of false discoveries in the reported pathway list
-
Group comparisons pooled all cells from 5 RCC_LM patients and 3 RCC_noM patients, with the patient treated implicitly as a grouping variable via metadata↳ Could also: A mixed-effects model (e.g., lme4 or MAST's hurdle model with patient as a random effect) could also be used to test gene-level differences while explicitly modeling within-patient cell correlations — Mixed-effects models formally partition variance between the patient level and the cell level, which can improve inference when patient-level variability is substantial relative to the small number of donors
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.
What was reproduced
The exact results taken into scope, with each reported value next to the value our attempt produced.
Scope — pmid-41898210
Paper: Ye Z, Zhang W, Qiu H, Luo F, Liao C, Lei K, Zhou Q (2026). Single-Cell Transcriptomic Landscape of Right-Sided Colon Cancer Reveals Cellular and Molecular Features of Metastatic Potential. Biomedicines 14(3), article 563. PMID 41898210 · PMCID PMC13024220 · DOI 10.3390/biomedicines14030563.
"Code" (brief): https://github.com/broadinstitute/inferCNV — a THIRD-PARTY tool. The authors ship no own code repository. The paper is a re-analysis built entirely from standard public tools, named with versions in the Methods: Seurat v4.3.0.1, Harmony v1.2.0, DoubletFinder v2.0.3, HGNChelper, biomaRt, inferCNV v1.16.0, Monocle2, CytoTRACE v0.3.3, fgsea/msigdbr. Per rule P16, applying these tools to the paper's data is an equally-valid reproduction.
Data: TWO public GEO datasets, sub-selected to 8 right-sided colon tumor samples (Methods §2.1) — all 10x matrices are PUBLICLY downloadable from GEO FTP (verified):
- GSE245552 (CellRanger 5.0.1, GRCh38) — 5 samples = RCC_LM (liver-metastasis): GSM7844812 (Pt01_T1), GSM7844823 (Pt06_T1), GSM7844833 (Pt10_T1), GSM7844837 (Pt13_T1), GSM7844844 (Pt17_T1).
- GSE188711 (CellRanger 3.0, GRCh38) — 3 samples = RCC_noM (non-metastatic): GSM5688709, GSM5688710, GSM5688711 (Right-sided CRC 1–3).
Auditor flag (transparency, not fabrication): the paper's Data Availability Statement reads "available from the corresponding authors upon request," yet the actual raw inputs are fully public GEO samples named in the Methods. The "upon request" wording understates availability; the data ARE reproducible without contacting the authors. Also note GSE245552 is the deposit of a different study (Th17/TWEAK CRLM, PMID 38335276) — this paper reuses a 5-sample subset of it.
In scope (pipeline-derived, ATTEMPTED — clean deterministic data-fidelity checks)
The cleanest, lowest-effort, highest-auditability reported numbers are the data-loading/QC counts in Methods §2.1. They test whether the stated inputs and filters actually yield the reported dataset size — a direct fabrication check.
| id | reported result | paper loc | pipeline step | determinism |
|---|---|---|---|---|
| C1 | 33,583 unique gene symbols after merging the 8 samples (Seurat merge, union of genes) |
§2.1 | Read10X → CreateSeuratObject → merge → nrow |
fully deterministic (depends only on the deposited feature lists) |
| C2 | 20,389 HGNC-approved genes used for all downstream analyses (after HGNChelper vs biomaRt's 41,365-symbol HGNC list) | §2.1 | HGNChelper alias-fix + intersect with HGNC approved symbols | version-dependent (biomaRt/HGNC snapshot) — within-tol expected |
| C3 | 31,674 high-quality single cells after QC (exclude <300 genes, >15% mito, >10,000 UMI) + DoubletFinder (rate 7.5%, pN=0.25, set.seed(1)) | §2.1 | per-sample QC + DoubletFinder | QC deterministic; DoubletFinder mildly stochastic → report deterministic post-QC count too |
Out of scope (NOT attempted — the hard ~20%, with reasons)
- inferCNV CNV scores + Tumor > Proliferating > B (p<0.0001), Friedman p=0.000912, RCC_LM > RCC_noM CNV (Fig 3A–C). Needs the full upstream pipeline (Harmony integration, Louvain clustering at res=0.1, manual cell-type annotation to define the B-cell reference + tumor cells) before inferCNV can even run; the headline p-values are direction claims. The defining tool, but the heavy 20%.
- 5 tumor subclusters + proportions (Fig 2): res=0.1 on 25 UMAP PCs + manual annotation — unpinned, qualitative.
- DEGs / HALLMARK+REACTOME GSEA / KEGG metabolic activity / Monocle2 / CytoTRACE (Figs 4–7): MSigDB-version-dependent, manual Monocle root selection, figure-only (no in-text scalars to compare). Not attempted.
Honest framing
This is a partial, 80/20 reproduction by design. What we reproduce 1:1 is the data-construction backbone (gene-union count, HGNC-filter
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.
Input data is fully public (8 GEO 10x samples) and the environment reproduced exactly (Seurat 4.3.0.1), so comparison was direct. The deviations sit at the data-construction/preprocessing layer: C1 reports 33583 'unique gene symbols (union)' which is mathematically impossible as a union (it falls below GSE188711's own 33694-gene reference; true union=43789, detected=32476), and C3's 31674 cells is reachable only by ignoring the explicitly-stated 10k-UMI cap (with the cap, pre-doublet=26245). These are authors'-side imprecise/contradictory Methods reporting, not our methodology, and reconcile under small reinterpretations — flagged as possible misreporting, not asserted fabrication. The central biological conclusion (CNV/subcluster/GSEA, Figs 2-7) was not attempted, so it remains untested rather than overturned.
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.