Deciphering the Immune Microenvironment at the Forefront of Tumor Aggressiveness by Constructing a Regulatory Network with Single-Cell and Spatial Transcriptomi
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
- ✓No relevant deviation in data/preprocessing
- ✓Any deviation was negligible
- 🟡Reported values were only indirectly comparable
- 🔴A deviation was attributed to the published material
- 🟡Reported values were not (fully) derivable from the shared data
- 🟡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
Described well enough for ONE clean data point; not for the headline result. The cIGRN repo (xukun01102021/cIGRN) is a stub: README + ~4 KB of thin R glue wrappers (runsMLnet/runSCENIC/GRNetwork) that require unshipped intermediate files (a processed Seurat tumor.rds with malignant-cluster labels already assigned, a precomputed pySCENIC sample_SCENIC.loom, DEGs.markers.csv), a custom uninstallable 'IGRN' R package, and hard-coded Windows paths (D:/IGRN/). The entire upstream pipeline (Seurat->Harmony->SingleR/CellMarker->inferCNV->clustering->manual malignant naming) that produces the headline numbers is absent, so the cIGRN network result (scMLnet 165->24 for m_cluster2 by macrophages; TFs FOXA1/EZH2/HDAC2/EGR1; L-R pairs) is not reproducible from the shipped code, and even P16 (run scMLnet/pySCENIC on the data) is blocked because those steps still need tumor.rds with the authors' cluster labels. We therefore reproduced the cleanest clearly-specified deterministic pipeline output: the normal-cohort QC cell count. Using the 13 'Normal Total cells' GSMs of GSE161529 (exactly 13 such samples) and the verbatim QC thresholds, we got 58063 cells vs 58058 reported -- a 5-cell (0.009%) match, essentially 1:1, with the gap fully explained by inclusive/exclusive boundary conventions. This strongly corroborates the sample selection + QC and shows no fabrication for C1. NOT attempted (out of 80/20 scope or unreproducible): C2 GSE176078 38121 ER+ (deposited matrix already post-QC), C3 (=C1+C2), C4 10 cell types, C5 5 malignant clusters, C6-C8 the network/TF/LR results (need unshipped upstream object), and the spatial/CIBERSORT/survival analyses. We make no fabrication claim on C6-C8: they are unverifiable from this repo, not demonstrably false.
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 85assessed: 2026-06-15 ⛓ a3b2e39c1199
✎ 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 combining single-cell RNA-seq and spatial transcriptomic data to construct intercellular gene regulatory networks (IGRNs) reveal the regulatory relationships between malignant cells at the invasive front of the tumor microenvironment and immune cells in ER-positive breast cancer, thereby identifying drivers of tumor aggressiveness?
- ★ Combining scRNA-seq and spatial transcriptomics enables inference of malignant cells at the invasive front of the ER+ breast cancer TME and dissection of events at the tumor infiltration forefront method
- ★ High expression of transcription factors FOXA1 and EZH2 plays a key role in driving tumor progression in malignant cells at the invasive front finding
- ★ Elevated levels of downstream target genes ESR1 and CDKN1A are associated with poor prognosis in breast cancer patients finding
- ★ A software pipeline (cIGRN) was developed to automatically construct intercellular gene regulatory networks, reducing errors from single-cell communication analysis and increasing confidence of selected cell communication signals resource
- ★ The malignant subgroup m_cluster2 at the invasive front is enriched in cell cycle and metabolism pathways (MYC, Oxidative Phosphorylation) and is implicated as a key driver of tumor progression mechanism
- Immune cells (T cells, myeloid cells, plasma cells) are enriched around malignant epithelial (luminal/cycling) cells, suggesting physical interaction between epithelial and immune cells in the TME finding
- Epithelial cells (luminal, cycling, basal) represent the malignant cell population in ER+ breast cancer based on chromosomal CNV finding
- Bidirectional signaling between invasive frontline malignant cells and immune cells reveals signaling molecules that may serve as biomarkers or therapeutic targets mechanism
| Assay | System | Perturbation | Readout | Platform |
|---|---|---|---|---|
| single-cell RNA sequencing (scRNA-seq) | normal human breast tissue and ER+ breast cancer patient tissue (GSE161529, GSE176078, GSE164898) | none | transcript counts / cell type composition and clustering | 10x Genomics |
| spatial transcriptomics (ST) | human ER+ breast cancer tissue (BRCA1, BRCA2) | none | spatially resolved gene expression and cell type spatial distribution | 10x Genomics Visium Spatial; Space Ranger v1.2 and v1.3 |
| copy number variation (CNV) inference | epithelial cells of ER+ breast tumor vs normal epithelial reference | none | large-scale chromosomal CNV score to classify malignant vs normal cells | inferCNV R package v1.1.1 |
| pseudotime / developmental trajectory analysis | malignant epithelial cells (scRNA-seq) | none | cell trajectory / pseudotime ordering | Monocle 2 v2.18.0 (DDRTree) |
| intercellular gene regulatory network construction (cIGRN) | malignant cells and immune cells (macrophages/T cells) in ER+ breast cancer TME | none | ligand-receptor, receptor-TF, TF-target regulatory links | scMLnet v0.1.0, pySCENIC v0.11.2 |
| immune cell infiltration deconvolution | TCGA BRCA bulk mRNA expression (luminal A/B, ER+) | none | estimated proportions of 22 immune cell types | CIBERSORT |
| survival / prognosis analysis | TCGA BRCA patients and breast cancer patients (target gene expression) | none | association of immune infiltration and target genes with patient survival | R survival v3.3.1, survminer v0.4.9, GEPIA2 |
| cell type mapping / spatial co-embedding | integrated scRNA-seq and ST data of ER+ breast cancer | none | spatial localization and K-distance between malignant and other cell populations | Seurat v4.3.0, CellTrek v0.0.94 |
- – After quality control, 58,058 cells from normal breast tissue and 38,121 cells from ER+ breast tissue were retained (test set); a total of 96,179 cells from 13 normal and 11 ER+ samples were integrated for analysis 96,179 cells
- – Ten main cell types were annotated in the TME (T cells, myeloid cells, B cells, plasma cells, luminal, basal, cycling epithelial cells, fibroblasts, endothelial, PVL cells) 10 cell types
- ▲ Epithelial cells (luminal, cycling) increased in proportion in ER+ tumors with disease progression and showed significant CNV changes, classifying them as malignant
- ▲ Strong correlation observed among luminal, basal, and cycling cells Spearman correlation coefficient > 0.80
- ▲ Clusters 1,2,4,5,8 had CNV scores significantly higher than reference and were classified as malignant cells (m_cluster1-5) p-value < 0.05
- – m_cluster2, m_cluster3, and m_cluster5 located at the beginning of the developmental trajectory and at the leading/invasive edge of tumor tissue; m_cluster2 is the dominating subgroup
- ▲ m_cluster2 at the invasive front is enriched in MYC and Oxidative Phosphorylation pathways, indicating a role in driving tumor progression
- ▲ FOXA1 and EZH2 high expression drives progression and downstream targets ESR1 and CDKN1A elevation is associated with poor prognosis
- correlation Spearman correlation coefficient > 0.80 (correlation among luminal, basal, and cycling epithelial cells)
- pvalue p-value < 0.05 (CNV scores of malignant clusters significantly higher than reference)
- count 96,179 (total cells integrated from 13 normal and 11 ER+ samples)
- count 58,058 (normal breast tissue cells retained after QC)
- count 38,121 (ER+ breast tissue cells retained after QC)
- count 69,405 (cells in validation dataset (18,688 normal; 50,717 ER+ tumor))
- count 22 (immune cell types estimated by CIBERSORT in TCGA BRCA)
- other approximately 75% (ER+ breast cancer proportion of all breast cancer cases)
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 bioinformatics study integrated scRNA-seq and spatial transcriptomics data from ER+ breast cancer patients to characterize the tumor immune microenvironment at the invasive front. Malignant cells were identified by inferCNV-based chromosomal copy number variation (CNV) scoring, cell types were annotated following PCA/Harmony/UMAP clustering, and intercellular gene regulatory networks were constructed using scMLnet and pySCENIC. Immune cell infiltration was estimated from TCGA bulk RNA data via CIBERSORT, and prognosis associations were assessed by survival analysis using the R survival package. Differential gene expression and GSEA were applied to characterize cluster identities and pathway enrichment throughout.
| Test | Applied to | n | Assumptions |
|---|---|---|---|
| inferCNV CNV score comparison; specific statistical test not named; p < 0.05 threshold applied | Distinguishing malignant vs. normal epithelial cell clusters (Figure 2C) | 8 epithelial subclusters from 11 ER+ tumor samples | not stated |
| Spearman correlation | Correlation between copy number profiles of luminal, basal, and cycling epithelial cell types | — | not stated |
| Wilcoxon rank-sum test (Seurat FindMarkers/FindAllMarkers default; test name not explicitly stated in text) | Differential gene expression for cluster marker identification and TF target gene screening | — | not stated |
| Gene set enrichment analysis (GSEA) with p.adjust < 0.05 | Functional pathway enrichment of malignant cell clusters against MSigDB HALLMARK gene sets | — | not stated |
| Survival analysis with optimal cutpoint dichotomization (surv_cutpoint); specific test (e.g. log-rank) not named | Association of CIBERSORT immune infiltration proportions and regulatory network target gene expression with survival in TCGA BRCA ER+ patients; also assessed via GEPIA2 | — | not stated |
| Multivariate prognostic analysis; method not specified; p < 0.05 criterion used for variable selection | Multivariate survival analysis using variables significant at p < 0.05 in univariate step | — | not stated |
-
Differential gene expression between cell clusters used Seurat's FindMarkers function, which by default applies a Wilcoxon rank-sum test treating each cell as an independent observation↳ Could also: A pseudo-bulk approach — aggregating raw counts per donor per cluster, then applying DESeq2 or edgeR — could also be used — Pseudo-bulk methods account for within-patient correlation (cells from the same donor are not statistically independent), which is particularly relevant here because multiple patients contributed cells to each cluster; this approach can reduce false-positive rates in multi-donor single-cell studies
-
Patient survival groups were dichotomized using a data-driven optimal cutpoint algorithm (surv_cutpoint in survminer) on immune infiltration proportions and gene expression values↳ Could also: Continuous Cox proportional hazards regression could also model the predictor without dichotomization — Continuous Cox regression preserves full information in the predictor, avoids the inflated Type I error risk associated with data-driven cutpoint selection, and produces hazard ratios with confidence intervals as a standard effect size measure
-
Malignant cells were identified by thresholding inferCNV CNV scores relative to a normal epithelial reference, with significance at p < 0.05 (specific test not named)↳ Could also: Complementary tools such as CopyKAT or SCEVAN could also be applied for CNV-based malignant cell inference from scRNA-seq — Different CNV inference tools employ distinct statistical models and reference assumptions; requiring concordance across two tools can increase confidence in malignant cell assignments and reduce reference-dependent artifacts
-
Immune cell infiltration in TCGA bulk RNA data was estimated using CIBERSORT (LM22 signature, 22 cell types)↳ Could also: Alternative deconvolution methods such as TIMER2, xCell, or EPIC could also be applied to the same bulk data — Different deconvolution algorithms make distinct assumptions about cell-type signatures and mixture linearity; comparing estimates across methods can identify which cell-type associations are robust to algorithmic choice
-
Pathway enrichment was assessed using GSEA against the MSigDB HALLMARK gene set collection with a p.adjust < 0.05 threshold↳ Could also: Over-representation analysis (ORA) using a hypergeometric test on the DEG list, or inclusion of additional gene set collections (KEGG, Reactome, GO), could also be applied — GSEA and ORA make different assumptions (ranked full gene list vs. discrete DEG threshold); using both, or broader gene set collections, can reveal enriched biology that either approach alone might miss
-
Quantitative results throughout were reported with p-value thresholds (< 0.05) only, without measures of dispersion, effect sizes, or confidence intervals↳ Could also: Reporting effect sizes (e.g., log-fold changes with confidence intervals, hazard ratios with 95% CIs) alongside p-value thresholds would also be standard practice — Effect sizes and confidence intervals convey the magnitude and precision of associations independently of sample size, which is particularly informative when results are proposed as potential biomarkers or therapeutic targets
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-38254989 (cIGRN, Genes 2024, 15:100)
Paper: "Deciphering the Immune Microenvironment at the Forefront of Tumor
Aggressiveness by Constructing a Regulatory Network with Single-Cell and Spatial
Transcriptomics." PMID 38254989 · PMCID PMC10815467 · DOI 10.3390/genes15010100.
Code: https://github.com/xukun01102021/cIGRN (branches main, cIGRN, CCnetwork).
Data: GSE161529 (normal breast), GSE176078 (ER+ tumor), 10x spatial (BRCA1/BRCA2), TCGA-BRCA.
What the repo actually ships (assessed 2026-06-15, commit cIGRN@6b96679)
The repo is essentially a stub / thin glue layer, NOT a runnable end-to-end pipeline:
mainbranch: README only.cIGRNbranch: 4 tiny R files (cIGRN.Rdriver,IGRNetwork.R=GRNetwork(),runSCENIC.R,runsMLnet.R). Total ~4 KB.CCnetworkbranch: 7 small R wrappers around CellPhoneDB / NicheNet / SCENIC / DEG.
The driver cIGRN.R requires inputs that are NOT provided and NOT producible
from anything in the repo:
D:/IGRN/tumor.rds— a fully processed Seurat object with malignant-cluster labels (m_cluster1-5) already assigned. The entire upstream pipeline that produces it (load → Seurat → Harmony → SingleR/CellMarker → inferCNV → clustering → malignant-cluster naming) is absent.D:/IGRN/scenic/.../sample_SCENIC.loom— a precomputed pySCENIC loom; the pySCENIC GRN step that produces it is absent (only a post-hoc loom reader is shipped).D:/Ryy/tumor/DEGs.markers.csv, hard-coded Windows paths, and a customIGRNR package (library(IGRN)) with no install instructions / not on CRAN/Bioc.
→ The headline cIGRN result (e.g. "scMLnet reduced signals 165→24 for m_cluster2 mediated by macrophages", the L-R-TF-target chains, the specific TFs FOXA1/EZH2/HDAC2/ EGR1) is not reproducible from the shipped code without re-implementing the entire unshipped upstream pipeline AND matching their exact (non-deterministic, under-specified) clustering/annotation. That is squarely the "hard last 20%+" the brief says to skip.
In-scope (attempted) — clearly-specified, low-hanging pipeline output
| id | result | pipeline | reproducible? |
|---|---|---|---|
| C1 | GSE161529 normal cohort: 58,058 cells retained after QC (13 normal samples) | Seurat-style QC: mito ≤20%, UMI 200–60000, genes >200 | YES — data public, QC fully specified, deterministic. ATTEMPTED. |
The 13 "normal samples" map cleanly to the 13 Normal-Total GSMs in GSE161529
(GSM4909253/254/257/261/263/265/266/268/270/271/272/274/276 — the *-Total matrices,
not the sorted epithelial-only samples). QC thresholds are stated verbatim in Methods.
This exercises the paper's own data + its documented QC pipeline → an honest 1:1 check.
Out of scope / not attempted (why)
- cIGRN network construction (C-headline) — needs unshipped upstream pipeline +
intermediate files; custom
IGRNpkg uninstallable; hard-coded local paths.docs_insufficientfor the network result itself. P16 (run scMLnet/pySCENIC on the data) is blocked because even those steps requiretumor.rdswith the authors' malignant-cluster labels, which we would have to reconstruct (non-deterministic) first. - GSE176078 38,121 ER+ cells — deposited matrix is already post-QC by Wu et al; re-applying QC is not a clean 1:1 of a raw→QC step. Lower priority; not attempted to avoid chasing the tail.
- inferCNV malignant clusters (5), 10 cell types, 8 epithelial clusters — depend on stochastic clustering + manual annotation; under-specified; out of 80/20 scope.
- CellTrek spatial localization, CIBERSORT proportions, GEPIA2/TCGA survival, GSEA — external/manual or rely on the unreproducible upstream object. Out of scope.
Verdict orientation
Code is a thin glue layer over third-party tools; the reproducible-as-specified surface is the QC cell count (C1). Everything downstream depends on an unshipped, partly non-deterministic upstream pipeline
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.
We could verify only one clearly-specified deterministic output — the GSE161529 normal-cohort QC count — and it matched essentially 1:1 (58,063 vs 58,058, 5 cells / 0.009%), with the gap fully explained by QC boundary conventions and no fabrication indicated. The study's central results (the cIGRN network: scMLnet 165→24 for m_cluster2, TFs FOXA1/EZH2/HDAC2/EGR1, L-R pairs) are unverifiable because the repo xukun01102021/cIGRN is a stub — thin glue wrappers, hard-coded Windows paths, an uninstallable custom IGRN package, and required intermediate files (tumor.rds, SCENIC loom) that are neither shipped nor producible. The defect is on the authors' side (incomplete code/intermediate sharing), not a measured discrepancy or fabrication; hence solid-where-checkable but headline untested → overall yellow.
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.