An end-to-end scRNA-seq analysis of breast cancer (GSE176078) that goes beyond standard cell-type mapping: it classifies malignant vs. non-malignant epithelial cells, characterizes malignant-state heterogeneity across clinical subtypes, infers cell-cell communication with CellChat, and validates a derived gene signature as a prognostic biomarker in an independent bulk cohort (METABRIC).
| Source | GSE176078 — Wu et al., 2021, Nat Genet |
| Cells | 100,064 total → 94,195 after QC |
| Samples | 26 patients (11 ER+, 5 HER2+, 10 TNBC) |
| Validation cohort | METABRIC (n = 1,980), via cBioPortal |
QC → Normalization → PCA/UMAP → Clustering → SingleR Annotation
│
▼
Epithelial Subset → CNV-Based Malignant Classification → Malignant States
│
▼
CellChat (full object + subtype-stratified comparison)
│
▼
Gene Signature → METABRIC Validation → Survival Analysis
- QC & clustering: Standard Seurat workflow; 31 clusters annotated via SingleR.
- Malignant classification: A chromosome-arm expression-deviation proxy (referenced against immune/stromal cells) classifies epithelial cells as CNV-high (malignant candidate) or CNV-low, used in place of inferCNV/CopyKAT for computational feasibility.
- Malignant states: Five states (basal-like, luminal progenitor-like, mature luminal-like, EMT/mesenchymal, cycling/proliferative) annotated via
AddModuleScore, compared across clinical subtypes with patient-level Kruskal-Wallis testing. - Cell-cell communication: CellChat (human CellChatDB) run on the full object and per subtype, with
liftCellChatharmonization before comparison. - Signature & validation: A 60-gene signature (malignant-state markers + MK pathway genes) scored in METABRIC; prognostic value tested via Kaplan-Meier and Cox regression.
| Metric | Value |
|---|---|
| Malignant candidate cells | 10,548 / 22,729 epithelial cells (46.4%) |
| Malignant states, subtype-associated | 5 / 5 (BH-adj. p < 0.05) |
| CellChat pathways detected | 111 |
| Top subtype-differential pathway | Midkine (MK) |
| Gene signature size | 60 genes |
| Survival HR (adjusted, METABRIC) | 1.31 (95% CI 1.12–1.53), p = 8.9 × 10⁻⁴ |
- Malignant classification uses a simplified CNV-score proxy, not inferCNV/CopyKAT — cells are "CNV-high candidates," not definitively malignant.
- CellChat was run on downsampled cells with reduced permutations (
nboot = 25) for feasibility. - The gene signature is largely proliferation-driven, a near-universally prognostic gene class in breast cancer; this result is best read as pipeline validation, and its value beyond Ki-67/PAM50 remains untested.
- Validated in a single bulk cohort (METABRIC); TCGA-BRCA replication is planned.
├── figures/ # all analysis plots (QC, UMAP, GO/KEGG, malignant subtyping, CellChat, signature)
├── results/ # tables and outputs (.csv/.pdf) underlying each figure
├── scripts/ # analysis scripts, run sequentially
├── main.R
└── README.md
install.packages(c("Seurat", "tidyverse", "patchwork", "data.table", "survival", "survminer"))
BiocManager::install(c("SingleR", "celldex", "clusterProfiler", "ReactomePA"))
devtools::install_github("sqjin/CellChat")
devtools::install_github("immunogenomics/presto")Scripts in scripts/ are run in order, from QC through survival analysis. CNV scoring and CellChat steps are memory/time-intensive (see in-script notes for downsampling parameters).
- Wu, S.Z. et al. A single-cell and spatially resolved atlas of human breast cancers. Nat Genet 53, 1334–1347 (2021).
- Curtis, C. et al. Nature 486, 346–352 (2012) — METABRIC.
- Hao, Y. et al. Nat Biotechnol 42, 293–304 (2024) — Seurat v5.
- Jin, S. et al. Nat Commun 12, 1088 (2021) — CellChat.
Bano Rani — BS Bioinformatics, Department of Computer Science, University of Agriculture Faisalabad
Supervisor: Dr. Sumaira Nishat
⭐ If this project is useful, consider starring the repository.