PANGOLIN: Pan-cancer Analysis of Gene regulatory landscape Of LIoness Networks

Overview

PANGOLIN is a comprehensive Snakemake pipeline for pan-cancer analysis of gene regulatory networks using TCGA data, with a specific focus on PD-1 pathway analysis. This pipeline reproduces all analyses and figures from our manuscript investigating gene regulatory landscapes across 33 cancer types.

🚀 Quick Start

Prerequisites

• Snakemake ≥ 7.23.1
• Conda for environment management
• R ≥ 4.2.1 with Bioconductor packages
• Python ≥ 3.8
• Sufficient disk space: ~1TB for full workflow, ~20GB for precomputed analysis

Installation

# Clone the repository
git clone https://github.com/kuijjerlab/PANGOLIN.git
cd PANGOLIN

Configuration

1. Choose analysis type in config.yaml:

   # For complete reproduction (takes several days/weeks due to the single-sample network reconstruction for over 9 000 samples)
   analysis_type: "full_workflow"
   
   # For a more rapid analysis and figure reproduction using precomputed data (RECOMMENDED)
   analysis_type: "precomputed"

Execution

# Dry run to check workflow
snakemake --use-conda --conda-frontend conda --cores 1 -np

# Execute pipeline
snakemake --use-conda --conda-frontend conda --cores 1

📊 Analysis Workflow

Data Processing Pipeline

1. 📥 Data Acquisition

   • Downloads TCGA expression data for 33 cancer types using TCGAbiolinks package
   • Retrieves clinical information
   • Processes batch information and produces batch figure

2. 🔧 Expression Data Normalization

   • Combines multi-cancer expression matrices
   • Applies qsmooth normalization using PySNAIL
   • Performs batch effect detection and correction

3. 🕸️ Network Inference (Full workflow only)

   • Constructs gene regulatory network using PANDA
   • Generates patient-specific networks with LIONESS
   • Calculates network-based features (in-degree)

4. 📈 Dimensionality Reduction & Pathway Analysis

   • t-SNE analysis of expression and network features
   • PORCUPINE pathway heterogeneity analysis
   • Reactome pathway enrichment

5. 🧬 PD-1 Pathway Analysis

   • Extracts PD-1 pathway components and scores
   • Correlates with immune infiltration (CIBERSORTx)
   • Clinical association analysis (survival, molecular features)

6. 🎯 Clustering Analysis

   • Consensus clustering using Cola
   • Cancer type-specific cluster characterization
   • Survival analysis of identified clusters
   • Comparison of clusters for PRAD

📁 Output Structure

results/
├── data_all/                              # Pan-cancer results
│   ├── gdc_data/                            # Raw TCGA data
│   ├── batch_analysis/                      # Batch effect analysis
│   ├── batch_corrected_expression/          # Batch-corrected expression data
│   ├── clinical_associations_PD1/           # PD-1 clinical associations
│   ├── cola_consensus_clustering/           # all cancers consensus clustering
│   ├── combined_gdc_data/                   # Combined normalized expression
│   ├── cox_results_all/                     # survival analysis
│   ├── porcupine/                           # PORCUPINE pathway analysis
│   ├── pysnail_normalized_individual_cancer_expression/ # Normalized data per cancer
│   ├── tsne_results/                        # t-SNE dimensionality reduction
│   └── logs/                                # Processing logs
├── data_individual_cancers/                 # Cancer-specific results
│   └── [CANCER]/                            # Individual cancer directories
│       ├── pd1_data/                          # PD-1 scores and mappings
│       ├── cox/                               # Survival analysis
│       ├── consensus_clustering/              # Cancer-specific clustering
│       ├── final_clusters/                    # Final cluster assignments
│       ├── clinical_associations/             # Clinical correlations
│       ├── indegrees_norm/                    # Network in-degree features
│       ├── porcupine/                         # Pathway analysis results
│       └── clinical/                          # Clinical data processing
├── panda_input/                           # PANDA network input files
└── figs/                                  # Publication-ready figures
    ├── MBatch_DSC.pdf                       # Batch effect summary
    ├── TSNE_*.pdf                           # t-SNE visualizations
    ├── PC_immune_correlations_cibersort.png # Immune correlations
    ├── cox_results_final_clusters_*.pdf     # Survival analysis plots
    ├── PRAD_clusters_*.pdf                  # PRAD-specific analyses
    ├── pathways_intersection_pcp.pdf        # Pathway intersection analysis
    ├── sankey_plot_indegree_expression.pdf  # Sankey diagrams
    └── summary_table_PD1.html               # Results summary table

🎯 Key Features

Analysis Types

• Full Workflow: Complete analysis from raw data (a very long runtime)

  • Downloads and processes all TCGA data
  • Constructs patient-specific regulatory networks
  • Performs all downstream analyses

• Precomputed Workflow: Rapid reproduction using intermediate files (~4 hours)

  • Downloads precomputed expression data from Zenodo
  • Downloads precomputed network features from Zenodo
  • Peforms most of the downstream analysis (exluding network generation)
  • Focuses on statistical analysis and figure generation

Cancer Types Analyzed

33 TCGA cancer types: ACC, BLCA, BRCA, CESC, CHOL, COAD, DLBC, ESCA, GBM, HNSC, KICH, KIRC, KIRP, LAML, LGG, LIHC, LUAD, LUSC, MESO, OV, PAAD, PCPG, PRAD, READ, SARC, SKCM, STAD, TGCT, THCA, THYM, UCEC, UCS, UVM

Software Versions

• Snakemake: 7.23.1
• R: 4.2.1
• Python: 3.8+
• Key R packages: check package_requirements.txt. Container is on the way...

📊 Generated Figures

The pipeline generates all manuscript figures

📚 Citation

If you use PANGOLIN in your research, please cite:

@article{.....,
  title={Pan-cancer analysis of patient-specific gene regulatory landscapes identifies recurrent PD-1 pathway dysregulation},
  author={Belova et al.},
  journal={Journal Name},
  year={2025},
  doi={10.xxxx/xxxxx}
}