Consensus Clustering Analysis

Source: https://github.com/aipoch/medical-research-skills

Consensus Clustering Analysis

When to Use

Use this skill when you need to identify stable sample subtypes from a bulk expression matrix with ConsensusClusterPlus, compare candidate clustering settings with PAC, and export consensus matrix/CDF visualizations.

Do not use this skill for differential expression analysis, single-cell clustering, or non-expression tabular data.

When to Read External Files

Usage

bashCopy code
Rscript scripts/main.R \
  --input_file ./expression_matrix.csv \
  --group_file ./groups.csv \
  --disease_group case \
  --max_k 4 \
  --output_dir ./output/ \
  --gene_selection highly_variable \
  --top_n 5000 \
  --reps 1000 \
  --p_item 0.8 \
  --p_feature 1.0 \
  --timeout_seconds 3600 \
  --seed 42

Arguments

Input Format

Expression Matrix (input_file)

Genes as rows, samples as columns, CSV/TSV/TXT format with gene ID in the first column.

csvCopy code
,Sample01,Sample02,Sample03
TSPAN6,1.8479,1.8318,3.8276
TNMD,0.0349,0.0533,1.3889

Group File (group_file)

Delimited text file with sample ID and group columns.

csvCopy code
sample,group
Sample01,case
Sample02,control
Sample03,case

Gene List (gene_list)

Optional plain text or single-column CSV file with one gene symbol per line.

csvCopy code
TNMD
DPM1
SCYL3

Output Files

Workflow

Step 1: Validate Input

• Check file existence
• Detect sample and group columns in the group file
• Validate sample matching between expression matrix and group file

Step 2: Prepare Clustering Matrix

• Filter samples by the requested disease group
• Select genes using highly_variable or custom
• Median-center genes if requested

Step 3: Run Consensus Clustering

• Evaluate supported distance and clustering algorithm combinations
• Compute PAC scores across candidate K values
• Select the optimal model by minimum PAC

Step 4: Generate Outputs

• Save result tables
• Generate consensus matrix and CDF plots
• Record session information for reproducibility

Methods

ConsensusClusterPlus

Repeated subsampling is used to estimate cluster stability across candidate K values and clustering settings.

PAC Score

The proportion of ambiguous clustering is computed as CDF(0.9) - CDF(0.1) from lower-triangle consensus values. Lower PAC indicates more stable clustering.

Gene Selection

• highly_variable: rank genes by median absolute deviation
• custom: use the intersection of the provided gene list and matrix row names

Examples

Basic Usage

bashCopy code
Rscript scripts/main.R \
  -i expression_matrix.csv \
  -g groups.csv \
  -d case \
  -k 3 \
  -r 20 \
  -o output/example_basic \
  -t 120

With a Custom Gene List

bashCopy code
Rscript scripts/main.R \
  -i expression_matrix.csv \
  -g groups.csv \
  -d case \
  -m custom \
  -l genes.csv \
  -k 4 \
  -r 20 \
  -o output/example_custom \
  -t 120

Without Median Centering

bashCopy code
Rscript scripts/main.R \
  -i expression_matrix.csv \
  -g groups.csv \
  -d case \
  -c FALSE \
  -k 3 \
  -r 20 \
  -o output/example_rawscale \
  -t 120

Error Handling

Common Errors

IF error persists, READ: references/troubleshooting.md

Testing

Smoke Check

bashCopy code
# Check help
Rscript scripts/main.R --help

# Run analysis
Rscript scripts/main.R \
  -i tests/data/expression_matrix.csv \
  -g tests/data/groups.csv \
  -d case \
  -k 3 \
  -r 20 \
  -o output/example_basic \
  -t 120

Validation Commands

bashCopy code
# Inspect selected model
cat output/example_basic/Cluster_res.csv

# Check output plots exist
ls -la output/example_basic

Implementation Checklist

• CLI parsing with optparse
• set.seed() for reproducibility
• requireNamespace() dependency checks
• Session info recording
• data.table::fread() input reading
• File reading instructions in SKILL.md
• Modular script structure (<150 lines per file)
• Test data provided
• Error handling with SKILL_* codes
• Scripts in scripts/ directory
• References in references/ directory

Last updated: 2026-04-17 | Version: 1.0.0