edgeR

Overview

Perform Differential expression using Quasi-Likelihood Pipeline, run glmQLFit() and glmQLFTest().

Version: latest
Language: r

Description

edgeR is a GenePattern module that performs differential gene expression analysis using the edgeR Bioconductor package's quasi-likelihood (QL) pipeline. This module implements a robust statistical framework for identifying differentially expressed genes in RNA-seq count data by modeling overdispersion through quasi-likelihood methods. The QL approach provides better error rate control than likelihood ratio tests, especially for experiments with small sample sizes.

The module runs the complete edgeR QL pipeline including:

• Data normalization and filtering
• Dispersion estimation with trend modeling
• Quasi-likelihood fitting (glmQLFit)
• Statistical testing (glmQLFTest)
• Multiple testing correction

Parameters

Required Parameters

count.matrix.file (FILE)

• Purpose: Input count matrix file containing raw RNA-seq read counts
• Format: Tab-delimited (.txt, .tsv), comma-separated (.csv), or GCT format
• Structure: Genes as rows, samples as columns with integer count values
• Requirements:
  • Must contain raw, unnormalized integer counts
  • Sample names in header must match those in sample groups file
  • Gene identifiers should be in first column or row names

sample.groups.file (FILE)

• Purpose: Define experimental groups for comparison
• Format: Two-column tab-delimited text file
• Structure:
  • Column 1: Sample IDs (exactly matching count matrix column names)
  • Column 2: Group assignments (factor levels for comparison)
• Example:

  Sample	Group
  Sample1	Control
  Sample2	Control  
  Sample3	Treatment
  Sample4	Treatment
  

comparison.groups (TEXT)

• Purpose: Specify which groups to compare for differential expression
• Format: Comma-separated pair of group names
• Example: "Treatment,Control" (compares Treatment vs Control)
• Note: First group is numerator, second is denominator for fold change calculation

Optional Parameters

output.prefix (TEXT)

• Purpose: Customize output file naming
• Default: "edgeR"
• Usage: All output files will use this prefix (e.g., "MyExp_results.txt")

min.count (INTEGER)

• Purpose: Minimum count threshold for gene filtering
• Default: 1
• Rationale: Removes very lowly expressed genes that provide little statistical power
• Range: ≥ 0 (typically 1-10)

min.samples (INTEGER)

• Purpose: Minimum samples requiring expression above min.count threshold
• Default: 2
• Rationale: Genes must be expressed in multiple samples to be considered
• Usage: Gene retained if ≥ min.samples have counts ≥ min.count

norm.method (CHOICE: TMM, RLE, upperquartile, none)

• Purpose: Normalization method for library size adjustment
• Default: TMM (Trimmed Mean of M-values)
• Options:
  • TMM: Recommended for most RNA-seq experiments
  • RLE: Relative Log Expression (similar to DESeq2)
  • upperquartile: Use upper quartile for normalization
  • none: Skip normalization (not recommended)

robust.estimation (CHOICE: true, false)

• Purpose: Use robust methods for dispersion estimation
• Default: false
• Usage: Set to true if dataset may contain outlier genes that could skew dispersion estimates
• Impact: More conservative but stable dispersion estimates

abundance.trend (CHOICE: true, false)

• Purpose: Model abundance-dependent trend in QL dispersion
• Default: true
• Rationale: Accounts for the relationship between gene expression level and variance
• Recommendation: Generally keep as true unless specifically advised otherwise

poisson.bound (CHOICE: true, false)

• Purpose: Apply Poisson bound to QL p-values
• Default: true
• Function: Prevents over-dispersion from making p-values unrealistically small
• Recommendation: Keep true for proper error rate control

fdr.threshold (FLOAT)

• Purpose: False Discovery Rate threshold for significance
• Default: 0.05
• Range: 0.0 to 1.0
• Usage: Genes with FDR ≤ this value are considered significantly differentially expressed

fold.change.threshold (FLOAT)

• Purpose: Minimum absolute log2 fold change for biological significance
• Default: 1.0 (equivalent to 2-fold change)
• Range: ≥ 0.0
• Usage: Combined with FDR threshold to define significant genes

top.genes (INTEGER)

• Purpose: Number of top DE genes to highlight in outputs and plots
• Default: 50
• Range: > 0
• Usage: Affects summary tables and plot annotations

create.plots (CHOICE: true, false)

• Purpose: Generate diagnostic and visualization plots
• Default: true
• Plots Generated:
  • MA plot (log2FC vs average expression)
  • Volcano plot (log2FC vs -log10(p-value))
  • Dispersion plot (biological coefficient of variation)
  • PCA plot (sample relationships)

Usage Examples

Basic Differential Expression Analysis

Scenario: Compare treated vs control samples

Input Files:
- count_matrix.txt (gene counts across samples)
- sample_info.txt (sample-to-group mapping)

Parameters:
- comparison.groups: "Treated,Control"  
- fdr.threshold: 0.05
- fold.change.threshold: 1.0

Stringent Analysis with Custom Filtering

Scenario: High-stringency analysis with custom thresholds

Parameters:
- min.count: 5
- min.samples: 3
- fdr.threshold: 0.01
- fold.change.threshold: 1.5
- robust.estimation: true

Large-Scale Study

Scenario: Analysis of large cohort with many samples

Parameters:  
- norm.method: TMM
- robust.estimation: true
- top.genes: 100
- create.plots: true

Output Files

Primary Results

• {prefix}_results.txt: Complete differential expression results for all tested genes
• {prefix}_significant.txt: Filtered results containing only significantly DE genes
• {prefix}_summary.txt: Analysis summary with key statistics

Diagnostic Plots (if create.plots=true)

• {prefix}_MA_plot.pdf: MA plot showing log2FC vs average expression
• {prefix}_volcano_plot.pdf: Volcano plot showing log2FC vs significance
• {prefix}_dispersion_plot.pdf: Dispersion estimates across expression levels
• {prefix}_PCA_plot.pdf: Principal component analysis of samples

Results File Columns

• Gene: Gene identifier
• logFC: Log2 fold change (positive = upregulated in first group)
• logCPM: Log2 counts per million (average expression)
• F: F-statistic from QL test
• PValue: Uncorrected p-value
• FDR: False discovery rate (Benjamini-Hochberg adjusted p-value)

Computational Requirements

• Memory: Minimum 4GB RAM (8GB+ recommended for large datasets)
• Runtime: Typically 5-30 minutes depending on dataset size
• Disk Space: ~100MB for outputs plus input file sizes

Troubleshooting

Input Data Issues

"Sample names don't match between files"

• Verify sample IDs in count matrix headers exactly match those in sample groups file
• Check for spaces, special characters, or case sensitivity differences

"No genes pass filtering criteria"

• Reduce min.count or min.samples parameters
• Check if count data contains very low counts across most genes
• Verify count matrix contains raw integer counts, not normalized values

"Insufficient samples for analysis"

• Ensure each comparison group has at least 2 samples
• Check sample groups file for correct group assignments

Analysis Issues

"Dispersion estimation fails"

• Try setting robust.estimation=true
• Check for extreme outlier samples or genes
• Verify adequate biological replication within groups

"No significantly DE genes found"

• Increase FDR threshold (e.g., 0.1 instead of 0.05)
• Reduce fold change threshold
• Check if experimental groups are well-separated
• Consider increasing sample sizes if possible

"Memory errors during analysis"

• Reduce dataset size by pre-filtering lowly expressed genes
• Increase available system memory
• Consider analyzing subsets of samples separately

Output Issues

"Plot generation fails"

• Set create.plots=false to skip plotting
• Check available disk space for output files
• Ensure R graphics packages are properly installed

Statistical Considerations

When to Use edgeR QL Pipeline

• Recommended for: Small to moderate sample sizes (n=3-20 per group)
• Best for: Experiments where precise error rate control is critical
• Advantages: Conservative approach, good performance with limited replication

Comparison with Other Methods

• vs DESeq2: edgeR QL often more conservative, similar performance overall
• vs limma-voom: QL pipeline specifically designed for count data
• vs classic edgeR: QL methods provide better error rate control

Design Considerations

• Replication: Minimum 3 biological replicates per group recommended
• Batch Effects: Consider adding batch variables to sample groups file
• Paired Samples: For paired designs, include pairing information in groups file

References

1. Robinson MD, McCarthy DJ, Smyth GK (2010). "edgeR: a Bioconductor package for differential expression analysis of digital gene expression data." Bioinformatics 26, 139-140.

2. McCarthy DJ, Chen Y, Smyth GK (2012). "Differential expression analysis of multifactor RNA-Seq experiments with respect to biological variation." Nucleic Acids Research 40, 4288-4297.

3. Lun ATL, Chen Y, Smyth GK (2016). "It's DE-licious: A Recipe for Differential Expression Analyses of RNA-seq Experiments Using Quasi-Likelihood Methods in edgeR." Methods in Molecular Biology 1418, 391-416.

4. Chen Y, Lun ATL, Smyth GK (2016). "From reads to genes to pathways: differential expression analysis of RNA-Seq experiments using Rsubread and the edgeR quasi-likelihood pipeline." F1000Research 5, 1438.

Version History

• latest - Current implementation with quasi-likelihood pipeline
• Support for multiple normalization methods
• Enhanced plot generation and diagnostic outputs
• Improved parameter validation and error handling