[Paper Review] MODA: MOdule Differential Analysis for weighted gene co-expression network

Dong Li, James Brown+4 more|arXiv (Cornell University)|May 16, 2016

Bioinformatics and Genomic Networks5 references4 citations

TL;DR

This paper introduces MODA, a novel method for module differential analysis in weighted gene co-expression networks that identifies condition-specific and conserved modules by comparing condition-specific networks using Jaccard similarity. It improves WGCNA by optimizing module cutting height via density maximization, uses a sample-saving approach for reliable correlation estimation, and integrates functional enrichment to link modules to biological processes, validated on synthetic and Daphnia magna data.

ABSTRACT

Gene co-expression network differential analysis is designed to help biologists understand gene expression patterns under different condition. By comparing different gene co-expression networks we may find conserved part as well as condition specific set of genes. Taking the network as a collection as modules, we use a sample-saving method to construct condition-specific gene co-expression network, and identify differentially expressed subnetworks as conserved or condition specific modules which may be associated with biological processes. We have implemented the method as an R package which establishes a pipeline from expression profile to biological explanations. The usefulness of the method is also demonstrated by synthetic data as well as Daphnia magna gene expression data under different environmental stresses.

Motivation & Objective

• To address the limitations of traditional gene co-expression network analysis in detecting biologically meaningful subnetworks under different conditions.
• To overcome the instability of correlation estimation with small sample sizes by using a sample-saving method for condition-specific network construction.
• To shift from individual gene differential analysis to module-level differential analysis, identifying subnetworks that are conserved or condition-specific.
• To provide a systematic pipeline from gene expression profiles to functional interpretation via module identification and enrichment analysis.

Proposed method

• Uses weighted correlation network analysis (WGCNA) with edge weights derived from gene pair correlations, preserving all correlation information.
• Introduces a novel module cutting strategy that maximizes average module density in hierarchical clustering, replacing subjective threshold selection.
• Applies a sample-saving approach to construct condition-specific networks, improving reliability of correlation estimates with limited replicates.
• Employs the Jaccard index to compute similarity between modules from two networks (one background, one excluding a specific condition), forming a similarity matrix.
• Defines condition-specific modules as those with low row sums in the similarity matrix (low overlap with background), and conserved modules as those with high row sums.
• Integrates functional enrichment analysis using DAVID via an R web service interface to link identified modules to biological processes.

Experimental results

Research questions

• RQ1 Can a module-level differential analysis detect biologically relevant subnetworks that are conserved or condition-specific across experimental conditions?
• RQ2 How can module cutting height in hierarchical clustering be objectively determined to maximize module quality in weighted networks?
• RQ3 To what extent does the sample-saving method improve the reliability of correlation estimation in low-sample-size conditions?
• RQ4 How effective is the Jaccard similarity-based module comparison in identifying condition-specific modules compared to individual gene differential analysis?
• RQ5 Can the proposed pipeline successfully link identified modules to meaningful biological functions through functional enrichment?

Key findings

• The method successfully identified modules containing genes 1–100, 101–200, and 201–300 as highly overlapping between networks, consistent with the simulation design where these groups were preserved.
• The module containing genes 301–500 showed minimal overlap with the second network, correctly reflecting that it was merged into a single group in the second dataset.
• The sample-saving approach enabled stable and reliable correlation estimation even with limited replicates, improving module detection robustness.
• The density-optimized cutting criterion outperformed subjective thresholding in identifying biologically coherent modules with higher internal connectivity.
• Functional enrichment analysis successfully linked identified modules to relevant biological processes, demonstrating the pipeline’s utility in biological interpretation.
• The R package MODA provides a complete, reproducible pipeline from expression data to functional annotation, with visualization and threshold-based classification of modules.