A Complete Guide to WGCNA: Applications in Correlation Network Analysis

by Antonino Zito
⏱ 10 min read

Introduction

Weighted gene co-expression network analysis (WGCNA) is a powerful all-in-one analysis method that allows biologists to understand the system-wide relationships of all measured features (genes, proteins, or metabolites) rather than each one in isolation.  

With WGCNA, researchers can identify clusters of genes, proteins, or metabolites (called modules) that share correlated expression or abundance patterns and explore how these clusters relate to one another. Importantly, WGCNA also provides data on the association between modules and external traits, such as recorded sample phenotypes.  Identification of correlation networks has high biological relevance as features within the same module could share regulatory mechanisms and be functionally related within a molecular pathway at the cellular and inter-cellular level (1-4). 

Ultimately, WGCNA could inform on candidate biomarkers and druggable targets. Although WGCNA was first applied to transcriptomic data, its principles are equally suited to other omics, such as proteomics, metabolomics, and methylation data.

So, how does WGCNA work in practice? What insights can you expect from the approach in your next RNA-seq, proteomics, or metabolomics study, and what are its limitations?

What is WGCNA?

WGCNA is a systems biology approach that researchers use to analyze complex data patterns in large numbers of samples (1). WGCNA is split into four main sequential analytical components:

  1. Construction of weighted correlation networks.
  2. Identification of coexpression or coabundance modules.
  3. Association of features with sample traits
  4. Inference of intramodular hub features as candidate drivers of phenotypes 

How does WGCNA work?

 WGCNA determines these outcomes by pairwise correlations between features or modules in a guilty-by-association approach, where information about a feature is gained from its close neighbors in the network. However, there are many options and outputs, which could seem overwhelming to biologists with limited knowledge of the analyses performed and the plots generated. 

Here, we break down the WGCNA method, the plots, and their interpretation into bite-size pieces for biologists with limited bioinformatic expertise.

The four main steps of WGCNA

Step 1: Construct weighted correlation networks of features

Typically, WGCNA begins with a matrix of data that contains the expression or abundance of each gene, protein, or metabolite in each sample. The method then measures pairwise correlations between features across all samples.

The correlation score of each feature pair indicates the similarity of their expression or abundance pattern and could suggest their potential functional relationship.

The ‘weighted’ aspect of WGCNA aims to amplify the differences between strong and weak correlations by raising the correlation to a power defined by the user (don’t worry, the Omics Playground takes care of selecting the most appropriate power value for the user). A high correlation indicates the features are strongly connected, whereas a low correlation suggests a weak connection.

These magnified weighted correlation values then make it easier to identify groups of features with similar behaviors in the subsequent step. The user must then select the type of network required.

The result is a network plot where features are depicted by circles (known as nodes), and the strength of the weighted correlation coefficient is shown by the thickness of a line (or edge) connecting two features. 

In the plot below, the thick green line between SVOP and AMER3 indicates a strong correlation and, therefore, potential association between these two features (Fig. 1).

Gene correlation network in Omics Playground
Figure 1. Example of gene correlation network indicating connected genes as nodes (circles). The thickness of lines (edges) connecting two nodes represents the strength of their correlation. Green edges represent positive correlations. Orange edges represent negative correlation.

Step 2: Divide the network into modules

Next, WGCNA uses the network’s weighted correlation coefficient information to place features exhibiting significantly similar expression or abundance profiles into groups called modules.

If features have similar correlations with many shared neighbors in the network or have a large overlap of their network neighbors, the features likely have similar expression or abundance patterns and can be grouped into the same module. 

To determine modules, hierarchical clustering is performed on the correlation network data. A dendrogram is generated where each branch identifies a specific module (Fig. 2). 

Methods like dynamic tree cut can be employed to determine discrete modules containing features with similar expression or abundance patterns. Each module is assigned a distinct ID and color. 

Be cautious when setting parameters ⚠️

The way you “cut” the dendrogram influences the size and number of modules and, if done incorrectly, the clustering can become misleading and reduce biological accuracy.

These modules serve as the foundation for the next step, where WGCNA examines how feature clusters correlate with phenotypic traits.

Dendogram WGCNA analysis in Omics Playground
Figure 2. Dendrogram to identify modules with dynamic tree cut algorithm. A different color represents each module.

Step 3: Correlate phenotypic traits with different feature modules

Once modules are defined using the dendrogram, the output must be simplified to one value per module, called the module eigengene. The eigengene is the first component from a principal component analysis and represents the overall module expression or abundance. 

As the module eigengene characterizes each module as a singular entity, it enables us to perform correlation analysis between modules to find those with similar expression or abundance behaviors or to determine how each module correlates with phenotypes. 

For instance, our example below shows that the module eigengenes of ME1 and ME4 are highly correlated, suggesting their biological function or sample types of origin might be related (Fig. 3A). 

To determine whether these two modules do have similar biological roles, we can next measure the degree to which each module’s eigengene correlates to different patient traits, sample types, or disease outcomes. These biological variables could include a patient’s age, gender, or weight, outcomes like remission or patient death, or whether samples originate from healthy or disease patients or from different organs or tumor locations. Researchers have used this approach to identify key modules in many diseases, such as glioblastoma, breast, and colorectal cancer (2, 3, 4).  

In our example, ME1 and ME4 both highly correlate to healthy samples, whereas ME2 and ME3 are highly correlated with glioblastoma samples (Fig. 3B). This suggests that the features contained in these modules could play a role in glioblastoma. 

We can get an overview of this possibility by performing gene set or pathway enrichment analysis (e.g., gene ontology) to determine modules enriched for features associated with particular pathways or functions. 

But, from this, we don’t yet know which features within each module might be the most important.

WGCNA heatmap
Figure 3. (A) Dendrogram and heatmap generated with the module eigengenes for ME0 to ME4. (B) Heatmap correlating phenotypic traits with module eigengenes. Deep red indicates a high correlation and deep blue indicates a low correlation between A and B.

Step 4: Identify potential driver features

Finally, once we have identified modules of interest, we can delve deeper into each module to find features that might be key factors for a particular trait or could influence other features in that module. Each module may contain many features; it is essential to identify so-called ‘hub features’ that can be ideal candidates for further study.

Hub features are identified as the most highly connected features within a module and, expectedly, the most strongly correlated with the phenotype of interest. The expression or abundance of a feature is also used to calculate the module membership, which measures the degree to which a feature’s profile correlates with a particular module within the network. Module membership is therefore a useful tool for prioritizing features for further study.

If the correlation is high, the feature is likely representative of the overall expression or abundance of the module as a whole and is well connected in the network. Similarly, the high correlation of this feature to the trait of interest further strengthens its likelihood as an important driver in that module.

Be aware of the limitations of WGCNA

While it is a powerful approach, many parameters in WGCNA can present problems to the user if not applied correctly. 

For instance, before generating the correlation networks, users must choose, among many other options:

  • Network type (signed vs. unsigned)

  • Correlation method (Pearson, Spearman, or others)

  • Soft-thresholding power values to weight the correlations

  • Cut-offs for defining modules

The wide swathe of options and parameters needed to conduct an end-to-end WGCNA could make the analyses highly error-prone. In fact, selecting an inappropriate method, parameter, or threshold for the type or spread of your data could lead to misinterpretation of correlations where outliers aren’t treated correctly, networks that aren’t biologically realistic, and ultimately inaccurate conclusions that could hinder future research.

Breaking barriers with Omics Playground

Another major problem many biologists face is that WGCNA is available predominantly for those with coding knowledge. For biologists unfamiliar with programming languages, this barrier can often seem insurmountable, regardless of the vast biological insights possible with WGCNA.

To overcome this hurdle, our Omics Playground is designed as a point-and-click entry point that allows scientists with no knowledge of coding to explore all the functionalities and parameters of WGCNA in an interactive and efficient way. 

How to access the WGCNA module in Omics Playground

To view the WGCNA module in Omics Playground, you need to select the option in the computation settings when you first upload your dataset. Below you can find a step-by-step guide to enable the module and access it when your dataset is loaded.

1. Register or Log In to Omics Playground

If you’re new to Omics Playground you can easily access the platform by registering for a trial account. The platform is designed to be as user-friendly as possible. However, to ensure a successful first upload, we recommend reading the guidelines on data preparation before proceeding.

If you already have an account, simply log in using your credentials.

2. Upload your new dataset

When you upload a new dataset, you will be prompted to follow these steps:

  1. Upload your counts or abundance matrix.
  2. Upload your samples.
  3. Select your comparisons.
  4. Name your dataset and select any additional computation options.
At step 4, make sure the WGCNA option is selected in the ‘Extra analysis’ box in the computation options section to activate the module for this dataset (see Figure 4).
Extra analysis options in Omics Playground showing where to enable the WGCNA analysis module.
Figure 4. WCGNA selection in 'Extra analysis' box in Omics Playground.

Once you’ve ensured that all your options are selected, click ‘Compute’ and wait for the dataset to be ready.

For more information on how to upload your data to Omics Playground, feel free to check our step-by-step uploading guide.

3. Go to the WGCNA module and explore your data

When your dataset is ready you’ll receive both an email and in-app notification. Once loaded, you can find the WGCNA module under Menu > SystemsBio > WGCNA (see Figure 5).

WGCNA analysis module location in the Omics Playground menu
Figure 5. WGCNA module location in Omics Playground.


You can then select different methods, parameters, and thresholds to fully explore your data using the settings menu on the right-hand side of your dashboard (Figure 6).

Select options and parameters for your WGCNA analysis
Figure 6. WGCNA options in Omics Playground.

Now that you can easily access the WGCNA module, you can start exploring the complex relationships within your dataset and uncovering new biological insights.

Overview of WGCNA Key Visualizations in Omics Playground

Tab 1: WGCNA (Module Detection)

This is where the magic happens: features are grouped into modules.

WGCNA Tab 1 in Omics Playground BigOmics Analytics
Figure 7. WGCNA Tab in Omics Playground

Key Visualizations:

(a) Dendrogram and modules. Shows a tree-like diagram where similar features are clustered together. The colored bar at the bottom shows which module each feature belongs to (turquoise, blue, brown, etc.).

(b) Scale independence and mean connectivity. Technical plots that help determine if the analysis parameters are appropriate. You don’t need to interpret these in detail; they’re mainly for quality control.

(c) TOM heatmap. A red-orange colored heatmap showing how strongly features are connected to each other. Darker red squares indicate features that are highly correlated.

(d) Feature UMAP. A scatter plot where each dot represents a feature, colored by its module assignment. Features in the same module cluster together spatially.

(e) Module size. A bar chart showing how many features are in each module. This helps you understand the scale of each module.

Tab 2: Eigengenes (Module Patterns)

An “eigengene” represents the average expression or abundance pattern of all features in a module. It’s like a summary or representative profile for that module.

WGCNA Tab 2 in Omics Playground BigOmics Analytics
Figure 8. Eigengenes Tab in the WGCNA module in Omics Playground

Key Visualizations:

(a) Module-Trait relationships. Shows which modules are associated with your experimental conditions (e.g., disease vs. control, different treatments). Modules with strong correlations to your traits are the most biologically relevant.

(b) Sample dendrogram + eigengenes. Shows how your samples cluster together, with module eigengene patterns displayed as heatmaps below.

(c) Sample dendrogram + traits. Similar to (b) but showing your experimental traits instead of eigengenes.

(d) Eigengene correlation heatmap. Shows which module eigengenes are correlated with each other. This can reveal higher-order relationships between modules.

(e) Eigengene dendrogram. Shows how module eigengenes cluster together based on their similarity.

(f) Module graph. A network visualization showing relationships between modules.

Tab 3: Modules (Detailed Module Analysis)

This tab lets you dive deep into individual modules to identify “hub features”, the most important features driving each module’s pattern.

WGCNA Tab 3 in Omics Playground BigOmics Analytics
Figure 9. Modules Tab in the WGCNA module in Omics Playground

Key Visualizations:

(a) Summary. Basic information about the selected module.

(b) Trait correlation. Shows how strongly the selected module’s eigengene correlates with different experimental traits. Higher bars indicate stronger associations.

(c) Circle network of hub features. A network diagram showing the most connected features (hub features) in the module. These are often the most biologically important features.

(d) Significance table. Lists features in the module ranked by their importance scores. Higher scores indicate features more central to the module.

(e) Feature significance. Small scatter plots showing the relationship between module membership and feature-trait correlation for individual features.

Tab 4: Enrichment (Biological Interpretation)

This tab tells you what biological functions or pathways are enriched in each module, helping you understand what the module actually does.

Key Visualizations:

(a) Feature-set heatmap. Shows which feature sets (pathways, GO terms, etc.) are enriched in the module. Red indicates strong enrichment.

(b) Feature heatmap. Shows expression or abundance patterns of features in the most enriched feature sets.

(c) Enrichment scores. A table listing the most significantly enriched biological terms, ranked by score. This is where you find out if your module is involved in “immune response,” “cell cycle,” etc.

(d) Feature frequency. A bar chart showing which features appear most frequently in the enriched feature sets.

WGCNA Tab 4 in Omics Playground BigOmics Analytics
Figure 10. Enrichment Tab in the WGCNA module in Omics Playground

How to Interpret Your WGCNA Results in Omics Playground: A Step-by-Step Workflow

1. Start with the WGCNA tab. Look at how many modules were detected and their sizes. For transcriptomics datasets, modules with 50-500 features are usually most interpretable.

2. Go to Eigengenes tab. Check the Module-Trait relationships heatmap. Identify modules that show strong correlations (positive or negative) with your experimental conditions.

3. Select interesting modules. Focus on modules that correlate with your traits of interest (from step 2).

4. Go to Modules tab. For each interesting module, examine the hub features in the network and check the significance table to identify key driver features.

5. Go to Enrichment tab. Look at what biological functions are enriched in your module. This tells you what biological process this feature group is involved in.

6. Biological interpretation. Combine the information: “Module X (e.g., MEblue) contains 200 features (e.g., genes in an RNA-seq dataset) that are upregulated in disease samples. The hub genes include IFIT2, IFIT3, and HERC5. This module is enriched for interferon response and immune activation.”

Bonus Tips

  • Module colors are arbitrary. The names like “turquoise,” “blue,” or “brown” are just labels and don’t have biological meaning.
  • Focus on correlated modules. Not all modules will be relevant to your experiment. Use the Module-Trait relationships to prioritize.
  • Hub features are key. These highly connected features often represent key regulators worth investigating further.
  • Enrichment gives biological context. A module is only useful if you can understand what biological process it represents. Always check the enrichment results.
  • Module size matters. For typical transcriptomics datasets, very small modules (<30 genes) may be noise. Very large modules (>1000 genes) may be too general to be useful.

In addition to WGCNA, Omics Playground offers a wide range of other analytical tools and modules, allowing you to continue your research and dive deeper into your data. Start exploring today and discover the many ways Omics Playground can enhance your biological discoveries.

Access WGCNA analysis and much more in Omics Playground.

References

  1. Langfelder P, Horvath S. WGCNA: an R package for weighted correlation network analysis. BMC Bioinformatics. 2008 Dec;9:1-3.
  2. Wang LB, Karpova A, Gritsenko MA, Kyle JE, Cao S, Li Y, Rykunov D, Colaprico A, Rothstein JH, Hong R, Stathias V. Proteogenomic and metabolomic characterization of human glioblastoma. Cancer Cell. 2021 Apr 12;39(4):509-28.
  3. Tian Z, He W, Tang J, Liao X, Yang Q, Wu Y, Wu G. Identification of important modules and biomarkers in breast cancer based on WGCNA. OncoTargets and Therapy. 2020 Jul 12:6805-17.
  4. Ghafouri-Fard S, Safarzadeh A, Taheri M, Jamali E. Identification of diagnostic biomarkers via weighted correlation network analysis in colorectal cancer using a system biology approach. Scientific Reports. 2023 Aug 21;13(1):13637.

About the Author

Antonino Zito

Antonino is a senior bioinformatics engineer at BigOmics with a strong background in bioinformatics and biostatistics. With a PhD in genetics and bioinformatics and an MSc in biotechnology, he has made significant contributions to computational analysis in numerous projects during his previous research at Harvard Medical School and King’s College London.