Subject: Clarification on WGCNA Module-Trait Correlations, Interpretation, and Functional Annotation
Hello All,
I recently began a postbac a few months ago and am working on a project involving bulk RNA-seq data from over 50 patient samples (~12,012 genes). My plan is to first correlate WGCNA module eigengenes to various clinical traits and then use the top WGCNA-to-clinical trait pairings to correlate with pre-defined modules for functional annotation. As I delve deeper into this analysis, I want to ensure that my understanding of certain key concepts and interpretations within WGCNA is accurate. I would greatly appreciate your insights on the following points:
Module Eigengenes:
My understanding is that a module eigengene represents the first principal component of the gene expression data for a given module, capturing the overall expression pattern of the genes within that module. Is this a correct and comprehensive understanding?
Signed vs. Unsigned Networks:
Signed Networks: In a signed network, the sign of the correlation between genes is preserved, meaning that modules consist of genes with consistent expression patterns (modules are made up of genes that are all positively correlated with one another, as per the WGCNA glossary). When correlating module eigengenes to clinical traits, a positive correlation suggests that an increase in the expression of the genes within that module is associated with an increase in the trait. Conversely, a negative correlation suggests that an increase in gene expression is associated with a decrease in the trait. Is it safe to make these statements regarding gene expression?
Unsigned Networks: In unsigned networks, the absolute value of the correlation is used, so both positive and negative correlations between genes are treated similarly. This could result in modules containing genes with opposite expression patterns. I assume that in this case, the sign of the correlation between a module eigengene and a clinical trait might be arbitrary and not reflect consistent gene behavior within the module.
Could you please confirm if my understanding of these network types and their implications is accurate? Are there any nuances or additional considerations that I should be aware of?
Correlating Module Eigengenes to Disease Activity Measures:
I am working with different types of disease activity measures, which have opposite scales. I’ve provided two examples below:
- Disease Activity Measure 1 (DAM1): A higher score indicates more severe disease activity, while a lower score indicates less disease activity.
- Disease Activity Measure 2 (DAM2): This measure is inversely scaled compared to DAM1, where a higher score indicates less severe disease activity (better health), and a lower score indicates more severe disease activity.
My Current Interpretation:
In a signed network:
- A positive correlation between a WGCNA module eigengene and DAM1 would indicate that as disease activity worsens (higher DAM1 score), the expression of the genes within that module increases.
- A negative correlation would suggest the inverse, where disease activity decreases (lower DAM1 score) as the expression of the genes within the module increases.
For DAM2:
- A positive correlation (without adjustment) would imply that as disease activity decreases (higher DAM2 score), the module’s gene expression increases, potentially suggesting involvement in recovery or protective processes.
- A negative correlation (without adjustment) would suggest that as disease activity worsens (lower DAM2 score), the module’s gene expression increases.
Concern:
Given that DAM2 is scaled oppositely to DAM1, interpreting the correlation directly could be counterintuitive. To maintain consistency, I’m considering correcting the sign of the correlation for DAM2 and other like it (by multiplying the correlation by -1). This adjustment would allow for a consistent interpretation across both measures:
- A positive correlation with both DAM1 and the adjusted DAM2 would indicate that the expression of the genes within that module increases with worsening disease activity.
- A negative correlation with both measures would suggest that the module is associated with improving disease activity.
Additional Plan for Functional Annotation:
After identifying the top WGCNA-clinical trait pairs, I plan to functionally annotate them by correlating these WGCNA module eigengenes to pre-defined modules (such as BloodGen3) that have established functions. My approach is to assess correlations between the WGCNA module eigengenes and the annotated module eigengenes. I assume I will need to carefully interpret the sign of the correlation in this context as well. Ideally, I am trying to generate a heatmap similar to Figure 1 here: https://pubmed.ncbi.nlm.nih.gov/29908154/
Clarification Needed:
- Is it appropriate to correct the sign of the correlation for DAM2 in this context? My aim is to ensure that the biological interpretation of the correlations remains consistent across both disease activity measures.
- What are your thoughts on the biological implications of the correlation before and after sign correction? Specifically, would the unadjusted correlation with DAM2 still provide meaningful insights, or is sign correction necessary for clear interpretation?
- Does my understanding of how to interpret module eigengenes and their correlation with clinical traits in both signed and unsigned networks align with best practices? Are there any additional nuances or best practices I should consider when conducting this analysis?
- A previous postbac in the lab mentioned the arbitrary assignment of correlation signs in WGCNA, but I don’t know why they claimed this. Could anyone clarify what this means? How should this influence the interpretation of correlations when using them for functional annotation against traits and pre-defined modules?
- Is there anything I said here that is completely wrong? I am still trying to piece things together about my project and WGCNA and any insight will help.
I want to ensure that I’m applying WGCNA correctly and interpreting the results in a biologically meaningful way. Any expertise and feedback would be invaluable in guiding this process.
Thank you!
Best regards,
[S]
