Hi all,

My colleague and I are working with bulk RNA-seq (and some single-cell data using hdWGCNA) and constructing co-expression modules. For each module, I summarize expression using the first principal component (module eigengene).

The dataset is normalized.

My question is about comparing eigengenes derived from different gene clusters after running wgcna/hdwgcna.

Since each module contains a different set of genes, the PCA is performed in different feature spaces. In that sense, the resulting PC1s (eigengenes) are not directly comparable as axes.

Is it valid to:

Correlate two different module eigengenes? Compare their association with phenotypes (e.g., disease and control)? Use them to assess whether two modules co-vary or represent related biological processes?

Or would comparing eigengenes across different gene sets be considered inappropriate due to differences in underlying gene composition and variance structure?

I’m particularly interested in best practices for interpreting such comparisons in the context of Weighted Gene Co-expression Network Analysis and hdWGCNA.

Thanks in advance!