Fig. 1
From: Comparison of confound adjustment methods in the construction of gene co-expression networks
A Distribution of gene-gene correlations for 5000 randomly selected genes in the skeletal muscle tissue dataset. We observe significant differences in distribution of gene-gene correlations between adjustment methods across tissues (Pairwise K-S test D = 0.018–0.361, all p < 0.0001). B Box plots showing module size for each module detection method. There is a significant difference in module size between adjustment methods for WGCNA, MEGENA, and ICA-derived modules (all Kruskal-Wallis test p < 0.0001). C Box plots showing the total number of modules detected. p-values are provided in figure for significant pairwise Tukey HSD tests. D Box plots showing intramodular density across confound adjustment methods and tissue datasets. Similar to module connectivity, module density measures how tight or cohesive genes are within a group, and is equal to the mean adjacency of a module [22]. Unlike MEGENA and WGCNA, ICA is not a clustering module detection method rooted in the pairwise similarity between genes; therefore, intramodular density was not calculated for ICA-derived modules. Outlier points are omitted for ease of visualization in panels B and D