Fig. 2

False discovery rate of WGCNA modules and graphical lasso networks based on canonical pathways (a–f). The density of networks inferred from PC-corrected data is sparser (g–l). a–c FDR of WGCNA networks obtained at varying cut heights. Each point corresponds to FDR of the network obtained at a specific cut height. Each color represents networks reconstructed with a specific correction approach. d–f Each point in the figure corresponds to false discovery rates of networks obtained at a specific L1 penalty parameter value (lambda) in the graphical lasso. Each color represents networks reconstructed with a specific correction approach—uncorrected, multi-covariate, RIN, and PC corrected. g–i Each point corresponds to a number of edges in networks inferred by WGCNA at a cut height. j–l Each point corresponds to a number of edges inferred by graphical lasso in networks obtained at a specific L1 penalty parameter value. Networks inferred by PC-corrected data have fewer edges compared to uncorrected or RIN-corrected data