Fig. 5
From: Module-based regularization improves Gaussian graphical models when observing noisy data
Application of Standard and Modular GLASSO to gene co-expression data. The Standard GLASSO (log-likelihood) and Modular GLASSO (codelength savings) suggest vastly different regularization strengths also for the gene co-expression data (a). The Standard GLASSO’s minimal regularization leads to a network with little modular structure (b). In contrast, Modular GLASSO disconnects nodes to maximize the modular structure during cross-validation, revealing more regularities in the underlying system (c)