Cross-validation of correlation networks using modular structure

Correlation networks derived from multivariate data appear in many applications across the sciences. These are usually dense and require sparsification to detect meaningful structure. However, current methods for sparsifying correlation struggle with balancing overfitting underfitting. We propose a module-based cross-validation procedure threshold these networks, making modular structure an integral part of thresholding. illustrate our approach using synthetic real find that its ability recover planted partition has step-like dependence on number samples. The reward sampling more varies non-linearly samples, minimal gains after critical point. A comparison well-established WGCNA method shows allows revealing used here.