Fig. 1

Thresholding correlation networks. The researcher must solve a model selection problem when thresholding a correlation network, here exemplified by gene co-expression data from the plant Arabidopsis thaliana. As the threshold increases, more communities appear, potentially leading to overfitting (left). Including too many links (low threshold) can lead to underfitting (middle). Our module-based cross-validation method guides the researcher to the most parsimonious model (right)