Fig. 1

Overview of our evaluation methodology. a The nine different datasets used in this evaluation. b We used three different module definitions to extract known modules from known regulatory networks for the evaluation on E. coli, yeast and synthetic data. c To avoid parameter overfitting on characteristics of particular datasets, we first optimized the parameters on every dataset using a grid search, and then used the optimal parameters on one dataset (training score) to assess the performance of a method on another dataset (test score). d We evaluated a total of 42 methods, which can be classified in 5 categories: clustering, biclustering, direct network inference (NI), decomposition, and iterative NI. e For the evaluation on human data, we compared how well the targets of each regulator is enriched in at least one of the modules. f We used four different regulatory networks in our evaluation, each generated from different types of data