Scalable learning of large networks

From Q-bio

Structure inference of cellular networks from microarray experiments provides important insight about the network of interactions in cells under different conditions. Unfortunately, many structure inference algorithms do not scale to whole-genome data that have several thousands of variables. We propose a two-step approach for learning the structure of large networks. We first pre-cluster the network nodes and learn networks per cluster. We then revisit the cluster assignment of selected variables with poor neighborhoods. Preliminary results suggest that our approach performs at par to approaches that learn the complete network without pre-clustering.