Bayesian methods have found many successful applications in high-throughput genomics. We focus on approaches for network-based inference from gene expression data. Methods that employ sparse priors have been particularly successful, as they are properly designed to analyze large datasets in which the amount of measured variables can be greater than the number of observations. Here, we describe Bayesian approaches for both undirected and directed networks; we discuss novel approaches that are computationally efficient, do not rely on linearity assumptions, and perform comparatively better than state-of-the-art methods. We demonstrate the utility of our methods via applications to glioblastoma gene expression data.