The investigation of the interplay between genes, proteins, metabolites and diseases plays a central role in molecular and cellular biology. Whole genome sequencing has made it possible to examine the behavior of all the genes in a genome by high-throughput experimental techniques and to pinpoint molecular interactions on a genome-wide scale, which form the backbone of systems biology. In particular, Bayesian network (BN) is a powerful tool for the ab-initial identification of causal and non-causal relationships between biological factors directly from experimental data. However, scalability is a crucial issue when we try to apply BNs to infer such interactions. In this paper, we not only introduce the Bayesian network formalism and its applications in systems biology, but also review recent technical devel-opments for scaling up or speeding up the structural leaing of BNs, which is important for the discovery of causal knowledge from large-scale biological datasets.Specifically, we highlight the basic idea, relative pros and cons of each technique and discuss possible ways to combine different algorithms towards making BN leaing more accurate and much faster.