Statistical and Machine Learning Approaches for Network Analysis

1.6 Discussion

In this chapter, we discussed a number of approaches for the reconstruction and partition of biological networks. We considered the case of both directed and undirected biological networks in each of the above classes of problems. Network reconstruction algorithms presented in this chapter were further categorized based on the type of measurements used in the inference procedure. The type of measurements which we considered were gene expression data and symbol data comprising of gene sets. Gene expression data are numerical matrices containing gene expression levels measured from different experiments, whereas gene sets are sets of genes and do not assume the availability of the corresponding gene expression levels.

For the reconstruction of directed networks, we presented six approaches Boolean networks, probabilistic Boolean networks, Bayesian networks, cGraph, frequency method, and NICO. Among these approaches Boolean networks, probabilistic Boolean networks, and Bayesian networks accommodate gene expression data, whereas cGraph, frequency method, and NICO are suitable to infer the underlying network topologies from gene sets. For the reconstruction of undirected biological networks from gene expression data, we presented two approaches relevance networks and graphical Gaussian models. Nonetheless, the aforementioned algorithms for network inference using gene expression data are also applicable when the inputs are given in the form of gene set compendiums and ...