Inference of Restricted Stochastic Boolean GRN's by Bayesian Error and Entropy Based Criteria

This work compares two frequently used criterion functions in inference of gene regulatory networks (GRN), one based on Bayesian error and another based on conditional entropy. The network model utilized was the stochastic restricted Boolean network model; the tests were realized in the well studied yeast cell-cycle and in randomly generated networks. The experimental results support the use of entropy in relation to the use of Bayesian error and indicate that the application of a fast greedy feature selection algorithm combined with an entropy-based criterion function can be used to infer accurate GRN’s, allowing to accurately infer networks with thousands of genes in a feasible computational time cost, even though some genes are influenced by many other genes.