Inference of regulatory networks with MCMC sampler guided by mutual information

Abstract

Computationally efficient and exact inference of regulatory network topology is an open problem in System Biology. In this work we investigate the use of prior information about the network topology as a guide to a Markov Chain Monte Carlo sampler of network structures. The prior information is obtained from a coarser and faster network inference method, the Relevance Networks with Mutual Information scores. Moreover, the regulatory networks are represented by the Bayesian Networks model. The results show that the use of prior information drastically improves the convergence of the MCMC sampler. Therefore, the use of a more refined method is justified as it is likely to lead to more reliable results with less MCMC iterations.