Markov chain Monte Carlo simulation of a Bayesian mixture model for gene network inference

Background: Simultaneous measurement of gene expression level for thousands of genes contains the rich information about many different aspects of biological mechanisms. A major computational challenge is to find methods to extract new biological insights from this wealth of data. Complex biological processes are often regulated under the various conditions or circumstances and associated gene interactions are dynamically changed depending on different biological contexts. Thus, inference of such dynamic relationships between genes with consideration of biological conditions is very challenging. Method: In this study, we propose a comprehensive and integrated approach to infer the dynamic relationships between genes and evaluate this approach on three distinct gene networks. Results: This study demonstrates the advantage of integrating Markov chain Monte Carlo (MCMC) simulation into a Bayesian mixture model to overcome the high-dimension, low sample size (HDLSS) problem as well as to identify context-specific biological modules. Such biological modules were identified through the summarization of sampled network structures obtained from MCMC simulation. Conclusion: This novel approach gives a comprehensive understanding of the dynamically regulated biological modules.