Integrate Qualitative Biological Knowledge to Build Gene Networks by Parallel Dynamic Bayesian Network Structure Learning

Abstract

Dynamic Bayesian Network has become a popular tool for reconstructing gene regulatory networks from microarray data. Developing high performance method is crucial to deal with the huge computational workload of Bayesian Network structure learning. Also noisy and under-sampled microarray data requires data integration mechanism to make use of legacy biological knowledge for more accurate gene network prediction. In this paper, we introduce a software system targeting on building large-scale gene networks realized by both parallel computing, and a novel data integration model which fuses qualitative gene interaction information with quantitative microarray data under the Dynamic Bayesian Networks framework. The experimental study shows that our method can accelerate the computation by using multiple CPUs, while still maintain its advantage in accuracy over non-integrative methods.