A new dynamic Bayesian network for integrating multiple data in estimating gene networks

We propose a novel method for estimating a gene regulatory network from both gene expression data and transcription factor binding location data. Based on dynamic Bayesian network (DBN) models, our method has two advantages. First, structural expectation maximization algorithm is embedded into a DBN framework, allowing it to learn the network with unknown structure and incomplete data, which cannot be tackled by expectation maximization algorithm. Second, since learning only from gene expression data is not enough to accurately estimate a gene network, we incorporate transcription factor binding location data through a structure prior. We demonstrate the effectiveness of the proposed method by the analysis of Saccharomyces Cerevisiae cell cycle data. The experimental results show that this new approach suits for dealing with missing values. Furthermore, combination of heterogeneous data from multiple sources ensures that our results are more accurate than those recovered from gene expression data alone.