Multi-Dynamic Bayesian Networks for Pattern Recognition in Genomic DNA Sequences

Leo Wang-Kit Cheung1, Angel Yee-Man Cheung2, Cancer Research lCenter of Hawaii, University of Hawaii ;, Department of Computer Science, Chu Hai College

In parallel with the recent development of the novel Hidden Multivariate Markov Models (HM3s) (Cheung, 2003), a family of Dynamic Bayesian Networks (DBNs) is introduced for analyzing multi-dimensional genomic DNA data. This family of Bayesian Networks, which we called Multi-Dynamic Bayesian Networks (MDBNs), is designed with an overall network architecture that connects multiple DBNs. It provides a flexible alternative tool to incorporate multiple sources of different kinds of data for recognition of multiple patterns. We have been exploring the implementation of MDBNs with both classical and empirical Bayesian ideas in order to enhance their applicability to various areas of Bioinformatics and Computational Molecular Biology. The focus of this poster is to show how a two-dimensional version of MDBNs can be applied to recognition and prediction of eukaryotic promoter regions. In particular, we make use of the discrete base-compositional data and the continuous bendability (or structural bending propensity) data as our two-dimensional DNA data in our applications. Illustrations are shown via case studies using real human DNA data provided by Dr. Anders Pedersen at the Center for Biological Sequence Analysis in Denmark. Comparisons with the HM3s are also discussed.