Integrate qualitative biological knowledge for gene regulatory network reconstruction with dynamic Bayesian networks

Abstract

Reconstructing gene regulatory networks, especially the dynamic gene networks that reveal the temporal program of gene expression from microarray expression data, is essential in systems biology. To overcome the challenges posed by the noisy and under-sampled microarray data, developing data fusion methods to integrate legacy biological knowledge for gene network reconstruction is a promising direction. However, large amount of qualitative biological knowledge accumulated by previous research, albeit very valuable, has received less attention for reconstructing dynamic gene networks due to its incompatibility with the quantitative computational models. In this dissertation, I introduce a novel method to fuse qualitative gene interaction information with quantitative microarray data under the Dynamic Bayesian Networks framework. This method extends the previous data integration methods by its capabilities of both utilizing qualitative biological knowledge by using Bayesian Networks without the involvement of human experts, and taking time-series data to produce dynamic gene networks. The experimental study shows that when compared with standard Dynamic Bayesian Networks method which only uses microarray data, our method excels by both accuracy and consistency.