Prior knowledge driven causality analysis in gene regulatory network discovery

Previous researches focus on applying the Granger causality (GC) model to time-series DNA micro array data to infer gene regulatory networks. However, in biological datasets, the number of available time points is usually much smaller than the number of target genes. Therefore, people widely used a bivariate GC model, which might lead to a significant amount of false discoveries in the causality analysis. In this study, we proposed a new framework to resolve the problem by incorporating the prior biological knowledge. These prior knowledge helps us to use/build a gene association network and cluster the candidate gene set into smaller groups. Within each small group, the more precise multivariate GC model is applied to discover causalities. We validated this new framework to a yeast metabolic cycle dataset and initial analysis revealed the potentials of our approach in discovering meaningful regulatory networks.