Comprehensive pharmacogenomic pathway screening by data assimilation

We propose a computational method to comprehensively screen for pharmacogenomic pathway simulation models. A systematic model generation strategy is developed; candidate pharmacogenomic models are automatically generated from some prototype models constructed from existing literature. The parameters in the model are automatically estimated based on time-course observed gene expression data by data assimilation technique. The candidate simulation models are also ranked based on their prediction power measured by Bayesian information criterion. We generated 53 pharmacogenomic simulation models from five prototypes and applied the proposed method to microarray gene expression data of rat liver cells treated with corticosteroid. We found that some extended simulation models have higher prediction power for some genes than the original models.