Can dynamic gene networks predict novel therapeutic targets and clinical outcomes?
Transcriptomic analysis is commonly performed in a time-dependent manner to uncover the temporal response of RNA expressions to a dosed chemical compound. By using drug response time-course gene expression data measured by microarray or RNA-seq technologies, statistical time-series analysis can describe the temporal dependencies among genes as a mode-of-action of a drug. In the first part of this talk, we introduce several statistical models including a state-space model and dynamic Bayesian networks for estimating dynamic gene networks from time-course gene expression data, together with parameter estimation and model structure algorithms. In the second part, we show several examples of real application of the dynamic gene network estimation, including survival prediction of lung cancer patients based on gefitinib -sensitive genes which are downstream of EGFR as identified by the state-space model.