Statistical Analysis of Discrete Dynamical System Models for Biological Networks.

Very few data-driven methods for dynamic biological networks reconstruction from gene expression data evaluate the statistical significance of a model. A hypothesis testing procedure examining the goodness of fit of trajectory-based modeling is designed, in contrast to transition-based model fitting. The former has substantially reduced the modeling error. Simulation studies on the residual between noisy observations and true system dynamics suggest the use of the statistical hypothesis testing, so that one can evaluate how significantly a model is supported by the observed data under certain noise distribution. This method can also evaluate the dynamic model for each individual gene. Through a biochemical reaction model in the yeast pheromone pathway the effectiveness of the proposed evaluation procedure is demonstrated.