Search

We consider the problem of learning dynamical models of genetic regulatory networks from time-lapse measurements of gene expression. In our previous work [Porreca et al,Bioinformatics,2010], we described a method for the structural and parametric identification of ODE models that makes use of concurrent measurements of concentrations and synthesis rates of the gene products, and requires the knowledge of the noise statistics. In this paper we assume all these pieces of information are not simultaneously available. In particular we propose extensions of [Porreca et al,Bioinformatics,2010] that make the method applicable to protein concentration measurements only. We discuss the performance of the method on experimental data from the network IRMA, a benchmark synthetic network engineered in yeast Saccharomices cerevisiae.

A common situation in System Biology is to use several alternative models of a given biochemical system, each with a different structure reflecting different biological hypotheses. These models then have to be ranked according to their ability to reproduce experimental data. In this paper, we use Bayesian model selection to test four alternative models of the yeast GATA-factor genetic network. We employ three different computational methods to calculate the necessary probabilities and evaluate their performance for medium-scale biochemical systems.