Bayesian network models have been proposed for predicting gene regulatory networks from expression data. A problem of this approach is the large amount of data needed to reliably infer the structure and dynamics of the underlying network.
The aim of this study is to simulate data corresponding to the expression levels of genes in IL-4 signaling pathway and to asses the amount of material needed to identify the initial network structure.
The data is generated from a known pathway model containing both interconnected nodes (interacting genes) and independent nodes (redundant genes). In order to get more realistic results we also apply a noise model to the simulated data, which mimics the types of random and systematic variation inherent in microarray experiments. We consider the data as time series to use the REVEAL [1] algorithm to learn the structure of the interacting genes.
Preliminary results showed that the amount of real microarray experiments required for reconstructing the genuine pathway model is infeasible, even in the case of noise-free data. It remains to be studied whether admissible reconstruction can be achieved from reasonable amount of data.
[1] Liang S. et al (1998), Pac. Symp. Biocomput. 3:18-29