On modeling gene regulatory networks using Markov random fields

Modeling the joint expression patterns of genes is a challenging task due to the large number of genes simultaneously studied, relative to the amount of microarray data available. To model the joint expression profiles of genes using a small number of observations, we use Ising models to approximate the joint expression profiles. This approach naturally lends itself to the study of gene interactions and has a close connection to clustering techniques, which we use to reconstruct E. coli gene interaction pathways from microarray data. In addition, we note that extending available partial network topology information can be done using very few microarray samples-logarithmic in the number of genes.