Modeling Perturbations using Gene Networks

External factors such as radiation, drugs or chemotherapy can alter the expressions of a subset of genes. We call these genes the primarily affected genes. Primarily affected genes eventually can change the expressions of other genes as they activate/suppress them through interactions. Measuring the gene expressions before and after applying an external factor (i.e., perturbation) in microarray experiments can reveal how the expression of each gene changes. It however can not identify the cause of the change. In this paper, we consider the problem of identifying primarily affected genes given the expression measurements of a set of genes before and after the application of an external perturbation. We develop a novel probabilistic method to quantify the cause of differential expression of each gene. Our method considers the possible gene interactions in regulatory and signaling networks, for a large number of perturbed genes. It uses a Bayesian model to capture the dependency between the genes. Our experiments on both real and synthetic datasets demonstrate that our method can find primarily affected genes with high accuracy. Our experiments also suggest that our method is significantly more accurate then SSEM, a recent method developed for single gene perturbations, and the Student’s t-test.