Inducing Pairwise Gene Interactions from Time-Series Data by EDA Based Bayesian Network

Recently a variety of high throughput experimental techniques, such as DNA microarray, are opening system-level perspectives of living organisms on molecular level. Inferring gene-gene interactions from time series data generated from these technologies is an important computational method to help us to understand the system behavior of living organisms. The Bayesian network (BN), which is a graph-based representation of a joint probability distribution that captures properties of conditional independence between variables, is a desirable tool. However, how to construct appropriate BNs that best fit the data profile is very difficult since the number of BNs on n variables is the super-exponential of n. To avert the combinational explosion, in this paper, we use estimation of distribution algorithms (EDAs) to search the space. Also, in order to generate meaningful individuals, we also propose depth-first search method to cut circles in the graphs. We have tested our method on cell-cycle gene expression data and found that it can not only discover some existing relationships in other literatures and gene ontology, but also reveal some previously unknown interactions