Abstract:
|
A significant bottleneck in the inference of networks from transcriptomic data is the selection of a meaningful group of genes on which to focus interest. In this work, we develop a novel and targeted causal approach to identify candidates of causal downstream relationships in gene expression experiments consisting of replicated observational and single or multiple knock-out (KO) data. In particular, for each potential target of a KO gene, we use the framework of Gaussian Bayesian networks to perform selection among the models representing correlated, upstream, and downstream relationships. The proposed procedure is fast and parallelizable, facilitating its implementation in real data analyses. Simulations confirm its ability to accurately separate downstream partners of KO genes from upstream and spurious associations, and we illustrate the interest of the approach on a set of in-house animal genetics expression data that includes both wild-type and KO experiments.
|