Invited: Reverse engineering whole-genome networks

A research on the development of parallel methods to eliminate piecemeal approach on building whole-genome networks and facilitate genome-scale network inference, are discussed. A parallel mutual information based method is developed, which can scale up to an arbitrary number of genes and gene expression measurements on sufficiently large parallel computers and clusters. A novel technique to carry out permutation testing for assessing the statistical significance of predicted gene-gene interactions is used to construct a whole genome network of the model plant Arabidopsis thaliana in 9 minutes on a 1024 core cluster. The paper also studied the construction of the whole genome network using mutual information, sample small subgraphs from the network, and refine these local networks through parallel Bayesian inference; and heuristic algorithms for large scale Bayesian inference.