Parallel machine learning approaches for reverse engineering genome-scale networks

Reverse engineering whole-genome networks from large-scale gene expression measurements and analyzing them to extract biologically valid hypotheses are important challenges in systems biology. While simpler models easily scale to large number of genes and gene expression datasets, more accurate models are compute intensive limiting their scale of applicability. In this talk, I will present our research on the development of parallel mutual information and Bayesian network based structure learning methods to eliminate such bottlenecks and facilitate genome-scale network inference. As a demonstration, we reconstructed genome-scale networks of the model plant Arabidopsis thaliana from 11,700 microarray experiments using 1.57 million cores of the Tianhe-2 Supercomputer. Such networks can be used as a guide to predicting gene function and extracting context-specific subnetworks.