Introduction: We present an algorithm to discover 'active subnetworks' in a weighted, undirected graph. Active subnetworks are connected regions of a network that are enriched for statistically significant associations with an outcome of interest.
Methods: We consider metabolic correlation networks, in which nodes represent distinct metabolites and undirected edges between nodes are weighted by the magnitude of the correlation between the pair of nodes (metabolites). Each node (metabolite) is scored proportional to the strength of its association with an outcome of interest (e.g. cardiovascular disease). The proposed approach builds on the work of Ditrrich et.al [2008, 2010, and 2012] by incorporating weighted edges in metabolite correlation networks to detect 'active subnetworks' that are enriched for significant associations with the outcome of interest.
Results: We compare the performance of our approach with other methods for the detection of functional modules under simulation settings that reflect real data applications. We illustrate the application of the proposed algorithm to data from a metabolic project to discover biomarkers of coronary heart disease.
|