Using Local Reaction Structure To Build A Global Metabolic Network Classifier

This work is motivated by the increasing requirement in bioinformatics to model complex networks, such as metabolism, in conjunction with data table observations, such as microarray experiments, to facilitate a greater understanding of experimental results. Our goal is to accurately classify a response variable using the local reaction structure of metabolic networks. We represent a metabolic network as a bipartite graph of compounds connected by the reactions. Each reaction is modeled as a function of its underlying gene expression, and the action of the compounds is to combine each reaction function into global network classifier. Our model has two major advantages: