Stacking of network based classifiers with application in breast cancer classification

Abstract

In this study we present the use of existing biological knowledge in the form of biological networks for the construction of a two level classification scheme. At the first level base classifiers are built using a given list of candidate “biomarkers” and the topology of the biological network. In particular, the network structure is taken into account by a search strategy based on random walks for the selection of the genes used in these classifiers. At the second level, a metaclassifier is trained to combine in the best possible way the results of the base classifiers. The proposed approach therefore aims to strengthen the classification ability of the initial list of genes and provide more robust generalization guarantees. Our methodology is explained in full detail and promising results in Breast Cancer related scenarios are presented.