Predicting distant metastasis in breast cancer using ensemble classifier based on context-specific miRNA regulation modules

Many methods based on building classifiers by selecting gene markers have been proposed to predict breast cancer patient´s outcome However, most of them suffer from the problem of poor robustness, which are mainly due to the fact that the overlap degree of gene markers derived by different methods is not high, hading that few of them are generalized and can be widely used for clinical practice. In this paper, we present a method based on context-specific miRNA regulation modules to predict distant metastasis in breast cancer. First, we describe the regulation activity of a miRNA on a specific context by using CoMi (Context-specific miRNA activity) score, based on which, several miRNAs regulate on the same context are regarded as a miRNA regulation module; then the discriminate regulation modules are selected and each is used to construct a classification model separately; finally, an ensemble classifier is established by combining all the models with a majority voting strategy. The evaluation experiment results show that our method performs better than previous works. In addition, the obtained discriminate modules show great stability across different data sets (withp-value of 1.119e-06).