Clinical Information Driven Ensemble Clustering for Inferring Robust Tumor Subtypes

Inferring tumor subtypes based on the gene expression data alone does not appear to be as powerful as expected for the lack of robustness and clinical meaning. The ultimate aim of clustering tumor samples should be to support clinical evaluation or treatment. Therefore, clustering procedure should closely integrate the clinical outcome and/or treatment information for final representation of the tumor homogeneity and heterogeneity. In this work, we developed an ensemble clustering method guided by the clinical outcome and treatment information for the identification of the robust and clinically meaningful tumor subtypes. Our method was expected to yield more robust and clinically relevant results than other commonly used methods and to give us comprehensive understanding of tumor heterogeneity.