We present robust classification algorithms for high-dimensional genomic data to find biomarkers, by analyzing variable importance, that enable a better diagnosis of disease, an earlier intervention, or a more effective assignment of therapies. The goal is to isolate a set of genes which play an important role in classification with respect to the prognosis or types of diseases in order to maximize efficacy or minimize toxicity in personalized treatment of a life-threatening disease using several methods for variable importance ranking. First, we present a highly accurate classification algorithm with a method for selecting highly ranked variables as a high-dimensional biomarker. We assess the prediction accuracy of the biomarker by cross-validation using published high-dimensional data sets. This approach can discover genomic biomarkers underlying both adverse and efficacious outcomes.