With increasing ubiquity of genome-wide assays it is now common to molecularly subtype cancers to predict patient therapy response. However identifying high-performing, robust molecular signatures for predictions remains difficult. Presented here is work towards a novel machine-learning algorithm that discovers intuitively understood and clinically relevant stratifying molecular signatures. This classification method, as well as many competing methods (SVM, random forests, Bayes nets, etc.), were applied by predicting drug-sensitivity to hundreds of compounds tested on the NCI60 cell lines. This new drug-sensitivity prediction method competes with and in many cases outperforms leading classifiers. The prediction results, as well as the molecular signatures that they are derived from, are publicly available for web-browsing through a new extension to the UCSC Cancer Genomics Browser, hgClassifications (http://genome-cancer.soe.ucsc.edu/hgClassifications). Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 102nd Annual Meeting of the American Association for Cancer Research; 2011 Apr 2-6; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2011;71(8 Suppl):Abstract nr 43. doi:10.1158/1538-7445.AM2011-43