| The triple-negative (TN) and basal-like breast cancers have poor outcomes, and lack both targeted therapies and accurate prognostic markers of outcome. Despite the application of microarray technologies to molecular profiling of breast tumors, most current genomics-derived predictors are incapable of stratifying TN or basal breast cancer patients by outcome. We have collected all publicly available breast cancer gene expression datasets to build a human-Compendium; from this, we selected TN and basal patient cohorts to build a TN-Compendium (TN-C) and basal-Compendium (basal-C). Using a de novo machine learning methodology, we have built 25-gene predictors of recurrence for TN and basal patients. Compared to previously reported predictors, these classifiers exhibit superior performance, and highlight multiple biological processes, including immune response, cytoskeletal regulation, signaling and ligand gated ion channels, as being differentially present between recurrent and non-recurrent patients. The small size of these predictors makes them potential candidates for use in clinical settings. |
