The ability to characterize and predict tumor phenotypes is crucial to precision medicine. In this study, we present an integrative computational approach using a genome-wide association analysis and an Elastic Net prediction method to analyze the relationship between DNA copy number alterations and an archive of gene expression signatures. Across breast cancers, we are able to quantitatively predict many gene signatures levels within individual tumors with high accuracy based upon DNA copy number features alone, including proliferation status and Estrogen-signaling pathway activity. We can also predict many other key phenotypes, including intrinsic molecular subtypes, estrogen receptor status, and TP53 mutation. This approach is also applied to TCGA Pan-Cancer, which identify repeatedly predictable signatures across tumor types including immune features in lung squamous and basal-like breast cancers. These Elastic Net DNA predictors could also be called from DNA-based gene panels, thus facilitating their use as biomarkers to guide therapeutic decision making.