Deep learning can be used to predict genomic alterations based on morphological features learned from digital histopathology. Two independent pan-cancer studies now show that automated learning from digital pathology slides and genomics can potentially decipher broader classes of molecular signatures and prognostic associations across cancer types.