Drug-induced gene expression profiles can reveal compound mechanisms but are difficult to generate at scale. Here, the authors present WAVE, a deep learning model that accurately predicts bulk and single-cell drug responses, with experimental validation supporting its potential for drug discovery.