Precision oncology uses genomic evidence to match patients with treatment but often fails to identify all patients who may respond. The transcriptome of these "hidden responders" may reveal responsive molecular states. We describe and evaluate a machine-learning approach to classify aberrant pathway …