Describir: Interpretable machine learning predicts cardiac resynchronization therapy responses from personalized biochemical and biomechanical features.