Research questionHow can molecular property predictors retain substructure and graph-distance information in compact representations without external pretraining?Compression can discard the substructures and spatial relationships that influence molecular properties. The challenge is to preserve these relationships while keeping representations and prediction models efficient to train and use.