Application of a state-of-the-art deep learning method to the imputation of sparse physicochemical and sensory data. Compare results with conventional QSAR methods and a multi-target graph convolutional neural network.