 Substructure mask explanations, SME, provide a transparent inspection into how graph neural networks, GNNs, learn from data to predict molecular properties such as aqueous solubility, genotoxicity, cardiotoxicity and blood-brain barrier permeability. This method is based on well-established molecular segmentation techniques and provides an interpretation that aligns with the understanding of chemists. It can be used to identify which parts of the molecule are responsible for the model's predictions, allowing chemists to optimize structures for desired properties. This article was authored by Shinshing Wu, Jack Wang, Hongyang Du, and others.