Abstract
Graph Neural Networks (GNNs) play a fundamental role in many deep learningproblems, in particular in cheminformatics. However, typical GNNs cannotcapture the concept of chirality, which means they do not distinguish betweenthe 3D graph of a chemical compound and its mirror image (enantiomer). Theability to distinguish between enantiomers is important especially in drugdiscovery because enantiomers can have very distinct biochemical properties. Inthis paper, we propose a theoretically justified message-passing scheme, whichmakes GNNs sensitive to the order of node neighbors. We apply that generalconcept in the context of molecular chirality to construct Chiral Edge NeuralNetwork (ChiENN) layer which can be appended to any GNN model to enablechirality-awareness. Our experiments show that adding ChiENN layers to a GNNoutperforms current state-of-the-art methods in chiral-sensitive molecularproperty prediction tasks.