Rare neglected diseases may be neglected but are hardly rare, affecting hundreds of millions of people around the world. Here, we present a hit identification approach using AtomNet, the world's first deep convolutional neural network for structure-based drug discovery, to identify inhibitors targeting aspartate N-acetyltransferase (ANAT), a promising target for the treatment of patients suffering from Canavan disease. Despite the lack of a protein structure or high sequence identity homologous templates, the approach successfully identified five low-micromolar inhibitors with drug-like properties.