Automated interpretation of Cryo-EM density maps with convolutional neural networks

Haruspex is a fully convolutional neural network that automatically annotates both protein secondary structure and nucleotides in experimentally derived Cryo-EM maps. The network was trained on a carefully curated dataset of EMDB (Electron Microscopy Data Bank) entries. Haruspex enables users to identify folds and can be used to guide model building as well as validate structures.