GAUSS dataset

The Gauss dataset is a large-scale collection comprising over 200,000 data points, each containing molecular geometries, local ligand properties, and Kramers doublet energies of dysprosium (Dy) complexes. This dataset was developed to support the training of a variational autoencoder (VAE)-based generative AI model, GAUSS (Generative Autoencoders for State-of-the-art Single-molecule Magnets), aimed at efficiently generating novel single-molecule magnets (SMMs) with record-high magnetic anisotropy. The dataset is organized into three main subsets, each including labels and SMILES representations of the axial ligands within the coordination complexes. Of these, one subset provides local property descriptors of the ligands, while another contains the full geometrical information of the optimized complexes. Together, these subsets support the training of deep neural network (DNN) models for predicting structure–property relationships, in addition to generative modeling tasks.