Processed CrossDocked2020 and ZINC datasets for CascadeDiff

Processed datasets used to train and evaluate CascadeDiff, a coarse-to-fine diffusion model for structure-based ligand design. CrossDocked2020 training/test data and the sequence-similarity-based split (about 100,000 training protein-ligand pairs and 100 test pockets), derived from the CrossDocked2020 release (Francoeur et al., 2020) ZINC molecules used for fine-stage pre-training, from the ZINC database. Code and trained model weights: https://github.com/lijianing0902/CascadeDiff