HOFDiff

HOFDiff: Synthetic HOF Structures and Pretrained Weights<br>This repository provides synthetic HOF structures generated by the <b>HOFDiff</b> model, along with the corresponding preprocessed datasets and pretrained model weights. These resources are designed to support pretraining and data augmentation for hydrogen-bonded organic framework (HOF) property prediction tasks.<br>Project repository: https://github.com/wangtaoyang/HOFDiff<br>Contents- <code><strong>hofdiff-20k.zip</strong></code> Contains 20,000 HOF structures generated by HOFDiff. These were filtered using <b>MOFChecker</b> with a relaxed criterion.- <code><strong>hofdiff-1300.zip</strong></code> A refined subset of 1,300 high-quality HOF structures selected from the above using stricter MOFChecker filters.<br>- <code><strong>tobacco_cifs.zip</strong></code> Synthetic HOF structures generated by ToBaCCo. - <code><strong>hof_save_path.zip</strong></code> Preprocessed HOFDiff dataset directory used for training. Contains serialized outputs such as structure graphs, embeddings, and intermediate features. <br>- <code><strong>lmdb_data.zip</strong></code> LMDB-format data used to accelerate HOFDiff training and evaluation.Refer to the official repo for usage: HOFDiff GitHub- <code><strong>hofdiff_bb_encoder.ckpt</strong></code> Pretrained checkpoint of the HOFDiff <b>building block encoder</b> module.<br>- <code><strong>hofdiff_diffusion_model.ckpt</strong></code> Pretrained checkpoint of the HOFDiff <b>diffusion model</b>, used for structure generation from latent space.