Official Pytorch implementation of HiMol model in the paper "Xuan Zang, Xianbing Zhao, Buzhou Tang. Hierarchical Molecular Graph Self-supervised Learning for Property Prediction".

python rdkit scipy torch torch-geometric torch-sparse tqdm networkx numpy pandas

You can pretrain the model by

mkdir saved_model
python pretrain.py

You can evaluate the pretrained model by finetuning on downstream tasks

Download the downstream data from https://github.com/deepchem/deepchem/tree/master/deepchem/molnet/load_function, and save the .csv files in the ./finetune/dataset/[dataset_name]/raw/, where [dataset_name] is replaced by the downstream dataset name. For example, bace.csv is saved in './finetune/dataset/bace/raw/bace.csv'.

cd finetune
mkdir model_checkpoints
python finetune.py