PyTorch implementation of MCPNET.
Make sure PyTorch 1.11.0 and RDKIT are installed, should resolve all dependencies. The program is run and tested using python3.
A complete example of running the program is provided in the attention_pred.py, including molecular point cloud feature generation, model training and evaluation. You can put your own data in the /data to train model. You should modify the parameters in the following command line to your data path.
python3 attention_pred.py --source_path [your data path]For visualizing MCPNET, you should provide your trained model, data, and scaler of data. Then call the get_important_points method in model_utils.py. The returned data can be rendered by calling the plot_points method in plot_utils.py. The following figure shows the results:
