PyTorch implementation for MolCPT. Author list withheld during review.
We provide a PyTorch implementation of MolCPT with GIN backbone. Pre-trained weights can be found by referencing the original GraphCL README. Datasets can be downloaded here.
Environment file can be found under the transferLearning_MoleculeNet_PPI subdirectory. If there are package incompatibility issues, you may want to instead create your environment using the file provided by GraphCL here.
To run MolCPT on GraphCL, navigate to the finetuning file here and edit the hyperparameters passed to the main function. We expose dropout, normalization, and filtering threshold as changeable hyperparameters.
Finally, run the following commands on terminal (assuming you already cd'd into the top-level of the MolCPT repository):
cd ./transferLearning_MoleculeNet_PPI/chem/
python finetune_motif.py
Results will be recorded in result.log.