This repository contains data and codes for PolyCL.

• Dependency: You will need only polycl.py in this repository and torch,transformers packages as the minimum requirement.
• Obtain the polymer embedding: Simply follow the demonstration in PolyCL_Easy_Usage.ipynb.

You might need to configure git lfs first and download git lfs following instructions on https://git-lfs.com/ . Then install git lfs using:

$ git lfs install 

After git lfs properly configured:

$ git clone https://github.com/JiajunZhou96/PolyCL.git 

# create a new environment $ conda create --name polycl python=3.9 $ conda activate polycl # install requirements #$ pip install numpy==1.26.4 #$ pip install pandas==1.3.3 #$ pip install scikit-learn==0.24.2 $ pip install torch==1.12.0+cu113 -f https://download.pytorch.org/whl/torch_stable.html $ pip install transformers==4.20.1 $ pip install -U torchmetrics $ pip install tensorboard $ pip install tqdm $ conda install -c conda-forge rdkit 

pip install torch-geometric==1.7.2 torch-sparse==0.6.18 torch-scatter==2.1.2 -f https://pytorch-geometric.com/whl/torch-1.12.0+cu113.html 

Run with key parameters for the pretraining summarized in config.json.

train.py 

Run with sample configurations described in config_tf_notebook.json.

transfer_learning.py 

Models available for benchmarking are stored in the ./benchmark/ directory.

• polyBERT
  

  • Run tf_polybert.py and polyBERT model will be automatically downloaded from https://huggingface.co/kuelumbus/polyBERT .

• Transpolymer
  

  • Download the model folder of Transpolymer "pretrain.pt" from https://github.com/ChangwenXu98/TransPolymer/tree/master/ckpt .
    
  • Put the folder to the directory "./model/Trasnpolymer/" to be referred to as "./model/Trasnpolymer/pretrain.pt".
  • Run tf_transpolymer.py.
    

• GNNs
  

  • Assign "gcn" or "gin" to the key "gnn_type" in config_graph.json to use different types of GNNs.
  • Run gnn.py.

• ECFP Methods

  • Run morgan_nn.py to use neural network.
  • Run rf.py to use random forest.
  • Run xgb.py to use XGBoost.