⭐ HeteroDTA: Deep Learning Model for Drug-Target Affinity Prediction

🥰 This repository contains the HeteroDTA deep learning model, which is designed for predicting drug-target affinity in drug discovery tasks. The model is implemented in Python and utilizes various deep learning techniques for accurate affinity predictions.

:satisfied: Requirements

• numpy == 1.17.4
• kreas == 2.3.1
• pytorch == 1.8.0
• matplotlib==3.2.2
• pandas==1.2.4
• PyG (torch-geometric) == 1.3.2
• rdkit==2009.Q1-1
• tqdm==4.51.0
• numpy==1.20.1
• scikit_learn==0.24.2
  

:construction: Note: There are some dependencies that are not listed, please install them independently according to the feedback from the console

:rainbow: Datasets

All publicly accessible datasets used can be accessed here:

:construction: Note: It is necessary to download the corresponding data set and then place it in the corresponding directory named "data" on your own machine for subsequent data preprocessing

:rocket: Pre-trained model

All publicly accessible models used can be accessed here:

:construction: Note: It is necessary to read the corresponding GitHub guidelines and use pre-trained models on your own machine to preprocess your chosen datasets or your own private datasets

• The role of GEM: obtain atomic embeddings in compounds
• The role of ESM: (a) predict protein structure; (b) obtain embeddings of amino acids in proteins

:aerial_tramway: Install tutorial

:construction: Notes:

• Before training or inference, the datasets used must be preprocessed
• Please refer to the individual script files for more detailed instructions and options.

1. Clone the repository

   git clone https://github.com/daydayupzzl/HeteroDTA.git
   cd your-repository

2. Install the required dependencies

   pip install SomePackage # Please follow comsole's feedback and install the missing packages until there is no lack of dependencies

   :page_with_curl: Training tutorial

3. Make sure you have the necessary Python libraries installed, including PyTorch.

4. Open a terminal (Command Prompt on Windows, or Terminal on Mac/Linux).

5. Navigate to the directory containing the file.

6. Run the following command:

   python training.py <dataset_index> <cuda_index> <dataset_type_index>

   Replace , , and with the following values:

   • dataset_index: Dataset index. 0 for 'davis', 1 for 'kiba'.
   • cuda_index: GPU index. 0 for 'cuda:0', 1 for 'cuda:1'.
   • dataset_type_index: Dataset type index. 0 for 'original', 1 for 'cold_drug', 2 for 'cold_protein', 3 for 'cold_pair'.

   For example:

   python your_script_name.py 0 0 0 v1

   This will run the script, using the "davis" dataset, utilizing the first GPU, and employing the original dataset type.

:stuck_out_tongue_closed_eyes: Inference tutorial

1. Ensure that you have trained one model
2. Open a terminal (Command Prompt on Windows, or Terminal on Mac/Linux).
3. Navigate to the directory containing the file.

4. Run the following command:

   python inference.py <dataset_index> <cuda_index> <dataset_type_index>

   Replace , , and with the following values:

   • dataset_index: Dataset index. 0 for 'davis', 1 for 'kiba'.
   • cuda_index: GPU index. 0 for 'cuda:0', 1 for 'cuda:1', 2 for 'cuda:2', 3 for 'cuda:3'.
   • dataset_type_index: Dataset type index. 0 for 'original', 1 for 'cold_drug', 2 for 'cold_protein', 3 for 'cold_pair'.

   For example:

   python inference.py 0 0 0

   This will execute the script, using the "davis" dataset, utilizing the first GPU, and employing the original dataset type.

:heartpulse: Citation

If you use this code or related methods in your research, please consider citing HeteroDTA paper (We will add the literature address in the future)

    @article {,
        author = {Zhang, Zuolong and He, Xin and Long, Dazhi and Luo, Gang and Chen, Shengbo},
        title = {Enhancing generalizability and performance in drug-target interaction identification by integrating pharmacophore and pre-trained models},
        year={2024},
        publisher = {Oxford University Press (OUP)},
        journal = {Bioinformatics}
    }