HeteroDTA: Deep Learning Model for Drug-Target Affinity Prediction

🥰 Introduction

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.

😆 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
  

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

🌈 Datasets

All publicly accessible datasets used can be accessed here:

Dataset Name	Link
Davis, KIBA	https://github.com/hkmztrk/DeepDTA/tree/master/data
Human and C.elegans	https://github.com/masashitsubaki/CPI_prediction

🚧 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

🚀 Pre-trained model

All publicly accessible models used can be accessed here:

Model Name	Link
GEM	https://github.com/PaddlePaddle/PaddleHelix
ESM	https://github.com/facebookresearch/esm

🚧 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

🚡 Install tutorial

🚧 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

📃 Training tutorial

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

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

   Replace <dataset_index>, <cuda_index>, and <dataset_type_index> 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.

😝 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 <dataset_index>, <cuda_index>, and <dataset_type_index> 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.

💗 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{10.1093/bioinformatics/btae240,
    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},
    journal = {Bioinformatics},
    volume = {40},
    number = {Supplement_1},
    pages = {i539-i547},
    year = {2024},
    month = {06},
    issn = {1367-4811},
    doi = {10.1093/bioinformatics/btae240},
    url = {https://doi.org/10.1093/bioinformatics/btae240},
    eprint = {https://academic.oup.com/bioinformatics/article-pdf/40/Supplement\_1/i539/58355122/btae240.pdf},
}