🎯 Predictive Modelling of Drug Target Interaction

📝 Overview

This project implements an improved deep learning framework for predicting drug-target interactions (DTI), combining multiple encoder architectures including CNN, Transformer, and Message-Passing Neural Networks (MPNN). Our approach utilizes both structural information of compounds through SMILES representation and protein sequences through amino acid encodings.

✨ Key Features

• Multiple encoder combinations (CNN, Transformer, MPNN)
• Benchmark evaluation on DAVIS and KIBA datasets
• Case study on SARS-CoV 3C-like protease
• Interactive web interface with PubChem integration

🚀 Getting Started

Prerequisites

• Python 3.8+
• PyTorch
• RDKit
• Streamlit
• Other dependencies (listed in requirements.txt)

Installation

git clone https://github.com/yourusername/dti-prediction.git
cd dti-prediction
pip install -r requirements.txt

💻 Usage

Training Models

Our models are implemented in Google Colab for easy access and GPU support.

Running Web Interface

cd web_interface
streamlit run app.py

📈 Results

Our CNN-Transformer model achieved superior performance with:

• DAVIS dataset: CI of 0.88 and MSE of 0.27
• KIBA dataset: CI of 0.88 and MSE of 0.19

🏗️ Project Structure

dti-prediction/
│
├── notebooks/          # Colab notebooks for model training
├── web_interface/     # Streamlit web application
├── data/              # Dataset handling and preprocessing
├── models/            # Model architectures
└── utils/            # Helper functions and utilities

👥 Authors

• Chang Liu
• Yitian Ma
• Yinuo Yang
• Yuning Zheng
• Zhuoqun Li

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.