Multisyn: A Multi-Source information Integration Approach for Synergistic Drug Combination Prediction

This repository contains the our implementation of the paper aboved.

1. Introduction

Our method, Multisyn, integrates pharmacophore information, PPI networks, and genomic data to predict synergistic anti-cancer drug combinations.

Paper Link:

1.1 Features

• Represents molecules using heterogeneous molecular graphs.
• Predicts synergistic drug combinations using attributed graph neural networks.
• Integrates multiple cell line data sources: genomics and protein-protein interaction (PPI) networks.
• Provides interpretable results with an attention mechanism for substructure identification.

1.2 Results

1.3 Citation

If you find this repository helpful, please cite our work:

2. Item

2.1 Requirements

Our operating environment is as follows：

• Python 3.11.5
• Pytorch 2.1.0
• numpy 1.24.3
• Pytorch Geometric 2.4.0

2.2 File Description

This project contains six files for preprocessing, training, and evaluating drug combination prediction models.

The project is mainly divided into data processing, model definition, training, and evaluation.

• const.py: The const.py file defines variables for common file paths.
• data_drug.py:process drug data, construct molecular graphs, and create datasets.
• dataset_drug.py:process drug-target data and convert it into PyTorch Geometric format.
• model_drug.py: Defines the model for drug molecular feature.
• model.py:predict drug combination.
• train.py: Main program for training the prediction model.
• utils.py: loss functions, evaluation metrics, etc.

2.3 Run

Preprocess the data by running:

When running for the first time, please run the file to create data.

python data_drug.py

Train the model：

python train.py