MMSyn

Introduction

MMSyn is a multimodal deep learning framework for prediction of synergistic drug combinations by integrating multimodal data.

Data

cell lines

• ONeil_31_gsva_1329_dim.csv - pathway scores of cell lines
• gene expressin data is downloaded from CCLE (the Cancer Cell Line Encyclopedia)
• DNA copy number data is downloaded from CCLE (the Cancer Cell Line Encyclopedia)

drug

• 38_drug_smiles.csv - SMILES (Simplified molecular input line entry system) of drugs

dataset

• drug_pair_cell_line_triple.csv - the effects of 538 pairwise drug combinations in 30 cell lines

Environment Requirement

The code has been tested running under Python 3.7. The required package are as follows:

• pytorch == 2.0.0+cu118
• numpy == 1.26.0
• sklearn == 1.0.2
• networkx == 2.8.4
• pandas == 1.2.4
• rdkit == 2023.3.1
• torch_geometric == 2.3.0

Source codes

• cell_autoencoder.py: learn low_dimensional representations from high-dimensional cell line features
• dataset.py: the dataset objects generated by PyG
• metric.py: evaluation metric functions
• model.py: details of MMSyn model
• preprocess.py: load data and convert to pytorch format
• pubchemfp.py: generate drug pubchem fingerprints
• simles2graph.py: convert SMILES sequence to graph
• train.py: train the model and make predictions
• trainer.py: training and evaluation functions

Step-by-step instructions

1. Install dependencies, including torch2.0, torch_geometric, sklearn, rdkit, and networkx.
2. Run cell_autoencoder.py to reduce the dimensionality of the DNA copynumberdata and gene expression data.
3. Run preprocess.py to convert label data and feature data into pytorch format.
4. Run train.py for training and prediction.